[{"data":1,"prerenderedAt":23603},["ShallowReactive",2],{"blog-post-/blogs/2-physical-ai-next-investment-theme":3,"blog-toc-/blogs/2-physical-ai-next-investment-theme":1671,"related-posts-/blogs/2-physical-ai-next-investment-theme":2862},{"id":4,"title":5,"body":6,"description":1650,"extension":1651,"meta":1652,"navigation":1666,"ogImage":1654,"path":1667,"seo":1668,"stem":1669,"__hash__":1670},"content/blogs/2-physical-ai-next-investment-theme.md","フィジカルAIが拓く次の投資フロンティア｜AMI Labsが示す世界モデルの可能性 | Physical AI: The Next Investment Frontier",{"type":7,"value":8,"toc":1604},"minimark",[9,805,807],[10,11,13,18,31,36,39,46,53,121,126,133,144,147,150,204,207,211,214,217,290,293,304,308,311,331,334,336,339,342,423,426,524,528,534,545,548,551,578,580,583,590,593,678,683,690,694,697,701,708,711,714,740,743,745,749,752,759,762,765,791,797,799],"lang-block",{"lang":12},"ja",[14,15,17],"h1",{"id":16},"フィジカルaiが拓く次の投資フロンティア","フィジカルAIが拓く次の投資フロンティア",[19,20,21,22,30],"p",{},"私はCVCで北米のEVリユース定置型蓄電や半固体電池スタートアップへの出資を担当してきた人間です。スタートアップ技術の系統分類とシナジーマップを自ら作る作業を通じて学んだのは、",[23,24,25,26,29],"strong",{},"「ハードウェアが絡むAI領域は、ソフトウェアAIと同じ速度では立ち上がらない」",[23,27,28],{},"ということでした。だからこそ、いま「LLMの次」と語られているフィジカルAI（Physical AI）を、ハイプではなく","バリューチェーンと参入タイミング","で読み解いてみたい——というのがこの記事の動機です。",[32,33,35],"h2",{"id":34},"llmブーム後の重心移動をどう読むか","LLMブーム後の「重心移動」をどう読むか",[19,37,38],{},"最初に、市場が向かい始めている方角をフラットに整理しておきます。",[19,40,41,42,45],{},"ChatGPTに代表される大規模言語モデル（LLM）が世界を席巻してから数年。生成AIへの直接投資ブームはひと段落して、投資家の視線はいま「次の波」に向かっています。",[23,43,44],{},"フィジカルAI","——AIが言語・画像だけでなく、現実の物理世界を理解し、その中で行動できるようになる技術領域——が、その筆頭候補です。",[19,47,48,49,52],{},"この潮流の象徴になっているのが、チューリング賞受賞者・Yann LeCunが2026年初頭に立ち上げた",[23,50,51],{},"AMI Labs（Advanced Machine Intelligence Labs）","。10億ドル超のシードラウンドという数字よりも、私には**「研究者がデジタルAIに見切りをつけて物理AIに移った」**という事実の方が重く見えています。",[54,55,56,69],"table",{},[57,58,59],"thead",{},[60,61,62,66],"tr",{},[63,64,65],"th",{},"項目",[63,67,68],{},"内容",[70,71,72,81,89,97,105,113],"tbody",{},[60,73,74,78],{},[75,76,77],"td",{},"正式名称",[75,79,80],{},"Advanced Machine Intelligence Labs（AMI Labs）",[60,82,83,86],{},[75,84,85],{},"創業",[75,87,88],{},"2026年（Yann LeCun × Alexandre Lebrun）",[60,90,91,94],{},[75,92,93],{},"本社",[75,95,96],{},"パリ（フランス）",[60,98,99,102],{},[75,100,101],{},"調達額",[75,103,104],{},"10.3億ドル（欧州史上最大のシードラウンド）",[60,106,107,110],{},[75,108,109],{},"バリュエーション",[75,111,112],{},"約35億ドル（pre-money）",[60,114,115,118],{},[75,116,117],{},"主要投資家",[75,119,120],{},"Cathay Innovation, Greycroft, HV Capital, Toyota Ventures, NVIDIA, Bezos Expeditions, Temasek 他",[122,123,125],"h3",{"id":124},"技術の核心世界モデルworld-models","技術の核心：世界モデル（World Models）",[19,127,128,129,132],{},"AMI Labsが開発するのは、",[23,130,131],{},"世界の物理的な仕組みを学習するAI","——「世界モデル」です。LLMとは設計思想が根本的に違うのが面白いところで、LeCunが2022年に提唱した**JEPA（Joint Embedding Predictive Architecture）**を土台にしています。",[134,135,140],"pre",{"className":136,"code":138,"language":139},[137],"language-text","LLMのアプローチ：    テキスト → テキスト生成\n世界モデルのアプローチ：センサーデータ → 物理世界の理解 → 行動計画\n","text",[141,142,138],"code",{"__ignoreMap":143},"",[19,145,146],{},"LeCunはかねてから「LLMだけでは汎用人工知能（AGI）には到達できない」と言い続けてきた人ですよね。AMI Labsはその哲学を10億ドル分の確信で実装した、というのが正確な表現だと思います。",[122,148,149],{"id":149},"主な応用分野",[54,151,152,162],{},[57,153,154],{},[60,155,156,159],{},[63,157,158],{},"分野",[63,160,161],{},"具体例",[70,163,164,172,180,188,196],{},[60,165,166,169],{},[75,167,168],{},"産業ロボティクス",[75,170,171],{},"工場での複雑な組み立て・検査作業",[60,173,174,177],{},[75,175,176],{},"自律走行",[75,178,179],{},"予測不能な道路環境への適応",[60,181,182,185],{},[75,183,184],{},"医療・ヘルスケア",[75,186,187],{},"手術支援ロボット、診断補助",[60,189,190,193],{},[75,191,192],{},"ウェアラブル",[75,194,195],{},"状況を理解して支援するデバイス",[60,197,198,201],{},[75,199,200],{},"宇宙・航空宇宙",[75,202,203],{},"遠隔環境での自律オペレーション",[205,206],"hr",{},[32,208,210],{"id":209},"なぜ今フィジカルaiなのか","なぜ今「フィジカルAI」なのか",[19,212,213],{},"ここは数字で見るのがいちばん早いです。",[122,215,216],{"id":216},"市場規模の爆発的成長",[54,218,219,238],{},[57,220,221],{},[60,222,223,226,229,232,235],{},[63,224,225],{},"指標",[63,227,228],{},"2025年",[63,230,231],{},"2030年予測",[63,233,234],{},"2033年予測",[63,236,237],{},"CAGR",[70,239,240,257,273],{},[60,241,242,245,248,251,254],{},[75,243,244],{},"フィジカルAI市場",[75,246,247],{},"52億ドル",[75,249,250],{},"—",[75,252,253],{},"497億ドル",[75,255,256],{},"32.5%",[60,258,259,262,265,268,270],{},[75,260,261],{},"身体化AI（Embodied AI）",[75,263,264],{},"44億ドル",[75,266,267],{},"231億ドル",[75,269,250],{},[75,271,272],{},"39.0%",[60,274,275,278,281,284,287],{},[75,276,277],{},"ヒューマノイド出荷台数",[75,279,280],{},"約1万台",[75,282,283],{},"50〜100万台",[75,285,286],{},"200万台超",[75,288,289],{},"60%+",[122,291,292],{"id":292},"ヒューマノイドロボットの量産化",[19,294,295,296,299,300,303],{},"Goldman Sachsは2026年のヒューマノイドロボット世界出荷台数を",[23,297,298],{},"5〜10万台","と予測していて、製造コストは",[23,301,302],{},"1台あたり1.5〜2万ドル","まで下がる見通しです。私はEV用半固体電池のDDをした経験があるので、この数字の裏側で何が起きているか想像がつくのですが、要するに**「セル・センサー・アクチュエータの3つが同時にコモディティ化に近づいている」**のが効いています。これが揃ったのは過去10年で初めてだと思います。",[122,305,307],{"id":306},"llmの次の壁","LLMの「次の壁」",[19,309,310],{},"LLMはテキストと画像の世界では驚異的な能力を発揮しましたが、以下の点で限界が露呈してきています。",[312,313,314,321,326],"ul",{},[315,316,317,320],"li",{},[23,318,319],{},"物理世界の常識を理解できない","（コップを置けば落ちることを「知らない」）",[315,322,323],{},[23,324,325],{},"長期的な計画・行動が苦手",[315,327,328],{},[23,329,330],{},"ロボット操作などリアルタイム制御に不向き",[19,332,333],{},"世界モデルはこれらを構造的に克服するアプローチで、「考えて動く」AIを実現する鍵として注目されているわけです。",[205,335],{},[32,337,338],{"id":338},"バリューチェーン別の投資アングル",[19,340,341],{},"ここからが本題です。投資判断をするとき、最初に**「どのレイヤーに張るか」**を決めるのが本質的に大事だと思っています。",[54,343,344,356],{},[57,345,346],{},[60,347,348,351,353],{},[63,349,350],{},"レイヤー",[63,352,68],{},[63,354,355],{},"代表企業・銘柄",[70,357,358,371,384,397,410],{},[60,359,360,365,368],{},[75,361,362],{},[23,363,364],{},"基盤モデル",[75,366,367],{},"世界モデル・ロボット基盤AI",[75,369,370],{},"AMI Labs、Physical Intelligence（π）",[60,372,373,378,381],{},[75,374,375],{},[23,376,377],{},"ハードウェア",[75,379,380],{},"センサー、チップ、アクチュエーター",[75,382,383],{},"NVIDIA（NVDA）、Mobileye、Sony Semiconductor",[60,385,386,391,394],{},[75,387,388],{},[23,389,390],{},"ロボット本体",[75,392,393],{},"ヒューマノイド・産業ロボット",[75,395,396],{},"Apptronik、Figure AI、1X、Boston Dynamics",[60,398,399,404,407],{},[75,400,401],{},[23,402,403],{},"プラットフォーム",[75,405,406],{},"データ収集・シミュレーション基盤",[75,408,409],{},"Foxglove、Dyna Robotics、NVIDIA Isaac Sim",[60,411,412,417,420],{},[75,413,414],{},[23,415,416],{},"エンドユーザー",[75,418,419],{},"製造・物流・医療への導入",[75,421,422],{},"Amazon、Tesla、Toyota、Stellantis",[122,424,425],{"id":425},"主要プレイヤー比較",[54,427,428,443],{},[57,429,430],{},[60,431,432,435,438,440],{},[63,433,434],{},"企業",[63,436,437],{},"調達総額",[63,439,109],{},[63,441,442],{},"技術的特徴",[70,444,445,461,477,493,509],{},[60,446,447,452,455,458],{},[75,448,449],{},[23,450,451],{},"AMI Labs",[75,453,454],{},"10.3億ドル",[75,456,457],{},"約35億ドル",[75,459,460],{},"世界モデル（JEPA）、LeCun主導",[60,462,463,468,471,474],{},[75,464,465],{},[23,466,467],{},"Physical Intelligence（π）",[75,469,470],{},"11億ドル",[75,472,473],{},"約56億ドル",[75,475,476],{},"ロボット基盤モデル（π0）、家庭・産業用",[60,478,479,484,487,490],{},[75,480,481],{},[23,482,483],{},"Apptronik",[75,485,486],{},"3.5億ドル以上",[75,488,489],{},"非公開",[75,491,492],{},"ヒューマノイド「Apollo」、NASA連携",[60,494,495,500,503,506],{},[75,496,497],{},[23,498,499],{},"Figure AI",[75,501,502],{},"6.75億ドル",[75,504,505],{},"約26億ドル",[75,507,508],{},"OpenAIとの提携、Figure 01/02",[60,510,511,516,519,521],{},[75,512,513],{},[23,514,515],{},"1X Technologies",[75,517,518],{},"1億ドル以上",[75,520,489],{},[75,522,523],{},"OpenAIバックアップの人型ロボット",[122,525,527],{"id":526},"ami-labs-vs-physical-intelligence-の違い","AMI Labs vs. Physical Intelligence の違い",[134,529,532],{"className":530,"code":531,"language":139},[137],"AMI Labs：\n  └── 「世界を理解するAI」が先。ロボットはその応用の一つ\n  └── 基礎研究主導。JEPAで物理法則の抽象表現を学習\n\nPhysical Intelligence (π)：\n  └── 「ロボットを動かすAI」に特化\n  └── π0モデルをロボットメーカーに提供（OpenAI的ポジション）\n",[141,533,531],{"__ignoreMap":143},[19,535,536,537,544],{},"両社は補完的な関係で、",[23,538,539,540,543],{},"世界モデル（AMI）× ロボット操作モデル（π）",[23,541,542],{},"の組み合わせが、汎用ロボットへの最短経路だと私は見ています。VC的に言うと、ここは","「両建てで持つ」のが安全","で、片方だけにベットする合理性はあまり感じません。",[122,546,547],{"id":547},"個人投資家の現実的アプローチ",[19,549,550],{},"スタートアップへの直接投資は上場前案件が多くて事実上難しいので、二次的な張り方が中心になります。",[552,553,554,560,566,572],"ol",{},[315,555,556,559],{},[23,557,558],{},"NVIDIA（NVDA）"," — フィジカルAIの「ピッケルと鍬」。GPUに加えIsaac Simでエコシステム支配",[315,561,562,565],{},[23,563,564],{},"ロボティクスETF"," — ROBO、ARKQなど（個別株リスクの分散）",[315,567,568,571],{},[23,569,570],{},"トヨタ・ホンダ等の自動車メーカー"," — 自律走行・工場自動化への先行投資",[315,573,574,577],{},[23,575,576],{},"間接受益銘柄"," — センサーメーカー、半導体設計、産業用接続部品",[205,579],{},[32,581,582],{"id":582},"シナリオ別の戦略とウォッチポイント",[19,584,585,586,589],{},"私の中ではここが今回の記事のいちばん大事なところで、フィジカルAIは",[23,587,588],{},"ハードウェア律速で立ち上がる","ことを忘れると判断を誤ります。CVCで電池スタートアップを見てきたので強く思うのですが、ソフトウェアAIと同じ速度・同じバリュエーション倍率を当てると、必ず失望します。",[122,591,592],{"id":592},"シナリオ別ポートフォリオ調整",[54,594,595,615],{},[57,596,597],{},[60,598,599,602,605,607,609,611,613],{},[63,600,601],{},"シナリオ",[63,603,604],{},"確率（私見）",[63,606,364],{},[63,608,377],{},[63,610,390],{},[63,612,403],{},[63,614,416],{},[70,616,617,639,658],{},[60,618,619,622,625,628,631,634,636],{},[75,620,621],{},"A. 漸進普及（ベース）",[75,623,624],{},"50%",[75,626,627],{},"コア配分",[75,629,630],{},"厚め",[75,632,633],{},"様子見",[75,635,627],{},[75,637,638],{},"中立",[60,640,641,644,647,649,651,653,655],{},[75,642,643],{},"B. 量産加速（メイン）",[75,645,646],{},"30%",[75,648,627],{},[75,650,630],{},[75,652,630],{},[75,654,627],{},[75,656,657],{},"上方修正",[60,659,660,663,666,669,671,673,676],{},[75,661,662],{},"C. ハードウェア律速で停滞（テール）",[75,664,665],{},"20%",[75,667,668],{},"縮小",[75,670,638],{},[75,672,668],{},[75,674,675],{},"維持",[75,677,638],{},[679,680,682],"h4",{"id":681},"シナリオa漸進普及ベース","シナリオA：漸進普及（ベース）",[19,684,685,686,689],{},"フィジカルAIは確実に立ち上がるが、ソフトウェアAIほどの直線的成長にはならない展開。この場合、",[23,687,688],{},"NVIDIAとIsaac Simエコシステム","を中心に、トヨタ・Stellantisなどエンドユーザー側の自動化恩恵企業を加えるのが、リスクリターン的には最も筋が良いと見ています。",[679,691,693],{"id":692},"シナリオb量産加速メイン","シナリオB：量産加速（メイン）",[19,695,696],{},"ヒューマノイドの量産単価が想定より早く下がり、エンドユーザーの導入が加速する展開。ここでは**Apptronik、Figure AIに近接する上場企業（Lockheedなど契約パートナー、産業用接続部品メーカー）**にもアロケーションを上げます。",[679,698,700],{"id":699},"シナリオcハードウェア律速で停滞テール","シナリオC：ハードウェア律速で停滞（テール）",[19,702,703,704,707],{},"センサー・アクチュエータのコスト低下が想定通り進まず、量産時期が後ろ倒しになる展開。私が一番警戒しているのは実はここで、",[23,705,706],{},"ロボット本体銘柄のバリュエーション圧縮","が起きやすい局面です。基盤モデル側にも資金調達の冷え込みが波及するので、ロボット本体のエクスポージャを軽くするのが合理的です。",[122,709,710],{"id":710},"リスク要因",[19,712,713],{},"正直、フィジカルAIはリスクも相当あります。",[312,715,716,722,728,734],{},[315,717,718,721],{},[23,719,720],{},"技術的不確実性","：世界モデルはまだ研究段階のものが多い",[315,723,724,727],{},[23,725,726],{},"ハードウェアのボトルネック","：ロボット本体のコスト・耐久性問題",[315,729,730,733],{},[23,731,732],{},"規制・安全基準","：医療・公道での自律機械への規制が整備途上",[315,735,736,739],{},[23,737,738],{},"長い開発サイクル","：ソフトウェアAIと違って、実証・量産に時間がかかる",[19,741,742],{},"CVCで投資委員会を経験した人間として一言だけ付け加えると、**「ハードウェア絡みのDDは、ソフトウェアの倍の時間をかけるのが正解」**です。私自身、追加出資案件でバリュエーション過大評価を指摘して条件交渉を主導した経験があるのですが、ロボティクス案件では特にこの感覚が効くと思っています。",[205,744],{},[32,746,748],{"id":747},"zyl0の投資判断と参入タイミング","ZYL0の投資判断と参入タイミング",[19,750,751],{},"LLMが「情報処理革命」だとすれば、フィジカルAIは**「物理世界との相互作用革命」**です。",[19,753,754,755,758],{},"AMI Labsの10億ドル調達は、単なる一スタートアップの成功ではなく、",[23,756,757],{},"AI産業の重心がデジタル空間から物理空間へと移行する歴史的な転換点のシグナル","だと見ています。",[19,760,761],{},"私の足元のスタンスを一言で言うと、**「基盤モデルとプラットフォームには今コア配分、ロボット本体は2027年に押し目で拾う」**です。",[122,763,764],{"id":764},"今後のウォッチポイント",[312,766,767,773,779,785],{},[315,768,769,772],{},[23,770,771],{},"ヒューマノイド量産単価","が$15K–$20Kラインを実際に割るか",[315,774,775,778],{},[23,776,777],{},"NVIDIA Isaac Simの採用社数","（これはエコシステム支配度の先行指標）",[315,780,781,784],{},[23,782,783],{},"AMI LabsとPhysical Intelligenceの最初の商用契約","（基盤モデルが実用段階に入るシグナル）",[315,786,787,790],{},[23,788,789],{},"米国・欧州のヒューマノイド向け安全規制","（量産化の制約条件）",[792,793,794],"blockquote",{},[19,795,796],{},"フィジカルAIは「SFの世界」ではなく、いま資本が現実に流れ込んでいる投資テーマです。\n基盤モデルとハードウェアのコスト低下が交差する2026〜2028年が、参入タイミングの鍵を握ると見ています。",[205,798],{},[19,800,801],{},[802,803,804],"em",{},"免責事項：本記事は情報提供を目的としており、投資アドバイスを構成するものではありません。すべての投資判断はご自身の判断と、資格を有するファイナンシャルアドバイザーへの相談に基づいて行ってください。",[205,806],{},[10,808,810,814,825,829,832,839,850,912,916,927,933,936,940,994,996,1000,1003,1007,1075,1079,1094,1098,1101,1119,1122,1124,1128,1135,1216,1220,1314,1318,1324,1331,1335,1338,1364,1366,1370,1377,1381,1464,1468,1475,1479,1486,1490,1497,1501,1527,1534,1536,1540,1550,1556,1562,1566,1592,1597,1599],{"lang":809},"en",[14,811,813],{"id":812},"physical-ai-the-next-investment-frontier","Physical AI: The Next Investment Frontier",[19,815,816,817,820,821,824],{},"I've spent years inside a corporate VC working on North American EV second-life storage and semi-solid battery startups, including the cross-functional task of mapping startup technologies into a synergy framework for the parent company. The single biggest lesson from that work: ",[23,818,819],{},"AI domains that touch hardware do not scale on the same clock as software AI",". So when \"Physical AI\" gets called the next big wave, my instinct is to pull it apart by ",[23,822,823],{},"value-chain layer and entry timing"," rather than narrative — and that's what this piece is about.",[32,826,828],{"id":827},"how-to-read-the-center-of-gravity-shift-after-the-llm-boom","How to Read the Center-of-Gravity Shift After the LLM Boom",[19,830,831],{},"Let me first reset the picture flat before adding any view on top.",[19,833,834,835,838],{},"LLMs have dominated the technology landscape for several years. The first generative-AI investment frenzy has settled, and capital is rotating toward the next wave. ",[23,836,837],{},"Physical AI"," — AI that goes beyond text and images to understand and act in the real world — is the leading candidate.",[19,840,841,842,845,846,849],{},"The defining signal of this shift came in early 2026 when Turing Award winner Yann LeCun launched ",[23,843,844],{},"AMI Labs (Advanced Machine Intelligence Labs)",", a Paris-based startup that closed Europe's largest seed round ever. The headline number — over $1 billion — matters less to me than the underlying fact that ",[23,847,848],{},"a top researcher walked away from digital AI to bet on physical AI",".",[54,851,852,862],{},[57,853,854],{},[60,855,856,859],{},[63,857,858],{},"Item",[63,860,861],{},"Detail",[70,863,864,872,880,888,896,904],{},[60,865,866,869],{},[75,867,868],{},"Full name",[75,870,871],{},"Advanced Machine Intelligence Labs (AMI Labs)",[60,873,874,877],{},[75,875,876],{},"Founded",[75,878,879],{},"2026 (Yann LeCun × Alexandre Lebrun)",[60,881,882,885],{},[75,883,884],{},"HQ",[75,886,887],{},"Paris, France",[60,889,890,893],{},[75,891,892],{},"Raised",[75,894,895],{},"$1.03B (Europe's largest seed round ever)",[60,897,898,901],{},[75,899,900],{},"Valuation",[75,902,903],{},"~$3.5B (pre-money)",[60,905,906,909],{},[75,907,908],{},"Key investors",[75,910,911],{},"Cathay Innovation, Greycroft, HV Capital, Toyota Ventures, NVIDIA, Bezos Expeditions, Temasek and others",[122,913,915],{"id":914},"the-core-world-models","The Core: World Models",[19,917,918,919,922,923,926],{},"AMI Labs is building AI that ",[23,920,921],{},"learns how the physical world works"," — what the field calls \"world models.\" It's a fundamentally different design from LLMs, grounded in ",[23,924,925],{},"JEPA (Joint Embedding Predictive Architecture)",", which LeCun proposed in 2022 as an alternative to the autoregressive paradigm.",[134,928,931],{"className":929,"code":930,"language":139},[137],"LLM approach:    Text → Generate more text\nWorld Model:     Sensor data → Understand physical world → Plan actions\n",[141,932,930],{"__ignoreMap":143},[19,934,935],{},"LeCun has long argued that LLMs alone won't reach AGI. AMI Labs is that thesis turned into $1B of conviction.",[122,937,939],{"id":938},"target-applications","Target Applications",[54,941,942,952],{},[57,943,944],{},[60,945,946,949],{},[63,947,948],{},"Domain",[63,950,951],{},"Examples",[70,953,954,962,970,978,986],{},[60,955,956,959],{},[75,957,958],{},"Industrial robotics",[75,960,961],{},"Complex assembly, quality inspection in factories",[60,963,964,967],{},[75,965,966],{},"Autonomous vehicles",[75,968,969],{},"Adapting to unpredictable road environments",[60,971,972,975],{},[75,973,974],{},"Healthcare",[75,976,977],{},"Surgical assist robots, diagnostic support",[60,979,980,983],{},[75,981,982],{},"Wearables",[75,984,985],{},"Context-aware assistive devices",[60,987,988,991],{},[75,989,990],{},"Aerospace",[75,992,993],{},"Autonomous operations in remote environments",[205,995],{},[32,997,999],{"id":998},"why-physical-ai-why-now","Why Physical AI, Why Now?",[19,1001,1002],{},"This is best read through the numbers.",[122,1004,1006],{"id":1005},"market-sizing","Market sizing",[54,1008,1009,1027],{},[57,1010,1011],{},[60,1012,1013,1016,1019,1022,1025],{},[63,1014,1015],{},"Metric",[63,1017,1018],{},"2025",[63,1020,1021],{},"2030 forecast",[63,1023,1024],{},"2033 forecast",[63,1026,237],{},[70,1028,1029,1044,1059],{},[60,1030,1031,1034,1037,1039,1042],{},[75,1032,1033],{},"Physical AI market",[75,1035,1036],{},"$5.2B",[75,1038,250],{},[75,1040,1041],{},"$49.7B",[75,1043,256],{},[60,1045,1046,1049,1052,1055,1057],{},[75,1047,1048],{},"Embodied AI market",[75,1050,1051],{},"$4.4B",[75,1053,1054],{},"$23.1B",[75,1056,250],{},[75,1058,272],{},[60,1060,1061,1064,1067,1070,1073],{},[75,1062,1063],{},"Humanoid shipments (units)",[75,1065,1066],{},"~10K",[75,1068,1069],{},"0.5–1M",[75,1071,1072],{},">2M",[75,1074,289],{},[122,1076,1078],{"id":1077},"humanoids-approach-mass-production","Humanoids approach mass production",[19,1080,1081,1082,1085,1086,1089,1090,1093],{},"Goldman Sachs forecasts global humanoid shipments of ",[23,1083,1084],{},"50,000–100,000 units in 2026",", with per-unit cost trending toward ",[23,1087,1088],{},"$15,000–$20,000",". From having worked on semi-solid battery DDs, I can read between those numbers: what's actually happening underneath is that ",[23,1091,1092],{},"cells, sensors, and actuators are converging on commodity pricing at the same time",". That's the first time in a decade that has been true.",[122,1095,1097],{"id":1096},"the-wall-llms-cannot-climb","The wall LLMs cannot climb",[19,1099,1100],{},"LLMs are extraordinary at language and images, but they have structural limits:",[312,1102,1103,1109,1114],{},[315,1104,1105,1108],{},[23,1106,1107],{},"No physical common sense"," — they don't \"know\" a cup left on a ledge falls",[315,1110,1111],{},[23,1112,1113],{},"Weak long-horizon planning and physical action",[315,1115,1116],{},[23,1117,1118],{},"Poor fit for real-time robotic control",[19,1120,1121],{},"World models are designed to overcome those structurally — that's the source of the optimism around them.",[205,1123],{},[32,1125,1127],{"id":1126},"the-investment-angle-layer-by-layer","The Investment Angle, Layer by Layer",[19,1129,1130,1131,1134],{},"This is where the whole piece comes together. Before picking names, ",[23,1132,1133],{},"decide which layer of the value chain you're betting on",". That's the most important call.",[54,1136,1137,1149],{},[57,1138,1139],{},[60,1140,1141,1144,1147],{},[63,1142,1143],{},"Layer",[63,1145,1146],{},"Description",[63,1148,951],{},[70,1150,1151,1164,1177,1190,1203],{},[60,1152,1153,1158,1161],{},[75,1154,1155],{},[23,1156,1157],{},"Foundation models",[75,1159,1160],{},"World models, robot foundation AI",[75,1162,1163],{},"AMI Labs, Physical Intelligence (π)",[60,1165,1166,1171,1174],{},[75,1167,1168],{},[23,1169,1170],{},"Hardware",[75,1172,1173],{},"Sensors, chips, actuators",[75,1175,1176],{},"NVIDIA (NVDA), Mobileye, Sony Semiconductor",[60,1178,1179,1184,1187],{},[75,1180,1181],{},[23,1182,1183],{},"Robots",[75,1185,1186],{},"Humanoid, industrial machines",[75,1188,1189],{},"Apptronik, Figure AI, 1X, Boston Dynamics",[60,1191,1192,1197,1200],{},[75,1193,1194],{},[23,1195,1196],{},"Platforms",[75,1198,1199],{},"Data infra, simulation, ops",[75,1201,1202],{},"Foxglove, Dyna Robotics, NVIDIA Isaac Sim",[60,1204,1205,1210,1213],{},[75,1206,1207],{},[23,1208,1209],{},"End users",[75,1211,1212],{},"Manufacturing, logistics, healthcare",[75,1214,1215],{},"Amazon, Tesla, Toyota, Stellantis",[122,1217,1219],{"id":1218},"competitive-landscape","Competitive landscape",[54,1221,1222,1237],{},[57,1223,1224],{},[60,1225,1226,1229,1232,1234],{},[63,1227,1228],{},"Company",[63,1230,1231],{},"Total raised",[63,1233,900],{},[63,1235,1236],{},"Technology",[70,1238,1239,1254,1270,1285,1300],{},[60,1240,1241,1245,1248,1251],{},[75,1242,1243],{},[23,1244,451],{},[75,1246,1247],{},"$1.03B",[75,1249,1250],{},"~$3.5B",[75,1252,1253],{},"World models (JEPA), LeCun-led",[60,1255,1256,1261,1264,1267],{},[75,1257,1258],{},[23,1259,1260],{},"Physical Intelligence (π)",[75,1262,1263],{},"$1.1B",[75,1265,1266],{},"~$5.6B",[75,1268,1269],{},"Robot foundation model (π0), home/industry",[60,1271,1272,1276,1279,1282],{},[75,1273,1274],{},[23,1275,483],{},[75,1277,1278],{},"$350M+",[75,1280,1281],{},"undisclosed",[75,1283,1284],{},"\"Apollo\" humanoid, NASA partnership",[60,1286,1287,1291,1294,1297],{},[75,1288,1289],{},[23,1290,499],{},[75,1292,1293],{},"$675M",[75,1295,1296],{},"~$2.6B",[75,1298,1299],{},"OpenAI partnership, Figure 01/02",[60,1301,1302,1306,1309,1311],{},[75,1303,1304],{},[23,1305,515],{},[75,1307,1308],{},"$100M+",[75,1310,1281],{},[75,1312,1313],{},"OpenAI-backed humanoids",[122,1315,1317],{"id":1316},"ami-labs-vs-physical-intelligence","AMI Labs vs Physical Intelligence",[134,1319,1322],{"className":1320,"code":1321,"language":139},[137],"AMI Labs:\n  └── \"Understand the world\" first; robots are one application\n  └── Research-led. JEPA learns abstract physical representations\n\nPhysical Intelligence (π):\n  └── Specifically \"make robots act\"\n  └── Provides the π0 model to robot manufacturers (an OpenAI-style position)\n",[141,1323,1321],{"__ignoreMap":143},[19,1325,1326,1327,1330],{},"These two are complementary, not competitive. The combination of ",[23,1328,1329],{},"world models (AMI) × action models (π)"," is, in my view, the shortest path to general-purpose robots. A VC perspective: hold both. The case for picking only one is weak.",[122,1332,1334],{"id":1333},"realistic-strategies-for-individual-investors","Realistic strategies for individual investors",[19,1336,1337],{},"Direct startup access is mostly closed at the pre-IPO stage, so secondary exposure dominates.",[552,1339,1340,1346,1352,1358],{},[315,1341,1342,1345],{},[23,1343,1344],{},"NVIDIA (NVDA)"," — picks-and-shovels. GPUs plus Isaac Sim's ecosystem dominance.",[315,1347,1348,1351],{},[23,1349,1350],{},"Robotics ETFs"," — ROBO, ARKQ (single-name risk dilution).",[315,1353,1354,1357],{},[23,1355,1356],{},"Auto OEMs"," — Toyota, Honda, Stellantis (autonomous + factory automation upside).",[315,1359,1360,1363],{},[23,1361,1362],{},"Indirect beneficiaries"," — sensor makers, chip designers, industrial connectors.",[205,1365],{},[32,1367,1369],{"id":1368},"scenario-based-strategy-watch-items","Scenario-Based Strategy & Watch Items",[19,1371,1372,1373,1376],{},"This is the most important part of this piece, in my opinion: Physical AI is ",[23,1374,1375],{},"hardware-rate-limited",". Forget that, and you'll mis-price the timing. Having watched battery startups inside a CVC, I'd say it strongly: do not apply software-AI valuation multiples or growth curves to Physical AI without adjustment.",[122,1378,1380],{"id":1379},"scenario-allocation-matrix","Scenario allocation matrix",[54,1382,1383,1404],{},[57,1384,1385],{},[60,1386,1387,1390,1393,1395,1397,1400,1402],{},[63,1388,1389],{},"Scenario",[63,1391,1392],{},"Probability",[63,1394,1157],{},[63,1396,1170],{},[63,1398,1399],{},"Robot bodies",[63,1401,1196],{},[63,1403,1209],{},[70,1405,1406,1427,1445],{},[60,1407,1408,1411,1413,1416,1419,1422,1424],{},[75,1409,1410],{},"A. Gradual adoption (base)",[75,1412,624],{},[75,1414,1415],{},"Core",[75,1417,1418],{},"Heavy",[75,1420,1421],{},"Wait & see",[75,1423,1415],{},[75,1425,1426],{},"Neutral",[60,1428,1429,1432,1434,1436,1438,1440,1442],{},[75,1430,1431],{},"B. Mass-production acceleration (main)",[75,1433,646],{},[75,1435,1415],{},[75,1437,1418],{},[75,1439,1418],{},[75,1441,1415],{},[75,1443,1444],{},"Upgrade",[60,1446,1447,1450,1452,1455,1457,1459,1462],{},[75,1448,1449],{},"C. Hardware bottleneck stalls scaling (tail)",[75,1451,665],{},[75,1453,1454],{},"Trim",[75,1456,1426],{},[75,1458,1454],{},[75,1460,1461],{},"Maintain",[75,1463,1426],{},[679,1465,1467],{"id":1466},"scenario-a-gradual-adoption-base","Scenario A — Gradual adoption (base)",[19,1469,1470,1471,1474],{},"Physical AI scales reliably, but not as straight a line as software AI. The cleanest risk-reward in this regime, in my view, is ",[23,1472,1473],{},"NVIDIA + Isaac Sim ecosystem players",", plus end-user automation beneficiaries (Toyota, Stellantis).",[679,1476,1478],{"id":1477},"scenario-b-mass-production-acceleration-main","Scenario B — Mass-production acceleration (main)",[19,1480,1481,1482,1485],{},"Humanoid unit costs fall faster than expected, end-user adoption accelerates. Here you also tilt into ",[23,1483,1484],{},"public-market companies adjacent to Apptronik / Figure AI"," — defense contracting partners, industrial connector makers.",[679,1487,1489],{"id":1488},"scenario-c-hardware-bottleneck-stalls-scaling-tail","Scenario C — Hardware bottleneck stalls scaling (tail)",[19,1491,1492,1493,1496],{},"Sensor and actuator cost declines miss the schedule, mass production slips. This is the scenario I personally watch most closely, because it's where ",[23,1494,1495],{},"valuation compression on robot-body names"," is most likely. Funding chill propagates back to the foundation-model layer, so trim robot-body exposure preemptively.",[122,1498,1500],{"id":1499},"risk-factors","Risk factors",[312,1502,1503,1509,1515,1521],{},[315,1504,1505,1508],{},[23,1506,1507],{},"Technical uncertainty"," — most world models are still research-stage",[315,1510,1511,1514],{},[23,1512,1513],{},"Hardware bottlenecks"," — robot body cost, durability, reliability",[315,1516,1517,1520],{},[23,1518,1519],{},"Regulation"," — autonomous machines in healthcare and public spaces face evolving rules",[315,1522,1523,1526],{},[23,1524,1525],{},"Long development cycles"," — unlike software AI, physical systems require validation and manufacturing scale-up",[19,1528,1529,1530,1533],{},"One thing from CVC investment-committee experience: ",[23,1531,1532],{},"hardware-touching DD takes twice as long as pure software DD, period",". I've personally been the person who flagged an over-valued hardware deal at IC and led the renegotiation, and that instinct applies double in robotics.",[205,1535],{},[32,1537,1539],{"id":1538},"the-zyl0-investment-call-entry-timing","The ZYL0 Investment Call & Entry Timing",[19,1541,1542,1543,1546,1547,849],{},"If LLMs were a revolution in ",[23,1544,1545],{},"information processing",", Physical AI is a revolution in ",[23,1548,1549],{},"interaction with the physical world",[19,1551,1552,1553,849],{},"AMI Labs' $1B raise isn't the success story of one startup — it's a signal that the ",[23,1554,1555],{},"center of gravity of AI is migrating from digital to physical space",[19,1557,1558,1559,849],{},"In one line: ",[23,1560,1561],{},"core allocation now to foundation models and platforms; pick up robot-body exposure on a 2027 dip",[122,1563,1565],{"id":1564},"watch-items-going-forward","Watch items going forward",[312,1567,1568,1574,1580,1586],{},[315,1569,1570,1573],{},[23,1571,1572],{},"Humanoid unit cost"," — does it actually break the $15K–$20K line?",[315,1575,1576,1579],{},[23,1577,1578],{},"NVIDIA Isaac Sim adoption count"," (a leading indicator of ecosystem dominance)",[315,1581,1582,1585],{},[23,1583,1584],{},"First commercial contracts from AMI Labs and Physical Intelligence"," (signals foundation models entering production)",[315,1587,1588,1591],{},[23,1589,1590],{},"US/EU safety regulations for humanoids"," (the gating constraint on mass deployment)",[792,1593,1594],{},[19,1595,1596],{},"Physical AI is not science fiction — it is an investment theme where real capital is flowing right now.\nThe convergence of declining hardware costs and maturing foundation models makes 2026–2028 the critical window to watch.",[205,1598],{},[19,1600,1601],{},[802,1602,1603],{},"Disclaimer: This article is for informational purposes only and does not constitute investment advice. All investment decisions should be made based on your own judgment and in consultation with a qualified financial advisor.",{"title":143,"searchDepth":1605,"depth":1605,"links":1606},2,[1607,1612,1617,1622,1626,1629,1633,1638,1643,1647],{"id":34,"depth":1605,"text":35,"children":1608},[1609,1611],{"id":124,"depth":1610,"text":125},3,{"id":149,"depth":1610,"text":149},{"id":209,"depth":1605,"text":210,"children":1613},[1614,1615,1616],{"id":216,"depth":1610,"text":216},{"id":292,"depth":1610,"text":292},{"id":306,"depth":1610,"text":307},{"id":338,"depth":1605,"text":338,"children":1618},[1619,1620,1621],{"id":425,"depth":1610,"text":425},{"id":526,"depth":1610,"text":527},{"id":547,"depth":1610,"text":547},{"id":582,"depth":1605,"text":582,"children":1623},[1624,1625],{"id":592,"depth":1610,"text":592},{"id":710,"depth":1610,"text":710},{"id":747,"depth":1605,"text":748,"children":1627},[1628],{"id":764,"depth":1610,"text":764},{"id":827,"depth":1605,"text":828,"children":1630},[1631,1632],{"id":914,"depth":1610,"text":915},{"id":938,"depth":1610,"text":939},{"id":998,"depth":1605,"text":999,"children":1634},[1635,1636,1637],{"id":1005,"depth":1610,"text":1006},{"id":1077,"depth":1610,"text":1078},{"id":1096,"depth":1610,"text":1097},{"id":1126,"depth":1605,"text":1127,"children":1639},[1640,1641,1642],{"id":1218,"depth":1610,"text":1219},{"id":1316,"depth":1610,"text":1317},{"id":1333,"depth":1610,"text":1334},{"id":1368,"depth":1605,"text":1369,"children":1644},[1645,1646],{"id":1379,"depth":1610,"text":1380},{"id":1499,"depth":1610,"text":1500},{"id":1538,"depth":1605,"text":1539,"children":1648},[1649],{"id":1564,"depth":1610,"text":1565},"Yann LeCunのAMI Labsを軸に、フィジカルAI・世界モデルが次の巨大投資テーマとなる理由を再点検。バリューチェーン別の投資アングル、参入タイミング、シナリオ別ポートフォリオ調整までを整理する。","md",{"date":1653,"image":1654,"alt":1655,"tags":1656,"tagsEn":1661,"published":1666},"21st Mar 2026","/blogs-img/blog-physical-ai.png","フィジカルAIと世界モデル：次世代投資テーマの全貌",[44,1657,1658,1659,1660],"AI投資","ロボティクス","世界モデル","スタートアップ",[837,1662,1663,1664,1665],"AI Investing","Robotics","World Models","Startups",true,"/blogs/2-physical-ai-next-investment-theme",{"title":5,"description":1650},"blogs/2-physical-ai-next-investment-theme","5Pl6--8ppjE9dHgKDp_u8dVhrfPiXqxiCuyC6PQOBIo",{"id":4,"title":5,"body":1672,"description":1650,"extension":1651,"meta":2858,"navigation":1666,"ogImage":1654,"path":1667,"seo":2861,"stem":1669,"__hash__":1670},{"type":7,"value":1673,"toc":2814},[1674,2236,2238],[10,1675,1676,1678,1684,1686,1688,1692,1696,1744,1746,1750,1755,1757,1759,1801,1803,1805,1807,1809,1863,1865,1871,1873,1875,1889,1891,1893,1895,1897,1961,1963,2039,2041,2046,2052,2054,2056,2074,2076,2078,2082,2084,2154,2156,2160,2162,2164,2166,2170,2172,2174,2192,2194,2196,2198,2200,2204,2206,2208,2226,2230,2232],{"lang":12},[14,1677,17],{"id":16},[19,1679,21,1680,30],{},[23,1681,25,1682,29],{},[23,1683,28],{},[32,1685,35],{"id":34},[19,1687,38],{},[19,1689,41,1690,45],{},[23,1691,44],{},[19,1693,48,1694,52],{},[23,1695,51],{},[54,1697,1698,1706],{},[57,1699,1700],{},[60,1701,1702,1704],{},[63,1703,65],{},[63,1705,68],{},[70,1707,1708,1714,1720,1726,1732,1738],{},[60,1709,1710,1712],{},[75,1711,77],{},[75,1713,80],{},[60,1715,1716,1718],{},[75,1717,85],{},[75,1719,88],{},[60,1721,1722,1724],{},[75,1723,93],{},[75,1725,96],{},[60,1727,1728,1730],{},[75,1729,101],{},[75,1731,104],{},[60,1733,1734,1736],{},[75,1735,109],{},[75,1737,112],{},[60,1739,1740,1742],{},[75,1741,117],{},[75,1743,120],{},[122,1745,125],{"id":124},[19,1747,128,1748,132],{},[23,1749,131],{},[134,1751,1753],{"className":1752,"code":138,"language":139},[137],[141,1754,138],{"__ignoreMap":143},[19,1756,146],{},[122,1758,149],{"id":149},[54,1760,1761,1769],{},[57,1762,1763],{},[60,1764,1765,1767],{},[63,1766,158],{},[63,1768,161],{},[70,1770,1771,1777,1783,1789,1795],{},[60,1772,1773,1775],{},[75,1774,168],{},[75,1776,171],{},[60,1778,1779,1781],{},[75,1780,176],{},[75,1782,179],{},[60,1784,1785,1787],{},[75,1786,184],{},[75,1788,187],{},[60,1790,1791,1793],{},[75,1792,192],{},[75,1794,195],{},[60,1796,1797,1799],{},[75,1798,200],{},[75,1800,203],{},[205,1802],{},[32,1804,210],{"id":209},[19,1806,213],{},[122,1808,216],{"id":216},[54,1810,1811,1825],{},[57,1812,1813],{},[60,1814,1815,1817,1819,1821,1823],{},[63,1816,225],{},[63,1818,228],{},[63,1820,231],{},[63,1822,234],{},[63,1824,237],{},[70,1826,1827,1839,1851],{},[60,1828,1829,1831,1833,1835,1837],{},[75,1830,244],{},[75,1832,247],{},[75,1834,250],{},[75,1836,253],{},[75,1838,256],{},[60,1840,1841,1843,1845,1847,1849],{},[75,1842,261],{},[75,1844,264],{},[75,1846,267],{},[75,1848,250],{},[75,1850,272],{},[60,1852,1853,1855,1857,1859,1861],{},[75,1854,277],{},[75,1856,280],{},[75,1858,283],{},[75,1860,286],{},[75,1862,289],{},[122,1864,292],{"id":292},[19,1866,295,1867,299,1869,303],{},[23,1868,298],{},[23,1870,302],{},[122,1872,307],{"id":306},[19,1874,310],{},[312,1876,1877,1881,1885],{},[315,1878,1879,320],{},[23,1880,319],{},[315,1882,1883],{},[23,1884,325],{},[315,1886,1887],{},[23,1888,330],{},[19,1890,333],{},[205,1892],{},[32,1894,338],{"id":338},[19,1896,341],{},[54,1898,1899,1909],{},[57,1900,1901],{},[60,1902,1903,1905,1907],{},[63,1904,350],{},[63,1906,68],{},[63,1908,355],{},[70,1910,1911,1921,1931,1941,1951],{},[60,1912,1913,1917,1919],{},[75,1914,1915],{},[23,1916,364],{},[75,1918,367],{},[75,1920,370],{},[60,1922,1923,1927,1929],{},[75,1924,1925],{},[23,1926,377],{},[75,1928,380],{},[75,1930,383],{},[60,1932,1933,1937,1939],{},[75,1934,1935],{},[23,1936,390],{},[75,1938,393],{},[75,1940,396],{},[60,1942,1943,1947,1949],{},[75,1944,1945],{},[23,1946,403],{},[75,1948,406],{},[75,1950,409],{},[60,1952,1953,1957,1959],{},[75,1954,1955],{},[23,1956,416],{},[75,1958,419],{},[75,1960,422],{},[122,1962,425],{"id":425},[54,1964,1965,1977],{},[57,1966,1967],{},[60,1968,1969,1971,1973,1975],{},[63,1970,434],{},[63,1972,437],{},[63,1974,109],{},[63,1976,442],{},[70,1978,1979,1991,2003,2015,2027],{},[60,1980,1981,1985,1987,1989],{},[75,1982,1983],{},[23,1984,451],{},[75,1986,454],{},[75,1988,457],{},[75,1990,460],{},[60,1992,1993,1997,1999,2001],{},[75,1994,1995],{},[23,1996,467],{},[75,1998,470],{},[75,2000,473],{},[75,2002,476],{},[60,2004,2005,2009,2011,2013],{},[75,2006,2007],{},[23,2008,483],{},[75,2010,486],{},[75,2012,489],{},[75,2014,492],{},[60,2016,2017,2021,2023,2025],{},[75,2018,2019],{},[23,2020,499],{},[75,2022,502],{},[75,2024,505],{},[75,2026,508],{},[60,2028,2029,2033,2035,2037],{},[75,2030,2031],{},[23,2032,515],{},[75,2034,518],{},[75,2036,489],{},[75,2038,523],{},[122,2040,527],{"id":526},[134,2042,2044],{"className":2043,"code":531,"language":139},[137],[141,2045,531],{"__ignoreMap":143},[19,2047,536,2048,544],{},[23,2049,539,2050,543],{},[23,2051,542],{},[122,2053,547],{"id":547},[19,2055,550],{},[552,2057,2058,2062,2066,2070],{},[315,2059,2060,559],{},[23,2061,558],{},[315,2063,2064,565],{},[23,2065,564],{},[315,2067,2068,571],{},[23,2069,570],{},[315,2071,2072,577],{},[23,2073,576],{},[205,2075],{},[32,2077,582],{"id":582},[19,2079,585,2080,589],{},[23,2081,588],{},[122,2083,592],{"id":592},[54,2085,2086,2104],{},[57,2087,2088],{},[60,2089,2090,2092,2094,2096,2098,2100,2102],{},[63,2091,601],{},[63,2093,604],{},[63,2095,364],{},[63,2097,377],{},[63,2099,390],{},[63,2101,403],{},[63,2103,416],{},[70,2105,2106,2122,2138],{},[60,2107,2108,2110,2112,2114,2116,2118,2120],{},[75,2109,621],{},[75,2111,624],{},[75,2113,627],{},[75,2115,630],{},[75,2117,633],{},[75,2119,627],{},[75,2121,638],{},[60,2123,2124,2126,2128,2130,2132,2134,2136],{},[75,2125,643],{},[75,2127,646],{},[75,2129,627],{},[75,2131,630],{},[75,2133,630],{},[75,2135,627],{},[75,2137,657],{},[60,2139,2140,2142,2144,2146,2148,2150,2152],{},[75,2141,662],{},[75,2143,665],{},[75,2145,668],{},[75,2147,638],{},[75,2149,668],{},[75,2151,675],{},[75,2153,638],{},[679,2155,682],{"id":681},[19,2157,685,2158,689],{},[23,2159,688],{},[679,2161,693],{"id":692},[19,2163,696],{},[679,2165,700],{"id":699},[19,2167,703,2168,707],{},[23,2169,706],{},[122,2171,710],{"id":710},[19,2173,713],{},[312,2175,2176,2180,2184,2188],{},[315,2177,2178,721],{},[23,2179,720],{},[315,2181,2182,727],{},[23,2183,726],{},[315,2185,2186,733],{},[23,2187,732],{},[315,2189,2190,739],{},[23,2191,738],{},[19,2193,742],{},[205,2195],{},[32,2197,748],{"id":747},[19,2199,751],{},[19,2201,754,2202,758],{},[23,2203,757],{},[19,2205,761],{},[122,2207,764],{"id":764},[312,2209,2210,2214,2218,2222],{},[315,2211,2212,772],{},[23,2213,771],{},[315,2215,2216,778],{},[23,2217,777],{},[315,2219,2220,784],{},[23,2221,783],{},[315,2223,2224,790],{},[23,2225,789],{},[792,2227,2228],{},[19,2229,796],{},[205,2231],{},[19,2233,2234],{},[802,2235,804],{},[205,2237],{},[10,2239,2240,2242,2248,2250,2252,2256,2262,2310,2312,2318,2323,2325,2327,2369,2371,2373,2375,2377,2431,2433,2441,2443,2445,2459,2461,2463,2465,2469,2533,2535,2611,2613,2618,2622,2624,2626,2644,2646,2648,2652,2654,2724,2726,2730,2732,2736,2738,2742,2744,2762,2766,2768,2770,2776,2780,2784,2786,2804,2808,2810],{"lang":809},[14,2241,813],{"id":812},[19,2243,816,2244,820,2246,824],{},[23,2245,819],{},[23,2247,823],{},[32,2249,828],{"id":827},[19,2251,831],{},[19,2253,834,2254,838],{},[23,2255,837],{},[19,2257,841,2258,845,2260,849],{},[23,2259,844],{},[23,2261,848],{},[54,2263,2264,2272],{},[57,2265,2266],{},[60,2267,2268,2270],{},[63,2269,858],{},[63,2271,861],{},[70,2273,2274,2280,2286,2292,2298,2304],{},[60,2275,2276,2278],{},[75,2277,868],{},[75,2279,871],{},[60,2281,2282,2284],{},[75,2283,876],{},[75,2285,879],{},[60,2287,2288,2290],{},[75,2289,884],{},[75,2291,887],{},[60,2293,2294,2296],{},[75,2295,892],{},[75,2297,895],{},[60,2299,2300,2302],{},[75,2301,900],{},[75,2303,903],{},[60,2305,2306,2308],{},[75,2307,908],{},[75,2309,911],{},[122,2311,915],{"id":914},[19,2313,918,2314,922,2316,926],{},[23,2315,921],{},[23,2317,925],{},[134,2319,2321],{"className":2320,"code":930,"language":139},[137],[141,2322,930],{"__ignoreMap":143},[19,2324,935],{},[122,2326,939],{"id":938},[54,2328,2329,2337],{},[57,2330,2331],{},[60,2332,2333,2335],{},[63,2334,948],{},[63,2336,951],{},[70,2338,2339,2345,2351,2357,2363],{},[60,2340,2341,2343],{},[75,2342,958],{},[75,2344,961],{},[60,2346,2347,2349],{},[75,2348,966],{},[75,2350,969],{},[60,2352,2353,2355],{},[75,2354,974],{},[75,2356,977],{},[60,2358,2359,2361],{},[75,2360,982],{},[75,2362,985],{},[60,2364,2365,2367],{},[75,2366,990],{},[75,2368,993],{},[205,2370],{},[32,2372,999],{"id":998},[19,2374,1002],{},[122,2376,1006],{"id":1005},[54,2378,2379,2393],{},[57,2380,2381],{},[60,2382,2383,2385,2387,2389,2391],{},[63,2384,1015],{},[63,2386,1018],{},[63,2388,1021],{},[63,2390,1024],{},[63,2392,237],{},[70,2394,2395,2407,2419],{},[60,2396,2397,2399,2401,2403,2405],{},[75,2398,1033],{},[75,2400,1036],{},[75,2402,250],{},[75,2404,1041],{},[75,2406,256],{},[60,2408,2409,2411,2413,2415,2417],{},[75,2410,1048],{},[75,2412,1051],{},[75,2414,1054],{},[75,2416,250],{},[75,2418,272],{},[60,2420,2421,2423,2425,2427,2429],{},[75,2422,1063],{},[75,2424,1066],{},[75,2426,1069],{},[75,2428,1072],{},[75,2430,289],{},[122,2432,1078],{"id":1077},[19,2434,1081,2435,1085,2437,1089,2439,1093],{},[23,2436,1084],{},[23,2438,1088],{},[23,2440,1092],{},[122,2442,1097],{"id":1096},[19,2444,1100],{},[312,2446,2447,2451,2455],{},[315,2448,2449,1108],{},[23,2450,1107],{},[315,2452,2453],{},[23,2454,1113],{},[315,2456,2457],{},[23,2458,1118],{},[19,2460,1121],{},[205,2462],{},[32,2464,1127],{"id":1126},[19,2466,1130,2467,1134],{},[23,2468,1133],{},[54,2470,2471,2481],{},[57,2472,2473],{},[60,2474,2475,2477,2479],{},[63,2476,1143],{},[63,2478,1146],{},[63,2480,951],{},[70,2482,2483,2493,2503,2513,2523],{},[60,2484,2485,2489,2491],{},[75,2486,2487],{},[23,2488,1157],{},[75,2490,1160],{},[75,2492,1163],{},[60,2494,2495,2499,2501],{},[75,2496,2497],{},[23,2498,1170],{},[75,2500,1173],{},[75,2502,1176],{},[60,2504,2505,2509,2511],{},[75,2506,2507],{},[23,2508,1183],{},[75,2510,1186],{},[75,2512,1189],{},[60,2514,2515,2519,2521],{},[75,2516,2517],{},[23,2518,1196],{},[75,2520,1199],{},[75,2522,1202],{},[60,2524,2525,2529,2531],{},[75,2526,2527],{},[23,2528,1209],{},[75,2530,1212],{},[75,2532,1215],{},[122,2534,1219],{"id":1218},[54,2536,2537,2549],{},[57,2538,2539],{},[60,2540,2541,2543,2545,2547],{},[63,2542,1228],{},[63,2544,1231],{},[63,2546,900],{},[63,2548,1236],{},[70,2550,2551,2563,2575,2587,2599],{},[60,2552,2553,2557,2559,2561],{},[75,2554,2555],{},[23,2556,451],{},[75,2558,1247],{},[75,2560,1250],{},[75,2562,1253],{},[60,2564,2565,2569,2571,2573],{},[75,2566,2567],{},[23,2568,1260],{},[75,2570,1263],{},[75,2572,1266],{},[75,2574,1269],{},[60,2576,2577,2581,2583,2585],{},[75,2578,2579],{},[23,2580,483],{},[75,2582,1278],{},[75,2584,1281],{},[75,2586,1284],{},[60,2588,2589,2593,2595,2597],{},[75,2590,2591],{},[23,2592,499],{},[75,2594,1293],{},[75,2596,1296],{},[75,2598,1299],{},[60,2600,2601,2605,2607,2609],{},[75,2602,2603],{},[23,2604,515],{},[75,2606,1308],{},[75,2608,1281],{},[75,2610,1313],{},[122,2612,1317],{"id":1316},[134,2614,2616],{"className":2615,"code":1321,"language":139},[137],[141,2617,1321],{"__ignoreMap":143},[19,2619,1326,2620,1330],{},[23,2621,1329],{},[122,2623,1334],{"id":1333},[19,2625,1337],{},[552,2627,2628,2632,2636,2640],{},[315,2629,2630,1345],{},[23,2631,1344],{},[315,2633,2634,1351],{},[23,2635,1350],{},[315,2637,2638,1357],{},[23,2639,1356],{},[315,2641,2642,1363],{},[23,2643,1362],{},[205,2645],{},[32,2647,1369],{"id":1368},[19,2649,1372,2650,1376],{},[23,2651,1375],{},[122,2653,1380],{"id":1379},[54,2655,2656,2674],{},[57,2657,2658],{},[60,2659,2660,2662,2664,2666,2668,2670,2672],{},[63,2661,1389],{},[63,2663,1392],{},[63,2665,1157],{},[63,2667,1170],{},[63,2669,1399],{},[63,2671,1196],{},[63,2673,1209],{},[70,2675,2676,2692,2708],{},[60,2677,2678,2680,2682,2684,2686,2688,2690],{},[75,2679,1410],{},[75,2681,624],{},[75,2683,1415],{},[75,2685,1418],{},[75,2687,1421],{},[75,2689,1415],{},[75,2691,1426],{},[60,2693,2694,2696,2698,2700,2702,2704,2706],{},[75,2695,1431],{},[75,2697,646],{},[75,2699,1415],{},[75,2701,1418],{},[75,2703,1418],{},[75,2705,1415],{},[75,2707,1444],{},[60,2709,2710,2712,2714,2716,2718,2720,2722],{},[75,2711,1449],{},[75,2713,665],{},[75,2715,1454],{},[75,2717,1426],{},[75,2719,1454],{},[75,2721,1461],{},[75,2723,1426],{},[679,2725,1467],{"id":1466},[19,2727,1470,2728,1474],{},[23,2729,1473],{},[679,2731,1478],{"id":1477},[19,2733,1481,2734,1485],{},[23,2735,1484],{},[679,2737,1489],{"id":1488},[19,2739,1492,2740,1496],{},[23,2741,1495],{},[122,2743,1500],{"id":1499},[312,2745,2746,2750,2754,2758],{},[315,2747,2748,1508],{},[23,2749,1507],{},[315,2751,2752,1514],{},[23,2753,1513],{},[315,2755,2756,1520],{},[23,2757,1519],{},[315,2759,2760,1526],{},[23,2761,1525],{},[19,2763,1529,2764,1533],{},[23,2765,1532],{},[205,2767],{},[32,2769,1539],{"id":1538},[19,2771,1542,2772,1546,2774,849],{},[23,2773,1545],{},[23,2775,1549],{},[19,2777,1552,2778,849],{},[23,2779,1555],{},[19,2781,1558,2782,849],{},[23,2783,1561],{},[122,2785,1565],{"id":1564},[312,2787,2788,2792,2796,2800],{},[315,2789,2790,1573],{},[23,2791,1572],{},[315,2793,2794,1579],{},[23,2795,1578],{},[315,2797,2798,1585],{},[23,2799,1584],{},[315,2801,2802,1591],{},[23,2803,1590],{},[792,2805,2806],{},[19,2807,1596],{},[205,2809],{},[19,2811,2812],{},[802,2813,1603],{},{"title":143,"searchDepth":1605,"depth":1605,"links":2815},[2816,2820,2825,2830,2834,2837,2841,2846,2851,2855],{"id":34,"depth":1605,"text":35,"children":2817},[2818,2819],{"id":124,"depth":1610,"text":125},{"id":149,"depth":1610,"text":149},{"id":209,"depth":1605,"text":210,"children":2821},[2822,2823,2824],{"id":216,"depth":1610,"text":216},{"id":292,"depth":1610,"text":292},{"id":306,"depth":1610,"text":307},{"id":338,"depth":1605,"text":338,"children":2826},[2827,2828,2829],{"id":425,"depth":1610,"text":425},{"id":526,"depth":1610,"text":527},{"id":547,"depth":1610,"text":547},{"id":582,"depth":1605,"text":582,"children":2831},[2832,2833],{"id":592,"depth":1610,"text":592},{"id":710,"depth":1610,"text":710},{"id":747,"depth":1605,"text":748,"children":2835},[2836],{"id":764,"depth":1610,"text":764},{"id":827,"depth":1605,"text":828,"children":2838},[2839,2840],{"id":914,"depth":1610,"text":915},{"id":938,"depth":1610,"text":939},{"id":998,"depth":1605,"text":999,"children":2842},[2843,2844,2845],{"id":1005,"depth":1610,"text":1006},{"id":1077,"depth":1610,"text":1078},{"id":1096,"depth":1610,"text":1097},{"id":1126,"depth":1605,"text":1127,"children":2847},[2848,2849,2850],{"id":1218,"depth":1610,"text":1219},{"id":1316,"depth":1610,"text":1317},{"id":1333,"depth":1610,"text":1334},{"id":1368,"depth":1605,"text":1369,"children":2852},[2853,2854],{"id":1379,"depth":1610,"text":1380},{"id":1499,"depth":1610,"text":1500},{"id":1538,"depth":1605,"text":1539,"children":2856},[2857],{"id":1564,"depth":1610,"text":1565},{"date":1653,"image":1654,"alt":1655,"tags":2859,"tagsEn":2860,"published":1666},[44,1657,1658,1659,1660],[837,1662,1663,1664,1665],{"title":5,"description":1650},[2863,4250,5606,7101,7654,10011,10644,11835,13445,14697,16117,17992,20281,21877],{"id":2864,"title":2865,"body":2866,"description":4233,"extension":1651,"meta":4234,"navigation":1666,"ogImage":4235,"path":4246,"seo":4247,"stem":4248,"__hash__":4249},"content/blogs/1-iran-us-war-stock-outlook.md","イラン・米国緊張が揺るがす米国株の行方 | US Stock Market Outlook: Iran-US Tensions",{"type":7,"value":2867,"toc":4201},[2868,3532,3534],[10,2869,2870,2874,2877,2881,2884,2887,2958,2965,2967,2971,2974,2977,3084,3087,3107,3114,3116,3120,3123,3126,3130,3137,3186,3190,3193,3219,3223,3230,3233,3237,3240,3244,3251,3255,3262,3264,3267,3270,3341,3348,3350,3353,3356,3358,3450,3454,3457,3461,3468,3472,3475,3477,3481,3484,3491,3493,3519,3526,3528],{"lang":12},[14,2871,2873],{"id":2872},"イラン米国緊張が揺るがす米国株の行方","イラン・米国緊張が揺るがす米国株の行方",[19,2875,2876],{},"私はCVCで海洋・エネルギー領域のディール分析を続けてきた人間です。中東情勢は「ニュースの見出し」ではなく、原油・LNG・運賃・配船パターンが連動して動く一個の系として見るのが現実的だと思っています。今回イランと米国の緊張が再燃した局面で、その視点でポートフォリオをどう組み替えるべきかを、自分の実務感覚も交えて書いてみます。",[32,2878,2880],{"id":2879},"いま市場で起きている中東プレミアムの正体","いま市場で起きている「中東プレミアム」の正体",[19,2882,2883],{},"ここで一度、市場が織り込み始めたリスクの中身をフラットに眺めておきたいです。",[19,2885,2886],{},"JCPOA崩壊以降、イラン・米国は「断続的な緊張＋限定的衝突」という、どっちつかずの状態が常態化してきましたよね。2026年初からは、複数の要因が同時並行で悪化していて、これが市場のリスクプレミアムを押し上げている、というのが現状の見立てです。",[54,2888,2889,2901],{},[57,2890,2891],{},[60,2892,2893,2896,2898],{},[63,2894,2895],{},"要因",[63,2897,68],{},[63,2899,2900],{},"市場への含意",[70,2902,2903,2914,2925,2936,2947],{},[60,2904,2905,2908,2911],{},[75,2906,2907],{},"核開発加速",[75,2909,2910],{},"濃縮ウラン蓄積量が国際的警戒水準を超過",[75,2912,2913],{},"制裁強化観測 → 原油・防衛株プラス",[60,2915,2916,2919,2922],{},[75,2917,2918],{},"ホルムズ海峡の緊張",[75,2920,2921],{},"世界の原油輸送の約20%が通過",[75,2923,2924],{},"海上保険料・運賃上昇",[60,2926,2927,2930,2933],{},[75,2928,2929],{},"米軍の中東増強",[75,2931,2932],{},"空母打撃群派遣・地域基地強化",[75,2934,2935],{},"防衛株モメンタム",[60,2937,2938,2941,2944],{},[75,2939,2940],{},"代理勢力の活発化",[75,2942,2943],{},"フーシ派・ヒズボラ等のエスカレーション",[75,2945,2946],{},"サプライチェーン分断リスク",[60,2948,2949,2952,2955],{},[75,2950,2951],{},"外交チャネルの閉塞",[75,2953,2954],{},"直接対話の欠如、第三国経由の間接交渉も停滞",[75,2956,2957],{},"不確実性プレミアムが剥がれにくい",[19,2959,2960,2961,2964],{},"特に",[23,2962,2963],{},"海上保険料の上昇","は私の中では今回いちばん「実体」に近いシグナルだと感じていて、ここが上抜けてくると、ニュースよりも先に船社・トレーダーの行動が変わり始めます。VLCC・LNG船の用船料スプレッドが広がり始めたら、株式市場よりも一足早く、コモディティ寄りのセクターに資金が動いている可能性が高い、と私は読んでいます。",[205,2966],{},[32,2968,2970],{"id":2969},"歴史が示す地政学ショックの賞味期限","歴史が示す「地政学ショックの賞味期限」",[19,2972,2973],{},"地政学イベントは派手に見えるわりに、株式市場では意外と短命です——というのが、過去の中東危機を一通り並べたときに最初に感じることでした。",[122,2975,2976],{"id":2976},"主要事例の株価反応",[54,2978,2979,2998],{},[57,2980,2981],{},[60,2982,2983,2986,2989,2992,2995],{},[63,2984,2985],{},"出来事",[63,2987,2988],{},"発生時期",[63,2990,2991],{},"初期反応（S&P500）",[63,2993,2994],{},"3ヶ月後",[63,2996,2997],{},"1年後",[70,2999,3000,3017,3033,3050,3067],{},[60,3001,3002,3005,3008,3011,3014],{},[75,3003,3004],{},"湾岸戦争勃発",[75,3006,3007],{},"1990年8月",[75,3009,3010],{},"−17%",[75,3012,3013],{},"−5%",[75,3015,3016],{},"+27%",[60,3018,3019,3022,3025,3028,3031],{},[75,3020,3021],{},"9.11同時多発テロ",[75,3023,3024],{},"2001年9月",[75,3026,3027],{},"−12%",[75,3029,3030],{},"−15%",[75,3032,3027],{},[60,3034,3035,3038,3041,3044,3047],{},[75,3036,3037],{},"イラク戦争開始",[75,3039,3040],{},"2003年3月",[75,3042,3043],{},"+3%（開戦後上昇）",[75,3045,3046],{},"+15%",[75,3048,3049],{},"+33%",[60,3051,3052,3055,3058,3061,3064],{},[75,3053,3054],{},"イラン核危機",[75,3056,3057],{},"2012年2月",[75,3059,3060],{},"−3%",[75,3062,3063],{},"+9%",[75,3065,3066],{},"+22%",[60,3068,3069,3072,3075,3078,3081],{},[75,3070,3071],{},"ソレイマニ暗殺",[75,3073,3074],{},"2020年1月",[75,3076,3077],{},"−0.5%",[75,3079,3080],{},"+4%",[75,3082,3083],{},"+17%",[19,3085,3086],{},"並べて見えるパターンは以下の3つです。",[552,3088,3089,3095,3101],{},[315,3090,3091,3094],{},[23,3092,3093],{},"初期のショック売り"," — 開戦前後に短期的な下落",[315,3096,3097,3100],{},[23,3098,3099],{},"「不確実性が剥がれた瞬間」の急反発"," — 開戦＝確定情報、として買い戻しが入る",[315,3102,3103,3106],{},[23,3104,3105],{},"長期では底堅い"," — 1年後はほぼ全ての事例でプラスリターン",[19,3108,3109,3110,3113],{},"ここで個人的に重要だと思っているのは、",[23,3111,3112],{},"「初期反応の大きさ」と「1年後のリターン」が逆相関に近い","という点です。9.11以外は、ショックが大きいほど1年後のリターンも大きい。短期で恐怖指数が跳ねている局面ほど、長期投資家にとっては入口になりやすい——というのは、過去のリターン分布から経験的に読み取れる、わりと固いパターンだと感じます。",[205,3115],{},[32,3117,3119],{"id":3118},"セクター別の効きどころを再点検する","セクター別の「効きどころ」を再点検する",[19,3121,3122],{},"セクター反応にも、毎回ほぼ同じ顔ぶれが並ぶのが面白いところです。",[122,3124,3125],{"id":3125},"恩恵を受けるセクター",[679,3127,3129],{"id":3128},"エネルギー原油天然ガス","エネルギー（原油・天然ガス）",[19,3131,3132,3133,3136],{},"ホルムズ海峡が部分的にでも機能不全になると、原油価格は上にレンジが切り替わります。私はLNG運賃データを見ていた人間なので余計に思うのですが、シェール上流よりも、",[23,3134,3135],{},"輸送・ターミナル・LNGリクィッファクション","といった「輸送ボトルネック」側の方が今回はリプライシングされやすい局面だと見ています。",[54,3138,3139,3151],{},[57,3140,3141],{},[60,3142,3143,3145,3148],{},[63,3144,601],{},[63,3146,3147],{},"原油価格レンジ",[63,3149,3150],{},"受益銘柄例",[70,3152,3153,3164,3175],{},[60,3154,3155,3158,3161],{},[75,3156,3157],{},"緊張継続（現状維持）",[75,3159,3160],{},"$80〜90/バレル",[75,3162,3163],{},"ExxonMobil, Chevron",[60,3165,3166,3169,3172],{},[75,3167,3168],{},"局地的衝突",[75,3170,3171],{},"$100〜120/バレル",[75,3173,3174],{},"ConocoPhillips, Pioneer, Cheniere",[60,3176,3177,3180,3183],{},[75,3178,3179],{},"ホルムズ封鎖",[75,3181,3182],{},"$150+/バレル",[75,3184,3185],{},"米シェール上流＋LNG輸出インフラ全般",[679,3187,3189],{"id":3188},"防衛航空宇宙","防衛・航空宇宙",[19,3191,3192],{},"ここはわりと教科書通りの動きをしやすいセクターという印象です。私はCVCで防衛関連スタートアップへのDDも経験しているのですが、有事局面では「契約スピード」が露骨に変わるので、四半期売上の前倒し計上が起きやすく、足元でモメンタムが出やすい。",[312,3194,3195,3201,3207,3213],{},[315,3196,3197,3200],{},[23,3198,3199],{},"Lockheed Martin（LMT）"," — F-35・ミサイル防衛",[315,3202,3203,3206],{},[23,3204,3205],{},"RTX","（旧Raytheon）— 精密誘導兵器・防空",[315,3208,3209,3212],{},[23,3210,3211],{},"Northrop Grumman（NOC）"," — 戦略爆撃機・宇宙防衛",[315,3214,3215,3218],{},[23,3216,3217],{},"L3Harris（LHX）"," — 通信・電子戦",[679,3220,3222],{"id":3221},"金貴金属","金・貴金属",[19,3224,3225,3226,3229],{},"地政学リスクが金価格を押し上げるのはほぼコンセンサスですが、私が今回少し違うかもと思っているのは、",[23,3227,3228],{},"実質金利との連動が以前より弱まっている","ことです。中央銀行の金購入が需要側を底上げしていて、米長期金利が上がる局面でも金が崩れにくい。Newmont、Barrick、GLDあたりは、従来の「保険」よりも一歩踏み込んだコア配分として持つ価値があると私は見ています。",[122,3231,3232],{"id":3232},"逆風にさらされるセクター",[679,3234,3236],{"id":3235},"航空旅行","航空・旅行",[19,3238,3239],{},"中東路線の運航停止と燃料コスト上昇のダブルパンチですよね。航空会社のコスト構造を見直すと、**燃料費は営業費用の25〜30%**を占めていて、ここが原油$120を超えると、ヘッジしていない会社は1〜2四半期で営業利益率が3〜5pt押し下げられる計算になります。Delta・American・United、いずれも避けたい局面です。",[679,3241,3243],{"id":3242},"消費財小売","消費財・小売",[19,3245,3246,3247,3250],{},"原油高の波及は遅れてやってきます。私が見ている範囲では、薄利の小売は",[23,3248,3249],{},"3〜4四半期遅れ","でグロスマージン悪化が表面化するので、「最初の決算ではセーフだった」と油断するのが一番危ないパターン。コスト転嫁できる強いブランドと、できない汎用小売をきちんと分けるのが大事です。",[679,3252,3254],{"id":3253},"ハイグローステック","ハイグロース・テック",[19,3256,3257,3258,3261],{},"長期金利が上昇すると割引率が上がり、PERの高いグロース株から資金が流出する——というのはほぼセオリーですが、今回特に注意したいのは、",[23,3259,3260],{},"AIキャペックスで株価が走ってきた銘柄群","です。地政学ストレスは「在庫を厚く積む」インセンティブを強化するので、半導体は短期的にむしろポジティブに働く可能性もあって、ここは一律に売るとミスを誘発しやすい局面だと思っています。",[205,3263],{},[32,3265,3266],{"id":3266},"マーケット指標で読む現在地",[19,3268,3269],{},"数字で頭を冷やすのが、有事局面でいちばんワークします。",[54,3271,3272,3284],{},[57,3273,3274],{},[60,3275,3276,3278,3281],{},[63,3277,225],{},[63,3279,3280],{},"意味",[63,3282,3283],{},"警戒ライン",[70,3285,3286,3297,3308,3319,3330],{},[60,3287,3288,3291,3294],{},[75,3289,3290],{},"VIX（恐怖指数）",[75,3292,3293],{},"市場の不安心理",[75,3295,3296],{},"25以上で警戒、40超で極度の恐怖",[60,3298,3299,3302,3305],{},[75,3300,3301],{},"原油先物（WTI）",[75,3303,3304],{},"エネルギー需給",[75,3306,3307],{},"$100超でstagflationリスク再浮上",[60,3309,3310,3313,3316],{},[75,3311,3312],{},"金価格",[75,3314,3315],{},"安全資産需要",[75,3317,3318],{},"$2,500超は強い逃避買い",[60,3320,3321,3324,3327],{},[75,3322,3323],{},"10年債利回り",[75,3325,3326],{},"リスク回避度",[75,3328,3329],{},"急低下はリセッション懸念",[60,3331,3332,3335,3338],{},[75,3333,3334],{},"ドル円",[75,3336,3337],{},"リスクオフ指標",[75,3339,3340],{},"円高は世界的リスク回避の証",[19,3342,3343,3344,3347],{},"私が一番先に動くと考えているのは、実は",[23,3345,3346],{},"WTIではなくBrentと中東向けタンカー運賃（BDTI）のスプレッド","です。ここが3週間で20%以上広がると、株式市場の感応度は一段上がる、というのが過去の経験則。表に出ない\"先行指標\"として置いておく価値があります。",[205,3349],{},[32,3351,3352],{"id":3352},"投資家へのシナリオ別戦略",[19,3354,3355],{},"ここはCVC的に言うと「ベース・メイン・テールの3シナリオ」で組むのが王道です。",[122,3357,592],{"id":592},[54,3359,3360,3383],{},[57,3361,3362],{},[60,3363,3364,3366,3368,3371,3374,3377,3380],{},[63,3365,601],{},[63,3367,604],{},[63,3369,3370],{},"エネルギー",[63,3372,3373],{},"防衛",[63,3375,3376],{},"金ETF",[63,3378,3379],{},"テック",[63,3381,3382],{},"現金",[70,3384,3385,3408,3427],{},[60,3386,3387,3390,3393,3396,3399,3402,3405],{},[75,3388,3389],{},"A. 外交緊張継続（ベース）",[75,3391,3392],{},"55%",[75,3394,3395],{},"+5〜10%",[75,3397,3398],{},"+3〜5%",[75,3400,3401],{},"+2〜3%",[75,3403,3404],{},"やや減",[75,3406,3407],{},"現状維持",[60,3409,3410,3413,3415,3418,3420,3423,3425],{},[75,3411,3412],{},"B. 局地的軍事衝突（メイン）",[75,3414,646],{},[75,3416,3417],{},"+10%",[75,3419,3417],{},[75,3421,3422],{},"+5%",[75,3424,3013],{},[75,3426,3422],{},[60,3428,3429,3432,3435,3438,3441,3444,3447],{},[75,3430,3431],{},"C. 全面戦争（テール）",[75,3433,3434],{},"15%",[75,3436,3437],{},"コモディティ全般",[75,3439,3440],{},"+12%",[75,3442,3443],{},"+5〜7%",[75,3445,3446],{},"−10〜15%",[75,3448,3449],{},"20〜30%",[679,3451,3453],{"id":3452},"シナリオa外交的緊張が継続ベースケース","シナリオA：外交的緊張が継続（ベースケース）",[19,3455,3456],{},"不確実性は続くが全面衝突には至らない——個人的にはこれが最もありうる展開だと見ています。エネルギー5〜10%、防衛3〜5%、GLD2〜3%を加え、テック比率をやや下げて配当株・ディフェンシブにシフトする、というのが現状のセオリー的な答えです。",[679,3458,3460],{"id":3459},"シナリオb局地的軍事衝突リスクケース","シナリオB：局地的軍事衝突（リスクケース）",[19,3462,3463,3464,3467],{},"ホルムズ海峡での限定衝突や米軍施設への攻撃。ここまで来るとボラは大きく上がりますが、",[23,3465,3466],{},"VIX急騰局面を「買い場」と認識する","ためには事前の戦略が要る、というのが過去の事例から繰り返し示されているところです。原油・LNG関連株を積み増し、防衛比率を10%超に引き上げ、株式比率全体を落として短期国債と金にシフトします。",[679,3469,3471],{"id":3470},"シナリオc全面戦争テールリスク","シナリオC：全面戦争（テールリスク）",[19,3473,3474],{},"可能性は高くないですが、起きた場合の影響が極端なので、保険として組み込んでおく価値はあります。現金20〜30%、地理的分散（欧州・アジア）、コモディティ全般への配分増加、オプションでのダウンサイドヘッジ、というのが現状の組み合わせです。",[205,3476],{},[32,3478,3480],{"id":3479},"zyl0の投資判断と今後のウォッチポイント","ZYL0の投資判断と今後のウォッチポイント",[19,3482,3483],{},"私の足元のスタンスを一言で言うと、**「短期は中立〜ややディフェンシブ、中期は押し目買いに備えて段階的に攻める」**です。",[19,3485,3486,3487,3490],{},"過去の事例が一貫して示すのは、地政学リスクは株式市場にとって「短期的な試練」であり、「長期的な敗因」にはならないということ。CVCの仕事を通じて学んだのは、",[23,3488,3489],{},"パニックで動くより、シナリオ通りに淡々と動く側に立つ方が、最終的なリターンは確実に良くなる","ということでした。",[122,3492,764],{"id":764},[312,3494,3495,3501,3507,3513],{},[315,3496,3497,3500],{},[23,3498,3499],{},"中東向けタンカー運賃（BDTI）"," が3週間で20%以上拡大するか",[315,3502,3503,3506],{},[23,3504,3505],{},"米長期金利","が4.5%超で粘る局面が続くか（→金とテックのリプライシング）",[315,3508,3509,3512],{},[23,3510,3511],{},"米国防予算の補正","が議会で動き始めるか（防衛セクターの上方修正トリガー）",[315,3514,3515,3518],{},[23,3516,3517],{},"OPEC+の対応スピード"," — イランからの供給減を相殺するスペアキャパが足りるか",[19,3520,3521,3522,3525],{},"恐れて逃げるリスクではなく、",[23,3523,3524],{},"アクティブに対処するリスク","として向き合うのが、有事局面の投資家の仕事だと私は思っています。",[205,3527],{},[19,3529,3530],{},[802,3531,804],{},[205,3533],{},[10,3535,3536,3540,3543,3547,3550,3553,3624,3631,3633,3637,3640,3644,3738,3741,3765,3772,3774,3778,3784,3788,3792,3799,3846,3850,3857,3882,3886,3897,3901,3905,3912,3916,3923,3927,3938,3940,3944,3947,4019,4026,4028,4032,4035,4037,4125,4129,4132,4136,4143,4147,4150,4152,4156,4161,4164,4166,4192,4195,4197],{"lang":809},[14,3537,3539],{"id":3538},"us-stock-market-outlook-iran-us-tensions","US Stock Market Outlook: Iran-US Tensions",[19,3541,3542],{},"I've spent years inside a corporate VC analyzing maritime and energy deals — From that vantage point, Middle East tensions don't read as a headline story. They read as a single coupled system in which crude, LNG, freight rates, and routing patterns all move together. So when Iran-US tensions flare again, my instinct is to walk through portfolio adjustments with that lens — and that's what this piece is built around.",[32,3544,3546],{"id":3545},"what-the-markets-middle-east-premium-is-actually-pricing-in","What the Market's \"Middle East Premium\" Is Actually Pricing In",[19,3548,3549],{},"Let me first reset the picture flat, before adding any view on top.",[19,3551,3552],{},"Since the JCPOA's collapse, US-Iran relations have settled into a steady-state of \"intermittent tension plus limited skirmishes.\" What's different in 2026 is that several factors have started worsening in parallel, and that's what's lifting the market's risk premium right now.",[54,3554,3555,3567],{},[57,3556,3557],{},[60,3558,3559,3562,3564],{},[63,3560,3561],{},"Factor",[63,3563,861],{},[63,3565,3566],{},"Market implication",[70,3568,3569,3580,3591,3602,3613],{},[60,3570,3571,3574,3577],{},[75,3572,3573],{},"Iranian nuclear acceleration",[75,3575,3576],{},"Enriched-uranium stockpiles exceeding international red lines",[75,3578,3579],{},"Sanctions tightening → oil & defense bid",[60,3581,3582,3585,3588],{},[75,3583,3584],{},"Strait of Hormuz stress",[75,3586,3587],{},"Strategic chokepoint for ~20% of global oil shipments",[75,3589,3590],{},"Marine insurance & freight rates rising",[60,3592,3593,3596,3599],{},[75,3594,3595],{},"US military buildup",[75,3597,3598],{},"Carrier strike group deployment, regional base reinforcement",[75,3600,3601],{},"Defense-stock momentum",[60,3603,3604,3607,3610],{},[75,3605,3606],{},"Proxy escalation",[75,3608,3609],{},"Houthis, Hezbollah and others ramping up",[75,3611,3612],{},"Supply-chain disruption risk",[60,3614,3615,3618,3621],{},[75,3616,3617],{},"Diplomatic deadlock",[75,3619,3620],{},"No direct talks, indirect channels stalled",[75,3622,3623],{},"Uncertainty premium hard to compress",[19,3625,3626,3627,3630],{},"The signal I personally weight most is ",[23,3628,3629],{},"marine insurance pricing",". Insurance moves before the headlines do — when VLCC and LNG charter spreads start widening, traders and shipping companies have already adjusted their behavior, and equity markets typically lag that move by several weeks.",[205,3632],{},[32,3634,3636],{"id":3635},"what-history-teaches-about-the-shelf-life-of-geopolitical-shocks","What History Teaches About the Shelf Life of Geopolitical Shocks",[19,3638,3639],{},"Geopolitical events look loud, but they have surprisingly short shelf lives in equity markets — that's the first thing you notice when you line them up.",[122,3641,3643],{"id":3642},"historical-sp-500-reactions","Historical S&P 500 Reactions",[54,3645,3646,3665],{},[57,3647,3648],{},[60,3649,3650,3653,3656,3659,3662],{},[63,3651,3652],{},"Event",[63,3654,3655],{},"Date",[63,3657,3658],{},"Initial S&P 500",[63,3660,3661],{},"3 months",[63,3663,3664],{},"1 year",[70,3666,3667,3681,3695,3710,3724],{},[60,3668,3669,3672,3675,3677,3679],{},[75,3670,3671],{},"Gulf War outbreak",[75,3673,3674],{},"Aug 1990",[75,3676,3010],{},[75,3678,3013],{},[75,3680,3016],{},[60,3682,3683,3686,3689,3691,3693],{},[75,3684,3685],{},"9/11 attacks",[75,3687,3688],{},"Sep 2001",[75,3690,3027],{},[75,3692,3030],{},[75,3694,3027],{},[60,3696,3697,3700,3703,3706,3708],{},[75,3698,3699],{},"Iraq War start",[75,3701,3702],{},"Mar 2003",[75,3704,3705],{},"+3% (rallied on start)",[75,3707,3046],{},[75,3709,3049],{},[60,3711,3712,3715,3718,3720,3722],{},[75,3713,3714],{},"Iran nuclear crisis",[75,3716,3717],{},"Feb 2012",[75,3719,3060],{},[75,3721,3063],{},[75,3723,3066],{},[60,3725,3726,3729,3732,3734,3736],{},[75,3727,3728],{},"Soleimani assassination",[75,3730,3731],{},"Jan 2020",[75,3733,3077],{},[75,3735,3080],{},[75,3737,3083],{},[19,3739,3740],{},"Three patterns:",[552,3742,3743,3749,3759],{},[315,3744,3745,3748],{},[23,3746,3747],{},"Initial shock selloff"," — short-term decline around the outbreak",[315,3750,3751,3754,3755,3758],{},[23,3752,3753],{},"Sharp rebound once uncertainty lifts"," — the moment \"war begins\" is paradoxically a ",[802,3756,3757],{},"removal"," of uncertainty, and markets usually rally",[315,3760,3761,3764],{},[23,3762,3763],{},"Long-term resilience"," — nearly every case shows positive returns one year out",[19,3766,3767,3768,3771],{},"The relationship I think gets undersold is the ",[23,3769,3770],{},"near-inverse correlation between initial drawdown and 1-year return",". With the exception of 9/11, the bigger the initial shock, the bigger the year-out gain. Periods of peak fear tend to function as long-term entry windows — that's a reasonably stable pattern, not a fluke.",[205,3773],{},[32,3775,3777],{"id":3776},"sector-sensitivities-re-checked","Sector Sensitivities, Re-checked",[19,3779,3780,3781,849],{},"The same sector cast tends to show up every time. What changes is which player benefits ",[802,3782,3783],{},"most",[122,3785,3787],{"id":3786},"sectors-that-benefit","Sectors that benefit",[679,3789,3791],{"id":3790},"energy-oil-natural-gas","Energy (oil & natural gas)",[19,3793,3794,3795,3798],{},"If the Strait of Hormuz becomes even partially unreliable, the entire oil price band resets higher. Having spent a lot of time staring at LNG freight curves, my intuition is that the part of the chain most likely to re-rate this cycle isn't shale upstream — it's the ",[23,3796,3797],{},"transport, terminal, and liquefaction layer"," where bottleneck pricing sits.",[54,3800,3801,3813],{},[57,3802,3803],{},[60,3804,3805,3807,3810],{},[63,3806,1389],{},[63,3808,3809],{},"Crude band",[63,3811,3812],{},"Beneficiaries",[70,3814,3815,3825,3835],{},[60,3816,3817,3820,3823],{},[75,3818,3819],{},"Sustained tension",[75,3821,3822],{},"$80–90/bbl",[75,3824,3163],{},[60,3826,3827,3830,3833],{},[75,3828,3829],{},"Localized conflict",[75,3831,3832],{},"$100–120/bbl",[75,3834,3174],{},[60,3836,3837,3840,3843],{},[75,3838,3839],{},"Hormuz closure",[75,3841,3842],{},"$150+/bbl",[75,3844,3845],{},"US shale upstream + LNG export infra",[679,3847,3849],{"id":3848},"defense-aerospace","Defense & aerospace",[19,3851,3852,3853,3856],{},"This is the most textbook of the cohort. From the CVC seat I've watched defense-adjacent startups during prior tension spikes, and the most consistent pattern is ",[23,3854,3855],{},"contracting cadence",": bookings get pulled forward, and that shows up early in revenue prints.",[312,3858,3859,3865,3870,3876],{},[315,3860,3861,3864],{},[23,3862,3863],{},"Lockheed Martin (LMT)"," — F-35, missile defense",[315,3866,3867,3869],{},[23,3868,3205],{}," — precision-guided munitions, air defense",[315,3871,3872,3875],{},[23,3873,3874],{},"Northrop Grumman (NOC)"," — strategic bombers, space defense",[315,3877,3878,3881],{},[23,3879,3880],{},"L3Harris (LHX)"," — comms, electronic warfare",[679,3883,3885],{"id":3884},"gold-precious-metals","Gold & precious metals",[19,3887,3888,3889,3892,3893,3896],{},"Gold catching a geopolitical bid is consensus. What I think is ",[802,3890,3891],{},"less"," appreciated this cycle is that ",[23,3894,3895],{},"gold's correlation with real rates has weakened"," — central bank buying has put a structural floor under demand, so gold doesn't crack as easily when long rates rise. Newmont, Barrick, and GLD increasingly behave like a core allocation, not just an insurance line.",[122,3898,3900],{"id":3899},"sectors-facing-headwinds","Sectors facing headwinds",[679,3902,3904],{"id":3903},"airlines-travel","Airlines & travel",[19,3906,3907,3908,3911],{},"Suspended Middle East routes plus higher fuel costs is a clean double hit. Fuel runs ",[23,3909,3910],{},"25–30% of operating costs",", and once crude clears $120, unhedged carriers tend to compress operating margin by 3–5 points within one or two quarters.",[679,3913,3915],{"id":3914},"consumer-goods-retail","Consumer goods & retail",[19,3917,3918,3919,3922],{},"The oil-price ripple arrives late. From what I've seen, thin-margin retail typically shows gross-margin damage on a ",[23,3920,3921],{},"3–4 quarter lag",", which is exactly the trap — the first earnings print \"looks fine,\" and that lulls people into staying long. The right cut is brand pricing power vs. commodity retail.",[679,3924,3926],{"id":3925},"high-growth-tech","High-growth tech",[19,3928,3929,3930,3933,3934,3937],{},"Higher long rates compress high-multiple growth stocks — that's the textbook. The wrinkle this cycle is the ",[23,3931,3932],{},"AI capex cohort",". Geopolitical stress strengthens the incentive to ",[802,3935,3936],{},"over-stock"," inventory, which means semiconductors can paradoxically benefit short-term, even within a \"tech\" basket. Selling the whole growth tape indiscriminately is the easiest way to lose money in this regime.",[205,3939],{},[32,3941,3943],{"id":3942},"market-indicators-to-watch","Market Indicators to Watch",[19,3945,3946],{},"Numbers are the cheapest way to keep your head clear when the news cycle gets loud.",[54,3948,3949,3962],{},[57,3950,3951],{},[60,3952,3953,3956,3959],{},[63,3954,3955],{},"Indicator",[63,3957,3958],{},"Meaning",[63,3960,3961],{},"Alert level",[70,3963,3964,3975,3986,3997,4008],{},[60,3965,3966,3969,3972],{},[75,3967,3968],{},"VIX (fear index)",[75,3970,3971],{},"Market anxiety",[75,3973,3974],{},"25+ = caution, 40+ = extreme fear",[60,3976,3977,3980,3983],{},[75,3978,3979],{},"WTI crude",[75,3981,3982],{},"Energy supply/demand",[75,3984,3985],{},"$100+ revives stagflation risk",[60,3987,3988,3991,3994],{},[75,3989,3990],{},"Gold price",[75,3992,3993],{},"Safe-haven demand",[75,3995,3996],{},"$2,500+ = strong flight to safety",[60,3998,3999,4002,4005],{},[75,4000,4001],{},"10-year Treasury yield",[75,4003,4004],{},"Risk appetite",[75,4006,4007],{},"Sharp drop = recession fears",[60,4009,4010,4013,4016],{},[75,4011,4012],{},"USD/JPY",[75,4014,4015],{},"Global risk-off signal",[75,4017,4018],{},"Yen strength = broad risk aversion",[19,4020,4021,4022,4025],{},"The instrument I personally watch first is actually the ",[23,4023,4024],{},"Brent–BDTI spread",": when the Middle-East tanker rate index gains more than 20% in three weeks, equity-market sensitivity steps up sharply. It's a useful \"off-screen\" leading indicator.",[205,4027],{},[32,4029,4031],{"id":4030},"scenario-based-strategy","Scenario-Based Strategy",[19,4033,4034],{},"CVC discipline says you build for three scenarios — base, main, tail — and adjust as the road forks.",[122,4036,1380],{"id":1379},[54,4038,4039,4063],{},[57,4040,4041],{},[60,4042,4043,4045,4048,4051,4054,4057,4060],{},[63,4044,1389],{},[63,4046,4047],{},"Probability (my view)",[63,4049,4050],{},"Energy",[63,4052,4053],{},"Defense",[63,4055,4056],{},"Gold ETF",[63,4058,4059],{},"Tech",[63,4061,4062],{},"Cash",[70,4064,4065,4087,4104],{},[60,4066,4067,4070,4072,4075,4078,4081,4084],{},[75,4068,4069],{},"A. Sustained diplomatic tension",[75,4071,3392],{},[75,4073,4074],{},"+5–10%",[75,4076,4077],{},"+3–5%",[75,4079,4080],{},"+2–3%",[75,4082,4083],{},"slight under",[75,4085,4086],{},"flat",[60,4088,4089,4092,4094,4096,4098,4100,4102],{},[75,4090,4091],{},"B. Localized military conflict",[75,4093,646],{},[75,4095,3417],{},[75,4097,3417],{},[75,4099,3422],{},[75,4101,3013],{},[75,4103,3422],{},[60,4105,4106,4109,4111,4114,4116,4119,4122],{},[75,4107,4108],{},"C. Full-scale war (tail)",[75,4110,3434],{},[75,4112,4113],{},"broad commodities",[75,4115,3440],{},[75,4117,4118],{},"+5–7%",[75,4120,4121],{},"−10–15%",[75,4123,4124],{},"20–30%",[679,4126,4128],{"id":4127},"scenario-a-sustained-tension-base","Scenario A — Sustained tension (base)",[19,4130,4131],{},"Uncertainty persists, no full-scale conflict. Add 5–10% energy, 3–5% defense, 2–3% GLD; trim tech modestly toward dividends and defensives. This is the path I currently weight highest.",[679,4133,4135],{"id":4134},"scenario-b-localized-conflict-main-risk","Scenario B — Localized conflict (main risk)",[19,4137,4138,4139,4142],{},"Limited Hormuz clashes or strikes on US installations. Volatility steps up — but ",[23,4140,4141],{},"VIX spikes only become buying opportunities if you've pre-committed to a plan",", which historical drawdowns repeatedly remind us. Add to oil & LNG, push defense above 10%, take down equity exposure broadly, rotate into short Treasuries and gold.",[679,4144,4146],{"id":4145},"scenario-c-full-scale-war-tail","Scenario C — Full-scale war (tail)",[19,4148,4149],{},"Low probability, large impact, so it has to be carried as a hedge. Cash 20–30%, geographic diversification (Europe, Asia), broader commodity exposure, and options for downside hedging.",[205,4151],{},[32,4153,4155],{"id":4154},"the-zyl0-investment-call-watch-items","The ZYL0 Investment Call & Watch Items",[19,4157,1558,4158,849],{},[23,4159,4160],{},"short-term neutral-to-defensive, medium-term staged buyer on dips",[19,4162,4163],{},"What history reliably shows is that geopolitical risk is a short-term trial for equity markets, not a long-term defeat. The lesson I keep coming back to from CVC work is that the investors who beat panic are the ones who pre-committed to a plan and executed it mechanically — emotional outperformance is rarely real outperformance.",[122,4165,1565],{"id":1564},[312,4167,4168,4174,4180,4186],{},[315,4169,4170,4173],{},[23,4171,4172],{},"Middle East tanker rates (BDTI)"," — does the index expand >20% over a three-week window?",[315,4175,4176,4179],{},[23,4177,4178],{},"US 10-year yield"," — does it stay sticky above 4.5%? (Repricing trigger for gold and tech)",[315,4181,4182,4185],{},[23,4183,4184],{},"US defense budget supplements"," — does Congress start moving incremental authorizations? (Defense-sector upgrade trigger)",[315,4187,4188,4191],{},[23,4189,4190],{},"OPEC+ response speed"," — is there enough spare capacity to offset Iranian-supply disruption?",[19,4193,4194],{},"This is risk to manage actively, not to flee. That framing — and the discipline to act on it — is, to me, what separates investors who survive Middle-East cycles from investors who get repeatedly chopped by them.",[205,4196],{},[19,4198,4199],{},[802,4200,1603],{},{"title":143,"searchDepth":1605,"depth":1605,"links":4202},[4203,4204,4207,4211,4212,4215,4218,4219,4222,4226,4227,4230],{"id":2879,"depth":1605,"text":2880},{"id":2969,"depth":1605,"text":2970,"children":4205},[4206],{"id":2976,"depth":1610,"text":2976},{"id":3118,"depth":1605,"text":3119,"children":4208},[4209,4210],{"id":3125,"depth":1610,"text":3125},{"id":3232,"depth":1610,"text":3232},{"id":3266,"depth":1605,"text":3266},{"id":3352,"depth":1605,"text":3352,"children":4213},[4214],{"id":592,"depth":1610,"text":592},{"id":3479,"depth":1605,"text":3480,"children":4216},[4217],{"id":764,"depth":1610,"text":764},{"id":3545,"depth":1605,"text":3546},{"id":3635,"depth":1605,"text":3636,"children":4220},[4221],{"id":3642,"depth":1610,"text":3643},{"id":3776,"depth":1605,"text":3777,"children":4223},[4224,4225],{"id":3786,"depth":1610,"text":3787},{"id":3899,"depth":1610,"text":3900},{"id":3942,"depth":1605,"text":3943},{"id":4030,"depth":1605,"text":4031,"children":4228},[4229],{"id":1379,"depth":1610,"text":1380},{"id":4154,"depth":1605,"text":4155,"children":4231},[4232],{"id":1564,"depth":1610,"text":1565},"イラン・米国の軍事的緊張が米国株に与える影響を、過去の中東危機の値動きと現在のセクター別感応度から再点検。エネルギー・防衛・安全資産の比率をどこまで動かすか、シナリオ別に整理する。",{"date":1653,"image":4235,"alt":4236,"tags":4237,"tagsEn":4242,"published":1666},"/blogs-img/blog-iran-us-war.png","地政学的リスクと市場の見通し：イラン・米国緊張の影響",[4238,4239,4240,3370,4241],"米国株","地政学リスク","投資","Market Trend",[4243,4244,4245,4050,4241],"US Stocks","Geopolitical Risk","Investing","/blogs/1-iran-us-war-stock-outlook",{"title":2865,"description":4233},"blogs/1-iran-us-war-stock-outlook","k-8I2Q8gB1IbcQLLjPYImkZoyR8ODtmtxVO9qqplh-8",{"id":4251,"title":4252,"body":4253,"description":5593,"extension":1651,"meta":5594,"navigation":1666,"ogImage":5596,"path":5602,"seo":5603,"stem":5604,"__hash__":5605},"content/blogs/10-gpt55-vs-claude-mythos.md","GPT-5.5は「仕事で使えるAI」——だがMythosの背中は遠い | GPT-5.5 Is Finally \"AI That Works at Work\" — But Mythos Is Still Far Ahead",{"type":7,"value":4254,"toc":5565},[4255,4908,4910],[10,4256,4257,4261,4268,4272,4275,4286,4289,4300,4302,4306,4309,4313,4316,4323,4334,4341,4345,4352,4371,4378,4385,4392,4395,4399,4406,4413,4420,4423,4568,4570,4574,4581,4588,4595,4675,4678,4685,4687,4691,4698,4701,4704,4711,4713,4717,4721,4792,4795,4798,4858,4860,4866,4872,4878,4884,4890,4897,4903],{"lang":12},[14,4258,4260],{"id":4259},"gpt-55は仕事で使えるaiだがmythosの背中は遠い","GPT-5.5は「仕事で使えるAI」——だがMythosの背中は遠い",[19,4262,4263,4264,4267],{},"私はCVCで投資メモ・決算分析・英文ピッチ仕分けを毎日AIに走らせ、Claude Codeでブログ実装を回している人間です。3社のフラッグシップを並列で",[23,4265,4266],{},"自腹のサブスクと業務アカウントの両方","で日常的に使っているからこそ、ベンチマークと実務感覚のずれが見えてきます。今回のGPT-5.5・Mythos・Gemini 3.1 Proの三つ巴は、過去2年で最も「数字より実務感覚が結論を変えた」局面でした。投資家としての含意までを、その手応えで書きます。",[32,4269,4271],{"id":4270},"gpt-55が来ただが主役はまだclaude","GPT-5.5が来た——だが主役はまだClaude",[19,4273,4274],{},"最初に、現状をフラットに置きます。",[19,4276,4277,4278,4281,4282,4285],{},"2026年4月23日、OpenAIは最新フラッグシップモデル",[23,4279,4280],{},"GPT-5.5","（コードネーム「Spud」）を発表しました。前モデルGPT-5.4からわずか6週間という異例のスピードでのリリース。GPT-4.5以来となるフルスクラッチの基盤モデル再学習を行った初のモデルであり、テキスト・画像・音声・動画を単一システムで処理する",[23,4283,4284],{},"ネイティブ・オムニモーダル設計","が採用されている。",[19,4287,4288],{},"率直に言って、GPT-5.5は「仕事でちゃんと使えるAI」に仕上がっている。事前学習の強化により推論の無駄が削ぎ落とされ、処理速度はGPT-5.4と同等を維持しながら性能が底上げされた。計画立案→ツール選択→出力の自己検証という一連のワークフローを、人間の介入を最小限に抑えて実行できる。AIが「質問に答える相手」から「仕事を一緒に片付けるパートナー」に変わったという実感がある。",[19,4290,4291,4292,4295,4296,4299],{},"しかし、",[23,4293,4294],{},"Mythosのレベルには達していない","。4月7日にAnthropicが発表したClaude Mythos Previewの衝撃が冷めやらぬなか、GPT-5.5はClaude陣営にすぐに追いつけない現実を露呈した。ベンチマークの数字を見れば一目瞭然だし、実際に両方を使い比べてみても、Claudeの「地頭の良さ」は明らかに一段上だ。裏を返せば、",[23,4297,4298],{},"Claudeの一人勝ち状態は当面揺るがない","。",[205,4301],{},[32,4303,4305],{"id":4304},"_3社フラッグシップモデルの全貌","3社フラッグシップモデルの全貌",[19,4307,4308],{},"2026年4月末時点で、OpenAI・Anthropic・Googleの3社がほぼ同時期に最新モデルをリリースし、AI業界は過去最大級の性能競争に突入している。各社のフラッグシップを整理してみたい。",[122,4310,4312],{"id":4311},"openai-gpt-55実務のパートナー路線","OpenAI GPT-5.5——「実務のパートナー」路線",[19,4314,4315],{},"GPT-5.5の設計思想は明確だ。OpenAI自身がこのモデルを「実務とエージェントのための新しい知能クラス」と位置づけている。社内コードネームの「Spud（じゃがいも）」が象徴的で、派手さよりも日常の泥くさい仕事をじっくり片付ける方向に振った印象だ。",[19,4317,4318,4319,4322],{},"コンテキストウィンドウは ",[23,4320,4321],{},"100万トークン","（API）。大規模なコードベースや社内ドキュメント群を丸ごと読み込み、そこに対してエージェントが自律的に作業する設計だ。最大出力は128,000トークン。学習・推論基盤にはNVIDIAのGB200とGB300 NVL72が使われている。",[19,4324,4325,4326,4329,4330,4333],{},"ベンチマークを見ると、GPT-5.5の特性がよく分かる。Terminal-Bench 2.0（CLI上の複合ワークフロー評価）で ",[23,4327,4328],{},"82.7%"," を記録し、Claude Opus 4.7の69.4%を明確に上回った。一方、実践的なソフトウェアエンジニアリング能力を測るSWE-Bench Proでは ",[23,4331,4332],{},"58.6%"," と、Claude Opus 4.7の64.3%には及ばない。つまり、GPT-5.5は「ツール連携を伴う長い作業フロー」に強く、「単独のコーディング精度」ではまだClaudeに分がある。この棲み分けは面白い。",[19,4335,4336,4337,4340],{},"ChatGPTの企業向けには ",[23,4338,4339],{},"Workspace Agents"," が実装された。ユーザーがPCをオフラインにした後もクラウド上で自律的に動作し続け、Google Driveなどの外部アプリと連携する。従来のCustom GPTsを置き換える位置づけだ。部門横断の手作業を削減し、長期的な業務プロセスを自動化する——これはエンタープライズ市場への本格的な攻勢と見るべきだろう。",[122,4342,4344],{"id":4343},"anthropic-claude-mythos-preview怪物の誕生","Anthropic Claude Mythos Preview——「怪物」の誕生",[19,4346,4347,4348,4351],{},"Claude Mythos Previewは間違いなく2026年最大のAIイベントだ。Anthropic史上最強のフロンティアモデルで、コードネームは「Capybara」。だが、このモデルは ",[23,4349,4350],{},"一般公開されていない","。理由はシンプルに「強すぎるから」だ。",[19,4353,4354,4355,4358,4359,4362,4363,4366,4367,4370],{},"スコアを並べると異次元さが伝わる。SWE-bench Verifiedで ",[23,4356,4357],{},"93.9%","（Opus 4.6の80.8%から大幅躍進）、SWE-bench Proで ",[23,4360,4361],{},"77.8%","（GPT-5.4の57.7%を圧倒）、自律的なPC操作能力のOSWorldで ",[23,4364,4365],{},"79.6%","、数学競技のUSAMO 2026で ",[23,4368,4369],{},"97.6%","。すべての主要ベンチマークで既存モデルをぶっちぎった。",[19,4372,4373,4374,4377],{},"非公開の最大の理由はサイバーセキュリティ能力にある。Anthropicのレッドチームによるテストで、MythosはWindows、macOS、Linuxの主要OSや全主要Webブラウザに潜むゼロデイ脆弱性を ",[23,4375,4376],{},"数千件"," 自律的に発見した。OpenBSDに27年間潜んでいた脆弱性さえ特定し、攻撃用のエクスプロイトコードまで自動生成した。米財務長官とFRB議長が金融インフラへのリスクを緊急協議したというのだから、その衝撃の大きさが分かる。",[19,4379,4380,4381,4384],{},"Anthropicはこの能力を防御側に活用するため、Apple・Google・Microsoft・NVIDIA等と連携する ",[23,4382,4383],{},"Project Glasswing"," を立ち上げた。参加組織に最大$100M相当のAI利用クレジットを提供し、重要ソフトウェアの脆弱性を先回りで修正する試みだ。Mythos PreviewのAPI料金は入力$25/百万トークン、出力$125/百万トークンと、GPT-5.5の5倍以上になる。",[19,4386,4387,4388,4391],{},"ここで考えるべきは、なぜAnthropicだけがこの性能水準に到達できたのか、だ。自分なりの仮説は ",[23,4389,4390],{},"「秘伝のタレ」"," ——事前学習データのキュレーション品質、RLHF（人間のフィードバックによる強化学習）の洗練されたノウハウ、あるいは潜在空間での推論と再帰的深さ（Recurrent Depth）という新アーキテクチャの組み合わせだろう。いずれか1つではなく、これらが複合的に作用して「地頭の良さ」を生み出している。この優位性は一朝一夕で模倣できるものではないと見ている。",[19,4393,4394],{},"4月16日にリリースされたClaude Opus 4.7もこの文脈で読むべきだ。Opus 4.6から顕著な改善が見られるが、Mythosほどの幅広い能力は備えていない。ただし、Mythosの危険な出力を防ぐセーフガードを最初にテストするモデルとして設計されており、将来的なMythos級モデルの一般公開への布石と理解している。",[122,4396,4398],{"id":4397},"google-gemini-31-pro静かな実力者のジレンマ","Google Gemini 3.1 Pro——「静かな実力者」のジレンマ",[19,4400,4401,4402,4405],{},"Gemini 3.1 ProはGoogle DeepMindが2月19日にリリースした最新モデルだ。Transformer MoEアーキテクチャを採用し、ARC-AGI-2（新しいロジックパターンを解く能力）で ",[23,4403,4404],{},"77.1%","（前モデルの2倍以上）と、推論能力の劇的な向上を見せた。リリース当時はClaude Opus 4.6を上回り、一時的にトップに立った。",[19,4407,4408,4409,4412],{},"100万トークンのコンテキストウィンドウ、テキスト・画像・動画・音声・PDFの統合理解、SVGアニメーションの自動生成やリアルタイムデータからのダッシュボード作成など、実務機能は正直かなり充実している。API料金は入力$2/百万トークン、出力$12/百万トークンと、3社の中で ",[23,4410,4411],{},"最もコスト効率が高い"," のも見逃せない。",[19,4414,4415,4416,4419],{},"問題は、4月のMythos発表以降に相対的なポジションが大きく後退したことだ。率直に言って、",[23,4417,4418],{},"Googleは焦っている"," と思う。Jules（エージェント型コーディングツール）やAntigravity（エージェント開発プラットフォーム）を急ピッチで展開しているが、Claude CodeやOpenAI Codexに対する差別化が見えにくい。モデル性能そのもので頂点を取れないフラストレーションが透けて見える。",[19,4421,4422],{},"ただし、GoogleにはAndroidエコシステム、Google Workspace統合、そして最大のコスト効率というディストリビューション優位がある。性能一点張りではない「別の勝ち筋」が存在することは認識しておくべきだ。",[54,4424,4425,4442],{},[57,4426,4427],{},[60,4428,4429,4431,4433,4436,4439],{},[63,4430,225],{},[63,4432,4280],{},[63,4434,4435],{},"Claude Mythos Preview",[63,4437,4438],{},"Claude Opus 4.7",[63,4440,4441],{},"Gemini 3.1 Pro",[70,4443,4444,4460,4476,4492,4507,4522,4536,4552],{},[60,4445,4446,4449,4451,4455,4457],{},[75,4447,4448],{},"SWE-bench Verified",[75,4450,250],{},[75,4452,4453],{},[23,4454,4357],{},[75,4456,250],{},[75,4458,4459],{},"80.6%",[60,4461,4462,4465,4467,4471,4474],{},[75,4463,4464],{},"SWE-bench Pro",[75,4466,4332],{},[75,4468,4469],{},[23,4470,4361],{},[75,4472,4473],{},"64.3%",[75,4475,250],{},[60,4477,4478,4481,4485,4487,4490],{},[75,4479,4480],{},"Terminal-Bench 2.0",[75,4482,4483],{},[23,4484,4328],{},[75,4486,250],{},[75,4488,4489],{},"69.4%",[75,4491,250],{},[60,4493,4494,4497,4499,4501,4503],{},[75,4495,4496],{},"ARC-AGI-2",[75,4498,250],{},[75,4500,250],{},[75,4502,250],{},[75,4504,4505],{},[23,4506,4404],{},[60,4508,4509,4512,4514,4518,4520],{},[75,4510,4511],{},"USAMO 2026",[75,4513,250],{},[75,4515,4516],{},[23,4517,4369],{},[75,4519,250],{},[75,4521,250],{},[60,4523,4524,4527,4530,4532,4534],{},[75,4525,4526],{},"コンテキスト",[75,4528,4529],{},"1Mトークン",[75,4531,250],{},[75,4533,4529],{},[75,4535,4529],{},[60,4537,4538,4541,4544,4547,4549],{},[75,4539,4540],{},"API入力料金/Mトークン",[75,4542,4543],{},"$5.00",[75,4545,4546],{},"$25.00",[75,4548,250],{},[75,4550,4551],{},"$2.00",[60,4553,4554,4557,4560,4563,4565],{},[75,4555,4556],{},"API出力料金/Mトークン",[75,4558,4559],{},"$30.00",[75,4561,4562],{},"$125.00",[75,4564,250],{},[75,4566,4567],{},"$12.00",[205,4569],{},[32,4571,4573],{"id":4572},"コスト高騰が生むai格差という構造問題","コスト高騰が生む「AI格差」という構造問題",[19,4575,4576,4577,4580],{},"今回の3社比較で最も見落とされがちだが、投資家として最も気になるのは ",[23,4578,4579],{},"推論コストの急激な高騰"," だ。",[19,4582,4583,4584,4587],{},"GPT-5.5のAPI料金は入力$5/百万トークン、出力$30/百万トークンで、GPT-5.4（入力$2.50、出力$15）から ",[23,4585,4586],{},"ちょうど2倍"," に上昇した。上位版GPT-5.5 Proは入力$30、出力$180とさらに高額だ。Claude Mythos Previewに至っては入力$25、出力$125と、GPT-5.5の5倍、Gemini 3.1 Proの10倍以上のコストがかかる。",[19,4589,4590,4591,4594],{},"このコスト構造は、将来的に深刻な ",[23,4592,4593],{},"「AI格差」"," を生み出すリスクを孕んでいる。最先端のAIモデルを使いこなせる企業・個人と、そうでない層の間で、生産性・創造性・情報アクセスの格差が拡大する構図だ。無料ユーザーはGPT-5.5の恩恵を受けられず、前世代を使い続けることになる。少なくともPlus（月額$20）以上が前提になる世界は、AIの民主化という理念とは逆方向に進んでいる。",[54,4596,4597,4613],{},[57,4598,4599],{},[60,4600,4601,4604,4607,4610],{},[63,4602,4603],{},"モデル",[63,4605,4606],{},"入力/Mトークン",[63,4608,4609],{},"出力/Mトークン",[63,4611,4612],{},"GPT-5.4比（出力）",[70,4614,4615,4626,4640,4651,4664],{},[60,4616,4617,4619,4621,4623],{},[75,4618,4441],{},[75,4620,4551],{},[75,4622,4567],{},[75,4624,4625],{},"0.8倍",[60,4627,4628,4631,4634,4637],{},[75,4629,4630],{},"GPT-5.4",[75,4632,4633],{},"$2.50",[75,4635,4636],{},"$15.00",[75,4638,4639],{},"1.0倍（基準）",[60,4641,4642,4644,4646,4648],{},[75,4643,4280],{},[75,4645,4543],{},[75,4647,4559],{},[75,4649,4650],{},"2.0倍",[60,4652,4653,4656,4658,4661],{},[75,4654,4655],{},"GPT-5.5 Pro",[75,4657,4559],{},[75,4659,4660],{},"$180.00",[75,4662,4663],{},"12.0倍",[60,4665,4666,4668,4670,4672],{},[75,4667,4435],{},[75,4669,4546],{},[75,4671,4562],{},[75,4673,4674],{},"8.3倍",[19,4676,4677],{},"特にエージェント系の処理はツールの試行錯誤で出力トークンが膨らみやすい。「気づくと月次費用が想定を大幅に超えていた」というケースは、今後かなり頻発するだろう。モデルの巨大化と性能向上は、必然的にインフラコストの上昇を伴う。これはNVIDIA株の追い風であると同時に、AIスタートアップの損益分岐点を押し上げる逆風でもある。",[19,4679,4680],{},[4681,4682],"img",{"alt":4683,"src":4684},"3社のフラッグシップモデルを擬人化したお遊びイラスト","/blogs-img/blog-gpt55-mythos-mid.png",[205,4686],{},[32,4688,4690],{"id":4689},"コーディングai競争次の覇権を握るのは誰か","コーディングAI競争——次の覇権を握るのは誰か",[19,4692,4693,4694,4697],{},"ここ数ヶ月で最も激化しているのが ",[23,4695,4696],{},"コーディングAI競争"," だ。OpenAIのCodex（アクティブユーザー400万人）、AnthropicのClaude Code、GoogleのJules/Gemini Code Assist/Gemini CLI——3社すべてがAIエージェントによるソフトウェア開発の自動化を最重要テーマに据えている。",[19,4699,4700],{},"なぜコーディングなのか。理由は明確で、コーディング支援はAIの「目に見える生産性向上」を最も直接的に示せる分野だからだ。GitHubのIssueを読んで、該当コードを探し、修正して、テストを流して、プルリクを作る——この一連の作業を自律的に実行できるAIは、エンジニアの生産性を何倍にも引き上げる。そして、その価値に対してエンタープライズ顧客は高い対価を払う用意がある。",[19,4702,4703],{},"GPT-5.5のCodexではブラウザ操作が拡張され、Webアプリとのインタラクション、テストフロー実行、スクリーンショットのキャプチャと反復改善が可能になった。ただし、コーディング精度のベンチマーク（SWE-bench系）では依然としてClaude陣営が優勢だ。Claude Mythos PreviewのSWE-bench Verified 93.9%は、自律的なコード修正能力が人間のシニアエンジニアに接近していることを意味する。この数字は衝撃的だ。",[19,4705,4706,4707,4710],{},"もう一つ注視すべきは、",[23,4708,4709],{},"「AIがAIを育てる」合成データ"," のトレンドだ。AIモデル自身が生成した学習データを使って次世代モデルを訓練する手法は、データの質と多様性を確保できる一方で、モデルの偏りが世代を超えて増幅されるリスクがある。この領域での技術差が、今後のモデル性能の分岐点になるのではないか。Anthropicがこの点で何らかのブレークスルーを持っている可能性は、Mythosの性能を見る限り、かなり濃厚だと考えている。",[205,4712],{},[32,4714,4716],{"id":4715},"投資テーマとしてのai三国志","投資テーマとしてのAI三国志",[122,4718,4720],{"id":4719},"_3社の戦略ポジション","3社の戦略ポジション",[54,4722,4723,4741],{},[57,4724,4725],{},[60,4726,4727,4729,4732,4735,4738],{},[63,4728,434],{},[63,4730,4731],{},"時価総額/バリュエーション",[63,4733,4734],{},"主力モデル",[63,4736,4737],{},"差別化の軸",[63,4739,4740],{},"収益モデル",[70,4742,4743,4759,4776],{},[60,4744,4745,4748,4751,4753,4756],{},[75,4746,4747],{},"OpenAI",[75,4749,4750],{},"~$300B（推定）",[75,4752,4280],{},[75,4754,4755],{},"実務エージェント+Codex",[75,4757,4758],{},"サブスク$20〜$200/月+API",[60,4760,4761,4764,4767,4770,4773],{},[75,4762,4763],{},"Anthropic",[75,4765,4766],{},"~$60B（推定）",[75,4768,4769],{},"Claude Mythos/Opus 4.7",[75,4771,4772],{},"技術的優位+安全性",[75,4774,4775],{},"API+Claude Code+企業向け",[60,4777,4778,4781,4784,4786,4789],{},[75,4779,4780],{},"Google",[75,4782,4783],{},"$2T+（GOOGL）",[75,4785,4441],{},[75,4787,4788],{},"エコシステム統合+コスト効率",[75,4790,4791],{},"広告+Cloud+サブスク",[19,4793,4794],{},"三者三様の戦い方が見えてきた。OpenAIは「実務で使える」を旗印にエンタープライズ浸透を狙い、Anthropicは技術的な突出で「頂点」を維持し、Googleはエコシステムとコスト効率で広く薄く取る。AIの「Windows vs Mac vs Linux」とでも言うべき構図が固まりつつある。",[122,4796,4797],{"id":4797},"シナリオ分析",[54,4799,4800,4812],{},[57,4801,4802],{},[60,4803,4804,4806,4809],{},[63,4805,601],{},[63,4807,4808],{},"想定される展開",[63,4810,4811],{},"投資含意",[70,4813,4814,4825,4836,4847],{},[60,4815,4816,4819,4822],{},[75,4817,4818],{},"ベースケース",[75,4820,4821],{},"3社が性能面で拮抗し、差別化はエコシステムとユースケースで決まる",[75,4823,4824],{},"Googleのディストリビューション優位が活きる。GOOGL安定",[60,4826,4827,4830,4833],{},[75,4828,4829],{},"メインシナリオ",[75,4831,4832],{},"Claudeの技術的優位が持続。Mythos級モデルの一般公開は2027年以降",[75,4834,4835],{},"Anthropic IPOが2027年の大型テーマに。プレIPO投資機会",[60,4837,4838,4841,4844],{},[75,4839,4840],{},"テールリスク（上振れ）",[75,4842,4843],{},"AI性能が急速に天井に到達し、差別化がアプリケーション層に移行",[75,4845,4846],{},"インフラ企業（NVIDIA等）よりアプリケーション企業が恩恵",[60,4848,4849,4852,4855],{},[75,4850,4851],{},"テールリスク（下振れ）",[75,4853,4854],{},"コスト高騰でAI投資ROIが疑問視される。「AIバブル」崩壊",[75,4856,4857],{},"テック株全般に調整圧力。ただしAI必需品化で底は浅い",[122,4859,764],{"id":764},[19,4861,4862,4865],{},[23,4863,4864],{},"Claude Mythosの一般公開時期"," — セーフガードの構築が進めば段階的な公開が見込まれるが、時期は不透明。公開されれば、AI業界の勢力図が一変する可能性がある。",[19,4867,4868,4871],{},[23,4869,4870],{},"OpenAI IPOの動向"," — ChatGPTの有料ユーザー900万人、Codexの400万人という数字がバリュエーションを支えるが、コスト構造の持続可能性が問われる。GPT-5.5のAPI料金2倍という値上げが、成長率にどう影響するかを注視すべきだ。",[19,4873,4874,4877],{},[23,4875,4876],{},"推論コスト低減技術の進展"," — モデルの蒸留、量子化、推論専用チップの開発が市場全体のTAMを決定する。NVIDIAの次世代チップロードマップ、GoogleのTPU、AmazonのTrainiumの動向が鍵になる。",[19,4879,4880,4883],{},[23,4881,4882],{},"エージェント型ワークフローのB2B浸透"," — GPT-5.5のWorkspace Agents、Claude Code、Gemini Agentsがどれだけエンタープライズ市場に食い込むかが、各社の収益成長を左右する。2026年後半の採用データが最初の試金石だ。",[19,4885,4886,4889],{},[23,4887,4888],{},"AI規制の加速"," — Mythosの「強すぎて出せない」という前例は、AI規制の議論に新しい次元を加えた。EU AI規制法との整合性、各国政府の対応が業界全体のバリュエーションに影響する。",[19,4891,4892,4893,4896],{},"結論として、AI競争は ",[23,4894,4895],{},"「性能の頂点」から「実務での浸透」"," へフェーズ移行しつつある。GPT-5.5は「仕事でちゃんと使えるAI」として着実に進化し、Mythosという「怪物」はAnthropicの技術的突出を証明した。Googleはコスト効率とエコシステムで別の勝ち筋を模索している。3社の戦略差がこれほど鮮明になった瞬間は過去になく、投資家にとってはポジションを見極める好機だろう。",[19,4898,4899],{},[4681,4900],{"alt":4901,"src":4902},"GPT-5.5・Claude Mythos・Gemini 3.1 Proの戦略ポジションサマリ","/blogs-img/blog-gpt55-mythos-summary.png",[19,4904,4905],{},[802,4906,4907],{},"本記事は情報提供を目的としたものであり、特定の金融商品の売買を推奨するものではありません。投資判断はご自身の責任において行ってください。",[205,4909],{},[10,4911,4912,4916,4927,4931,4934,4944,4954,4964,4966,4970,4973,4977,4984,4991,5008,5014,5018,5025,5040,5047,5053,5064,5067,5071,5077,5083,5090,5093,5220,5222,5226,5232,5239,5246,5321,5328,5333,5335,5339,5346,5353,5360,5370,5372,5376,5380,5447,5450,5454,5514,5518,5524,5530,5536,5542,5548,5555,5560],{"lang":809},[14,4913,4915],{"id":4914},"gpt-55-is-finally-ai-that-works-at-work-but-mythos-is-still-far-ahead","GPT-5.5 Is Finally \"AI That Works at Work\" — But Mythos Is Still Far Ahead",[19,4917,4918,4919,4922,4923,4926],{},"I run investment memos, earnings analyses, and English pitch-email triage through AI every day from the CVC seat, and I maintain this blog through Claude Code. I pay for all three flagships out of pocket ",[802,4920,4921],{},"and"," have enterprise accounts — which means I see where benchmarks diverge from real working feel. The three-way clash between GPT-5.5, Mythos, and Gemini 3.1 Pro is the ",[23,4924,4925],{},"most \"live working feel changed my conclusion\""," moment of the last two years. The investor implications follow that hands-on read.",[32,4928,4930],{"id":4929},"gpt-55-has-arrived-but-claude-still-owns-the-stage","GPT-5.5 Has Arrived — But Claude Still Owns the Stage",[19,4932,4933],{},"Let me put the state of play down flat first.",[19,4935,4936,4937,4939,4940,4943],{},"On April 23, 2026, OpenAI announced its newest flagship, ",[23,4938,4280],{}," (codename \"Spud\"), shipped just six weeks after GPT-5.4 — an unusually short cadence. It is the first OpenAI model since GPT-4.5 trained from scratch as a foundation model, and the first to use a ",[23,4941,4942],{},"native omnimodal architecture"," that handles text, images, audio, and video in a single system.",[19,4945,4946,4947,4950,4951],{},"To put it plainly: GPT-5.5 is finally ",[23,4948,4949],{},"\"AI that actually works at work.\""," Stronger pre-training has stripped reasoning waste, latency stays roughly on par with GPT-5.4, and end-to-end performance moves up. It can run plan → tool selection → self-verification cycles with minimal human intervention. AI has shifted, in my view, from \"something I ask questions to\" into ",[23,4952,4953],{},"\"a partner that gets work done with me.\"",[19,4955,4956,4957,4960,4961,849],{},"But — ",[23,4958,4959],{},"it has not reached the level of Mythos",". Anthropic's Claude Mythos Preview was announced on April 7, and the shockwave has not faded. GPT-5.5 makes it brutally clear that catching up to Claude is not happening overnight. The benchmarks tell the story, and side-by-side use confirms it: Claude's raw \"smarts\" are visibly a tier above. The flip side is that ",[23,4962,4963],{},"Claude's solo lead is unlikely to be challenged for some time",[205,4965],{},[32,4967,4969],{"id":4968},"the-three-flagship-models-in-detail","The Three Flagship Models in Detail",[19,4971,4972],{},"As of late April 2026, OpenAI, Anthropic, and Google have all refreshed their flagships in near lockstep, and the AI industry is in the most intense capability race we've seen.",[122,4974,4976],{"id":4975},"openai-gpt-55-the-working-partner-bet","OpenAI GPT-5.5 — The \"Working Partner\" Bet",[19,4978,4979,4980,4983],{},"GPT-5.5's design intent is unambiguous. OpenAI itself frames it as ",[23,4981,4982],{},"\"a new intelligence class for real work and agents.\""," The internal codename \"Spud\" (potato) feels apt: not flashy, but built to grind through mundane day-to-day work.",[19,4985,4986,4987,4990],{},"Context window: ",[23,4988,4989],{},"1M tokens"," (API). The design assumption is that an agent loads an entire codebase or internal documentation set and works autonomously over it. Max output is 128K tokens. Training and inference run on NVIDIA GB200 and GB300 NVL72.",[19,4992,4993,4994,4996,4997,4999,5000,5003,5004,5007],{},"Benchmarks define the personality. Terminal-Bench 2.0 (CLI multi-step workflow evaluation) hits ",[23,4995,4328],{},", clearly ahead of Claude Opus 4.7 at 69.4%. But on SWE-Bench Pro — the harder real-world software-engineering benchmark — GPT-5.5 lands at ",[23,4998,4332],{},", behind Claude Opus 4.7's 64.3%. Translation: GPT-5.5 is strong on ",[23,5001,5002],{},"long, tool-using workflows",", while Claude still leads on ",[23,5005,5006],{},"pure coding precision",". That split is interesting.",[19,5009,5010,5011,5013],{},"For enterprise ChatGPT, OpenAI shipped ",[23,5012,4339],{},": cloud-resident agents that keep running after the user closes their laptop, integrated with Google Drive and other external apps. They effectively replace Custom GPTs. The pitch — eliminate cross-team manual work, automate long-running business processes — reads as a serious enterprise push.",[122,5015,5017],{"id":5016},"anthropic-claude-mythos-preview-the-birth-of-a-monster","Anthropic Claude Mythos Preview — The Birth of a \"Monster\"",[19,5019,5020,5021,5024],{},"Claude Mythos Preview is, without question, the biggest AI event of 2026. It is Anthropic's strongest frontier model ever, codename \"Capybara.\" But it is ",[23,5022,5023],{},"not generally available",". The reason is simple: it is too strong.",[19,5026,5027,5028,5030,5031,5033,5034,5036,5037,5039],{},"The numbers are out of distribution. SWE-bench Verified at ",[23,5029,4357],{}," (Opus 4.6 was 80.8%). SWE-bench Pro at ",[23,5032,4361],{}," (GPT-5.4: 57.7%). OSWorld autonomous PC control at ",[23,5035,4365],{},". USAMO 2026 math at ",[23,5038,4369],{},". It dominates every major benchmark.",[19,5041,5042,5043,5046],{},"The deepest reason for the limited release is cybersecurity capability. In Anthropic red-team tests, Mythos autonomously discovered ",[23,5044,5045],{},"thousands"," of zero-day vulnerabilities across Windows, macOS, Linux, and every major web browser. It even surfaced a 27-year-old OpenBSD vulnerability and auto-generated working exploit code. The U.S. Treasury Secretary and the Fed Chair reportedly held an emergency consultation about systemic risk to financial infrastructure — that scale of impact is the signal.",[19,5048,5049,5050,5052],{},"To redirect the capability to defense, Anthropic launched ",[23,5051,4383],{},", partnering with Apple, Google, Microsoft, NVIDIA, and others. Participating organizations get up to $100M of AI usage credits to proactively patch critical software. Mythos Preview API pricing is $25 per million input tokens and $125 per million output tokens — over 5× GPT-5.5.",[19,5054,5055,5056,5059,5060,5063],{},"The real question is ",[23,5057,5058],{},"why only Anthropic reached this level",". My working hypothesis is the ",[23,5061,5062],{},"\"secret sauce\""," — pre-training data curation quality, refined RLHF craftsmanship, and possibly an architectural combination involving latent-space reasoning and \"Recurrent Depth.\" Not any one of these alone, but the compound effect, is what produces the raw intelligence. I don't think the gap closes in months.",[19,5065,5066],{},"Claude Opus 4.7, released April 16, should be read in this same context. It improves notably on Opus 4.6 but lacks Mythos-class breadth. Its real role is to be the first vehicle for testing Mythos-grade safety guardrails — a stepping stone for eventually shipping Mythos-class capability to the public.",[122,5068,5070],{"id":5069},"google-gemini-31-pro-the-quiet-strong-dilemma","Google Gemini 3.1 Pro — The \"Quiet Strong\" Dilemma",[19,5072,5073,5074,5076],{},"Gemini 3.1 Pro is Google DeepMind's latest, released February 19. It uses a Transformer MoE architecture and posted ",[23,5075,4404],{}," on ARC-AGI-2 (more than 2× the prior generation), a step-change in abstract reasoning. At launch it actually overtook Claude Opus 4.6 and briefly held the top spot.",[19,5078,5079,5080,849],{},"It also delivers strong product surface: 1M-token context, unified text/image/video/audio/PDF understanding, automated SVG animation, real-time data → dashboard generation. API pricing of $2 per million input tokens and $12 per million output tokens makes it ",[23,5081,5082],{},"the most cost-efficient flagship of the three",[19,5084,5085,5086,5089],{},"The problem is positioning post-Mythos. Frankly, ",[23,5087,5088],{},"Google looks rattled",". Jules (agentic coding tool) and Antigravity (agent development platform) are shipping fast, but their differentiation against Claude Code and OpenAI Codex is hard to read. The frustration of not topping pure model performance is visible.",[19,5091,5092],{},"That said, Google still has the Android ecosystem, Workspace integration, and best-in-class cost efficiency — a distribution moat. There is a real \"different way to win\" beyond raw capability that should not be underestimated.",[54,5094,5095,5109],{},[57,5096,5097],{},[60,5098,5099,5101,5103,5105,5107],{},[63,5100,1015],{},[63,5102,4280],{},[63,5104,4435],{},[63,5106,4438],{},[63,5108,4441],{},[70,5110,5111,5125,5139,5153,5167,5181,5194,5207],{},[60,5112,5113,5115,5117,5121,5123],{},[75,5114,4448],{},[75,5116,250],{},[75,5118,5119],{},[23,5120,4357],{},[75,5122,250],{},[75,5124,4459],{},[60,5126,5127,5129,5131,5135,5137],{},[75,5128,4464],{},[75,5130,4332],{},[75,5132,5133],{},[23,5134,4361],{},[75,5136,4473],{},[75,5138,250],{},[60,5140,5141,5143,5147,5149,5151],{},[75,5142,4480],{},[75,5144,5145],{},[23,5146,4328],{},[75,5148,250],{},[75,5150,4489],{},[75,5152,250],{},[60,5154,5155,5157,5159,5161,5163],{},[75,5156,4496],{},[75,5158,250],{},[75,5160,250],{},[75,5162,250],{},[75,5164,5165],{},[23,5166,4404],{},[60,5168,5169,5171,5173,5177,5179],{},[75,5170,4511],{},[75,5172,250],{},[75,5174,5175],{},[23,5176,4369],{},[75,5178,250],{},[75,5180,250],{},[60,5182,5183,5186,5188,5190,5192],{},[75,5184,5185],{},"Context",[75,5187,4989],{},[75,5189,250],{},[75,5191,4989],{},[75,5193,4989],{},[60,5195,5196,5199,5201,5203,5205],{},[75,5197,5198],{},"API input $/1M tokens",[75,5200,4543],{},[75,5202,4546],{},[75,5204,250],{},[75,5206,4551],{},[60,5208,5209,5212,5214,5216,5218],{},[75,5210,5211],{},"API output $/1M tokens",[75,5213,4559],{},[75,5215,4562],{},[75,5217,250],{},[75,5219,4567],{},[205,5221],{},[32,5223,5225],{"id":5224},"surging-costs-and-the-emerging-ai-divide","Surging Costs and the Emerging \"AI Divide\"",[19,5227,5228,5229,849],{},"The most overlooked storyline in the three-way comparison — and the one I care about most as an investor — is ",[23,5230,5231],{},"the sharp rise in inference cost",[19,5233,5234,5235,5238],{},"GPT-5.5 API pricing is $5 per million input tokens and $30 per million output tokens — ",[23,5236,5237],{},"exactly 2×"," GPT-5.4 ($2.50/$15). The premium GPT-5.5 Pro is $30/$180. Claude Mythos Preview at $25/$125 is over 5× GPT-5.5 and more than 10× Gemini 3.1 Pro.",[19,5240,5241,5242,5245],{},"This pricing structure carries real risk of generating an ",[23,5243,5244],{},"\"AI divide\""," — a widening gap in productivity, creativity, and information access between those who can wield frontier models and those who cannot. Free users won't get GPT-5.5 and will stay on the prior generation. A world where Plus ($20/month) is the de facto floor moves in the opposite direction of \"democratizing AI.\"",[54,5247,5248,5264],{},[57,5249,5250],{},[60,5251,5252,5255,5258,5261],{},[63,5253,5254],{},"Model",[63,5256,5257],{},"Input $/1M",[63,5259,5260],{},"Output $/1M",[63,5262,5263],{},"vs GPT-5.4 (output)",[70,5265,5266,5277,5288,5299,5310],{},[60,5267,5268,5270,5272,5274],{},[75,5269,4441],{},[75,5271,4551],{},[75,5273,4567],{},[75,5275,5276],{},"0.8×",[60,5278,5279,5281,5283,5285],{},[75,5280,4630],{},[75,5282,4633],{},[75,5284,4636],{},[75,5286,5287],{},"1.0× (baseline)",[60,5289,5290,5292,5294,5296],{},[75,5291,4280],{},[75,5293,4543],{},[75,5295,4559],{},[75,5297,5298],{},"2.0×",[60,5300,5301,5303,5305,5307],{},[75,5302,4655],{},[75,5304,4559],{},[75,5306,4660],{},[75,5308,5309],{},"12.0×",[60,5311,5312,5314,5316,5318],{},[75,5313,4435],{},[75,5315,4546],{},[75,5317,4562],{},[75,5319,5320],{},"8.3×",[19,5322,5323,5324,5327],{},"Agentic workloads in particular tend to balloon output tokens via tool retries and self-correction. ",[23,5325,5326],{},"\"My monthly bill came in way over plan\""," is going to be an extremely common story going forward. Larger, smarter models inevitably mean rising infrastructure cost — a tailwind for NVIDIA, but a headwind on the breakeven economics of every AI startup.",[19,5329,5330],{},[4681,5331],{"alt":5332,"src":4684},"A playful illustration personifying the three flagship AI models",[205,5334],{},[32,5336,5338],{"id":5337},"the-coding-ai-race-who-takes-the-next-crown","The Coding-AI Race — Who Takes the Next Crown?",[19,5340,5341,5342,5345],{},"The hottest contested battlefield of the past few months is ",[23,5343,5344],{},"coding AI",". OpenAI's Codex (4M active users), Anthropic's Claude Code, Google's Jules / Gemini Code Assist / Gemini CLI — all three labs have made agentic software development their top priority.",[19,5347,5348,5349,5352],{},"Why coding? Because it is the cleanest place to demonstrate ",[23,5350,5351],{},"visible AI productivity gains",". An agent that can read a GitHub issue, find the relevant code, fix it, run tests, and open a PR multiplies engineer output. And enterprise customers are willing to pay real money for that value.",[19,5354,5355,5356,5359],{},"GPT-5.5's Codex extends browser control: web app interaction, full test-flow execution, screenshot capture, and iterative refinement. But on coding accuracy benchmarks (SWE-bench family), Claude still leads. Mythos Preview's SWE-bench Verified at 93.9% means ",[23,5357,5358],{},"autonomous code repair is approaching senior-engineer level",". That number is genuinely jarring.",[19,5361,5362,5363,5366,5367,849],{},"The other thing to watch is ",[23,5364,5365],{},"\"AI training AI\" via synthetic data",". Using model-generated data to train the next generation secures quality and diversity, but risks amplifying biases across generations. Differentiation in this layer is, I suspect, the next inflection point for model performance. Looking at Mythos's level, my prior is high that ",[23,5368,5369],{},"Anthropic has some kind of breakthrough here",[205,5371],{},[32,5373,5375],{"id":5374},"ai-as-an-investment-theme-a-three-kingdoms-view","AI as an Investment Theme — A Three-Kingdoms View",[122,5377,5379],{"id":5378},"strategic-positioning-of-the-three-players","Strategic Positioning of the Three Players",[54,5381,5382,5399],{},[57,5383,5384],{},[60,5385,5386,5388,5390,5393,5396],{},[63,5387,1228],{},[63,5389,900],{},[63,5391,5392],{},"Flagship",[63,5394,5395],{},"Differentiation",[63,5397,5398],{},"Revenue model",[70,5400,5401,5416,5432],{},[60,5402,5403,5405,5408,5410,5413],{},[75,5404,4747],{},[75,5406,5407],{},"~$300B (est.)",[75,5409,4280],{},[75,5411,5412],{},"Practical agents + Codex",[75,5414,5415],{},"Subs $20–$200/mo + API",[60,5417,5418,5420,5423,5426,5429],{},[75,5419,4763],{},[75,5421,5422],{},"~$60B (est.)",[75,5424,5425],{},"Claude Mythos / Opus 4.7",[75,5427,5428],{},"Technical lead + safety",[75,5430,5431],{},"API + Claude Code + enterprise",[60,5433,5434,5436,5439,5441,5444],{},[75,5435,4780],{},[75,5437,5438],{},"$2T+ (GOOGL)",[75,5440,4441],{},[75,5442,5443],{},"Ecosystem + cost efficiency",[75,5445,5446],{},"Ads + Cloud + subs",[19,5448,5449],{},"Three distinct shapes are emerging. OpenAI pushes \"practical at work\" to win enterprise. Anthropic holds the summit through technical dominance. Google goes wide and thin on ecosystem and pricing. The AI version of \"Windows vs Mac vs Linux\" is taking form.",[122,5451,5453],{"id":5452},"scenario-analysis","Scenario Analysis",[54,5455,5456,5468],{},[57,5457,5458],{},[60,5459,5460,5462,5465],{},[63,5461,1389],{},[63,5463,5464],{},"Likely path",[63,5466,5467],{},"Investment implication",[70,5469,5470,5481,5492,5503],{},[60,5471,5472,5475,5478],{},[75,5473,5474],{},"Base case",[75,5476,5477],{},"The three converge on capability; differentiation moves to ecosystem and use cases",[75,5479,5480],{},"Google's distribution moat shines. Stable GOOGL",[60,5482,5483,5486,5489],{},[75,5484,5485],{},"Main case",[75,5487,5488],{},"Claude's technical lead persists; Mythos-class GA ships in 2027+",[75,5490,5491],{},"Anthropic IPO becomes a marquee 2027 theme; pre-IPO opportunity",[60,5493,5494,5497,5500],{},[75,5495,5496],{},"Tail (upside)",[75,5498,5499],{},"AI capability hits a fast ceiling; differentiation shifts to the application layer",[75,5501,5502],{},"Application companies benefit more than infra (NVIDIA et al.)",[60,5504,5505,5508,5511],{},[75,5506,5507],{},"Tail (downside)",[75,5509,5510],{},"Cost spike calls AI ROI into question; \"AI bubble\" concerns surface",[75,5512,5513],{},"Tech sees broad correction pressure, but bottom is shallow given AI's now-essential nature",[122,5515,5517],{"id":5516},"what-to-watch-next","What to Watch Next",[19,5519,5520,5523],{},[23,5521,5522],{},"Mythos GA timing"," — As guardrails mature, a phased release is plausible, but the schedule is unclear. Once it ships, the industry power map can flip overnight.",[19,5525,5526,5529],{},[23,5527,5528],{},"OpenAI IPO trajectory"," — 9M paying ChatGPT users and 4M Codex users underwrite the valuation, but cost-structure sustainability is the open question. Watch how the GPT-5.5 2× price hike affects growth.",[19,5531,5532,5535],{},[23,5533,5534],{},"Inference-cost reduction tech"," — Distillation, quantization, and inference-specific silicon will determine the total addressable market. NVIDIA's next-gen chip roadmap, Google's TPU cadence, and Amazon's Trainium are key signals.",[19,5537,5538,5541],{},[23,5539,5540],{},"Agentic workflow B2B adoption"," — How well GPT-5.5 Workspace Agents, Claude Code, and Gemini Agents penetrate enterprise will dictate revenue growth. H2 2026 adoption data is the first real test.",[19,5543,5544,5547],{},[23,5545,5546],{},"AI regulation acceleration"," — Mythos's \"too strong to ship\" precedent has added a new dimension to the regulatory conversation. EU AI Act alignment and government responses will move sector valuations.",[19,5549,5550,5551,5554],{},"The bottom line: AI competition is shifting ",[23,5552,5553],{},"from \"peak capability\" to \"practical adoption.\""," GPT-5.5 has matured into an \"AI that works at work.\" Mythos has proven Anthropic's technical lead. Google is hunting a different win via cost and ecosystem. The strategic gap between the three has never been this sharp — for investors, it's a real moment to size positioning carefully.",[19,5556,5557],{},[4681,5558],{"alt":5559,"src":4902},"Strategic positioning summary of GPT-5.5, Claude Mythos, and Gemini 3.1 Pro",[19,5561,5562],{},[802,5563,5564],{},"This article is for informational purposes only and is not a recommendation to buy or sell any financial instrument. Please make investment decisions at your own discretion.",{"title":143,"searchDepth":1605,"depth":1605,"links":5566},[5567,5568,5573,5574,5575,5580,5581,5586,5587,5588],{"id":4270,"depth":1605,"text":4271},{"id":4304,"depth":1605,"text":4305,"children":5569},[5570,5571,5572],{"id":4311,"depth":1610,"text":4312},{"id":4343,"depth":1610,"text":4344},{"id":4397,"depth":1610,"text":4398},{"id":4572,"depth":1605,"text":4573},{"id":4689,"depth":1605,"text":4690},{"id":4715,"depth":1605,"text":4716,"children":5576},[5577,5578,5579],{"id":4719,"depth":1610,"text":4720},{"id":4797,"depth":1610,"text":4797},{"id":764,"depth":1610,"text":764},{"id":4929,"depth":1605,"text":4930},{"id":4968,"depth":1605,"text":4969,"children":5582},[5583,5584,5585],{"id":4975,"depth":1610,"text":4976},{"id":5016,"depth":1610,"text":5017},{"id":5069,"depth":1610,"text":5070},{"id":5224,"depth":1605,"text":5225},{"id":5337,"depth":1605,"text":5338},{"id":5374,"depth":1605,"text":5375,"children":5589},[5590,5591,5592],{"id":5378,"depth":1610,"text":5379},{"id":5452,"depth":1610,"text":5453},{"id":5516,"depth":1610,"text":5517},"OpenAIが発表したGPT-5.5の実力を、Claude Mythos Preview・Gemini 3.1 Proと徹底比較。3社の戦略差・コスト高騰・コーディングAI競争の行方を投資家視点で読み解く。",{"date":5595,"image":5596,"alt":5597,"tags":5598,"tagsEn":5600,"published":1666},"25th Apr 2026","/blogs-img/blog-gpt55-mythos.png","GPT-5.5 vs Claude Mythos vs Gemini 3.1 Proの比較",[1657,4240,5599,4241,1660],"Deep Dive",[1662,4245,5599,4241,5601],"Startup","/blogs/10-gpt55-vs-claude-mythos",{"title":4252,"description":5593},"blogs/10-gpt55-vs-claude-mythos","XIC-ZsMdCIBnyhFFlRatPLkSnIADfu5wv_DW9LdWdVo",{"id":5607,"title":5608,"body":5609,"description":7090,"extension":1651,"meta":7091,"navigation":1666,"ogImage":7093,"path":7097,"seo":7098,"stem":7099,"__hash__":7100},"content/blogs/11-big-tech-q1-2026-comparison.md","Big Tech決算を読み解く：Meta・Amazon・Google・Microsoftの実力比較 | Decoding Big Tech Q1 2026 Earnings: A Comparative Analysis of Meta, Amazon, Google, and Microsoft",{"type":7,"value":5610,"toc":7062},[5611,6331,6333],[10,5612,5613,5617,5624,5628,5631,5638,5645,5648,5650,5654,5657,5830,5833,5835,5838,5842,5849,5852,5855,5859,5862,5865,5872,5876,5879,5882,5885,5888,5892,5895,5898,5905,5908,5910,5914,5917,5990,5993,5995,5998,6161,6163,6167,6254,6260,6266,6268,6270,6275,6280,6285,6290,6292,6295,6301,6307,6313,6319,6326],{"lang":12},[14,5614,5616],{"id":5615},"big-tech決算を読み解くmetaamazongooglemicrosoftの実力比較","Big Tech決算を読み解く：Meta・Amazon・Google・Microsoftの実力比較",[19,5618,5619,5620,5623],{},"私はCVCで毎四半期Big Techの決算を読み解いて、自社のクラウド・AIツール選定や投資判断に落とし込む役割を持っている人間です。AIキャペックスがいよいよ$6,000億ドルを超えてくる局面で、",[23,5621,5622],{},"「広告系」「クラウド系」「エコシステム系」の3つに分けて読まないと意味がない","——というのがここ2年で得た現場感覚です。今回のQ1 2026決算は、その3つのROI構造の差が初めて数字として可視化された四半期でした。",[32,5625,5627],{"id":5626},"一斉出そろったai黒字化元年の決算","一斉出そろった「AI黒字化元年」の決算",[19,5629,5630],{},"最初に、四半期の構図をフラットに置きます。",[19,5632,5633,5634,5637],{},"2026年4月末、米国Big Tech各社が相次いで2026年第1四半期の決算を発表しました。AIへの超大規模な資本支出が「本当に収益に変換されているのか」という問いは、2024年から2025年にかけて市場を悩ませてきました。Meta一社だけで2026年の設備投資計画を1,350億ドルに引き上げ、Amazon・Microsoft・Googleを合わせた4社の合算設備投資は2026年に",[23,5635,5636],{},"6,000億ドル","を超える見込み。文字通り史上前例のない規模です。",[19,5639,5640,5641,5644],{},"その答えが今四半期の決算で、ある程度明確な形で出た。GoogleはCloud部門が年率換算で約1,500億ドルのGemini API売上を積み上げ、MicrosoftのAIビジネスは年率370億ドルに達し前年比123%増を記録。MetaはAIが広告のコンバージョン率や時間消費を押し上げることで、まさに今この瞬間に収益をマネタイズしている。Amazonも同様にAWSのAIワークロード需要が牽引する形で成長が加速した。「AIは本当に稼げるのか」という問いに対する答えは、今四半期の決算を通じて ",[23,5642,5643],{},"「Yes」"," に傾きつつある。",[19,5646,5647],{},"ただし、4社の「AI稼ぎ方」は構造的に異なる。バリュエーション、モメンタム、リスクプロファイルは4社それぞれで大きく違う。本稿では、Morningstarのアナリストレポートおよびリアルタイムデータをもとに、4社を多角的に比較し、どこに投資機会があるかを検討する。",[205,5649],{},[32,5651,5653],{"id":5652},"_4社の決算主要kpi比較","4社の決算：主要KPI比較",[19,5655,5656],{},"以下に2026年Q1決算の主要数値を整理する。",[54,5658,5659,5677],{},[57,5660,5661],{},[60,5662,5663,5665,5668,5671,5674],{},[63,5664,225],{},[63,5666,5667],{},"Meta",[63,5669,5670],{},"Amazon",[63,5672,5673],{},"Alphabet (Google)",[63,5675,5676],{},"Microsoft",[70,5678,5679,5696,5713,5729,5746,5763,5779,5796,5813],{},[60,5680,5681,5684,5687,5690,5693],{},[75,5682,5683],{},"売上（前年比）",[75,5685,5686],{},"$560億（+33%）",[75,5688,5689],{},"$1,815億（+15%）",[75,5691,5692],{},"$1,100億（+22%）",[75,5694,5695],{},"$829億（+18%）",[60,5697,5698,5701,5704,5707,5710],{},[75,5699,5700],{},"営業利益率",[75,5702,5703],{},"41%（-90bp）",[75,5705,5706],{},"13.1%（+130bp）",[75,5708,5709],{},"34%（+220bp）",[75,5711,5712],{},"46.3%",[60,5714,5715,5718,5721,5724,5726],{},[75,5716,5717],{},"EPS",[75,5719,5720],{},"未公開",[75,5722,5723],{},"$8.47(E/F)",[75,5725,5720],{},[75,5727,5728],{},"$4.27（予想$4.06超）",[60,5730,5731,5734,5737,5740,5743],{},[75,5732,5733],{},"クラウド/AI KPI",[75,5735,5736],{},"広告単価+12%、量+19%",[75,5738,5739],{},"AWS+28%、年率$1,500億",[75,5741,5742],{},"Cloud+63%、$200億",[75,5744,5745],{},"Azure+40%、AI年率$370億",[60,5747,5748,5751,5754,5757,5760],{},[75,5749,5750],{},"設備投資（2026年計画）",[75,5752,5753],{},"$1,350億",[75,5755,5756],{},"$2,000億",[75,5758,5759],{},"大幅増（未公開）",[75,5761,5762],{},"$1,900億",[60,5764,5765,5768,5771,5774,5776],{},[75,5766,5767],{},"Morningstar評価",[75,5769,5770],{},"★★★★ 割安",[75,5772,5773],{},"★★★ ほぼフェアバリュー",[75,5775,5770],{},[75,5777,5778],{},"★★★★★ 大幅割安",[60,5780,5781,5784,5787,5790,5793],{},[75,5782,5783],{},"フェアバリュー推定値",[75,5785,5786],{},"$850",[75,5788,5789],{},"$280（$260→引上げ）",[75,5791,5792],{},"$433（$340→引上げ）",[75,5794,5795],{},"$600",[60,5797,5798,5801,5804,5807,5810],{},[75,5799,5800],{},"現在株価",[75,5802,5803],{},"$669",[75,5805,5806],{},"$263",[75,5808,5809],{},"$350",[75,5811,5812],{},"$424",[60,5814,5815,5818,5821,5824,5827],{},[75,5816,5817],{},"フェアバリュー乖離",[75,5819,5820],{},"-21%",[75,5822,5823],{},"-6%",[75,5825,5826],{},"-19%",[75,5828,5829],{},"-29%",[19,5831,5832],{},"この表だけ見ても、バリュエーション上最も魅力的なのはMicrosoftであり、最もリスク・リターンが均衡しているのはAmazonだということが直感的にわかる。",[205,5834],{},[32,5836,5837],{"id":5837},"各社の詳細分析",[122,5839,5841],{"id":5840},"meta-広告aiの完成形問われる設備投資の行方","Meta — 広告AIの完成形、問われる設備投資の行方",[19,5843,5844,5845,5848],{},"Metaの今四半期を一言で表すなら ",[23,5846,5847],{},"「AI広告エンジンの完成形が姿を現した」"," だ。Q1売上は前年比33%増の560億ドル。Instagram Reels上でAIが推薦するコンテンツへの視聴時間が10%増加し、Facebook動画が8%増、Instagramのコンバージョン率が1.6%改善した。これらの数字は、MetaがAI投資から直接的に広告収益を引き出している証拠である。",[19,5850,5851],{},"AI最適化された広告アルゴリズムが、広告単価（前年比+12%）と広告量（+19%）の両方を押し上げるという理想的な構造になっている。Morningstarが強調するのは、38億人以上のデイリーアクティブユーザーというMetaのリーチが「スケールの経済」をAI投資に乗せて機能させているという点だ。データ収集の規模が大きければ大きいほどAIモデルの精度が上がり、広告効果が高まり、広告単価が上昇するという好循環が実現している。",[19,5853,5854],{},"一方で、設備投資ガイダンスの引き上げ（1,250億ドル→1,350億ドル）が株価を時間外で押し下げた。Reality Labs（VR/AR部門）が依然として年間160億ドル超の営業赤字を垂れ流し続けていることも、投資家のコスト懸念を増幅させている。Morningstarは$850のフェアバリューを維持しており、現在株価$669は依然として21%の割安水準にある。株価下落を「AI設備投資への過剰懸念」と見るか、「妥当な警戒」と見るかが判断の分かれ目だ。",[122,5856,5858],{"id":5857},"amazon-awsが本格加速e-commerceの底堅さも証明","Amazon — AWSが本格加速、E-commerceの底堅さも証明",[19,5860,5861],{},"Amazonの今四半期最大のサプライズはAWSの成長加速だ。AWS売上の前年比成長率は28%に達し、年率換算1,500億ドルの収益基盤を持つまでになった。AIとトラディショナルなクラウドワークロードの両方が、それぞれ強い需要を示している。全セグメントがMorningstarの予想を超えており、Morningstarは従来の公正価値見積もりを260ドルから280ドルへと引き上げた。",[19,5863,5864],{},"AWSが全社収益の17%しか占めない一方で、全社の営業利益の大部分を稼ぐという構造は変わらない。広告事業も引き続き高成長・高マージンの収益源として機能しており、Eコマースのオペレーション効率改善（ロボティクス、多拠点最適化）が営業利益率の改善（11.8%→13.1%）に直結している。",[19,5866,5867,5868,5871],{},"ただし、Morningstarの評価は現在 ",[23,5869,5870],{},"「ほぼフェアバリュー（★★★、3スター）」"," だ。株価は過去1ヶ月で25%反発しており、好材料の多くが既に織り込まれた水準にある。2026年の設備投資計画2,000億ドルは、中期的なフリーキャッシュフローを大幅に押し下げる。長期保有の観点では依然として魅力的だが、短期のエントリーポイントとしては「待ち」のスタンスが合理的かもしれない。",[122,5873,5875],{"id":5874},"alphabet-google-クラウドの急加速でai敗者の烙印を返上","Alphabet (Google) — クラウドの急加速で「AI敗者」の烙印を返上",[19,5877,5878],{},"1年前、市場はAlphabetをAI競争の「敗者」と見ていた。OpenAI台頭後の検索市場縮小懸念が株価を下押しし、Morningstarも慎重なトーンを維持していた。しかし今四半期の決算がその見方を根本から変えた。",[19,5880,5881],{},"Google Cloudの成長率は前年比63%という驚異的な水準に達し、売上は200億ドルを記録した。Morningstarによると、Gemini APIだけで年率約150億ドルの収益を上げており、これは前四半期の90億ドルからわずか1四半期で急加速した。Morningstarはフェアバリュー推定値を340ドルから433ドルへと大幅に引き上げた。この引き上げの背景には、TPU（自社カスタムシリコン）のマネタイズ可能性への信頼感の高まりと、GeminiというLLMの商業展開の加速がある。",[19,5883,5884],{},"検索事業も健在だ。AI OverviewとAI Modeの導入が、AIチャットボットへの流出ではなく、既存ユーザーのエンゲージメントを高める形で機能しており、検索広告売上は前年比19%増を記録した。「AIで検索が食われる」という最大の懸念が、今のところ杞憂に終わっていることを数字が証明している。",[19,5886,5887],{},"全社の営業利益率は前年比220ベーシスポイント改善しており、クラウド・サービス両セグメントともに利益率が拡大した。現在株価$350に対してMorningstarのフェアバリュー$433は19%の割安を示しており、バリュエーションの観点からもMetaと並んで魅力的な水準だ。",[122,5889,5891],{"id":5890},"microsoft-aiの商業化で最も盤石圧倒的な割安感","Microsoft — AIの商業化で最も盤石、圧倒的な割安感",[19,5893,5894],{},"4社の中でMicrosoftは別格の存在感を示した。売上は前年比18%増の829億ドルでガイダンス上限を上回り、EPS4.27ドルは市場予想4.06ドルを大幅に超過した。Azure成長率は40%に達し、AIビジネスだけで年率370億ドルという規模に達した（前年比+123%）。",[19,5896,5897],{},"特筆すべきはAzureのキャパシティ制約だ。旺盛な需要に対してサプライが追いついておらず、マネジメントは「少なくとも2026年末まで供給制約が続く」と言明した。これは弱材料のように見えるが、裏を返せば需要は確実に存在しており、設備が稼働し次第そのまま売上に変換されるという好材料でもある。2026年の設備投資計画として1,900億ドルという数字を示したが、これはまさにその旺盛な需要に応えるための投資だ。",[19,5899,5900,5901,5904],{},"Morningstarはフェアバリュー推定値を600ドルに据え置きつつも（成長予想を引き上げ、利益率予測は一部引き下げ）、現在株価424ドルに対して ",[23,5902,5903],{},"★★★★★（5スター）"," という最高評価を維持した。これは「大幅割安」を意味し、4社の中で最も強いバイシグナルだ。OpenAIとのパートナーシップ、Copilotの法人展開（1,500万シート）、Azure・Microsoft 365の企業向け需要の複合的な成長エンジンが、他社にはない安定性を提供している。",[19,5906,5907],{},"Remaining Performance Obligation（RPO）が前年比99%増の6,270億ドルに膨らんでいることも、先行き安心感の根拠だ。この数字は今後12ヶ月以内に25%が売上認識されるため、短期的な売上の可視性が極めて高い。",[205,5909],{},[32,5911,5913],{"id":5912},"ai投資効率誰が最もroiを上げているか","AI投資効率：誰が最もROIを上げているか",[19,5915,5916],{},"4社の「AI投資→リターン」の構造は以下の通り整理できる。",[54,5918,5919,5935],{},[57,5920,5921],{},[60,5922,5923,5926,5929,5932],{},[63,5924,5925],{},"会社",[63,5927,5928],{},"AI投資の主な用途",[63,5930,5931],{},"AI ROIのメカニズム",[63,5933,5934],{},"ROI実現の確実性",[70,5936,5937,5950,5963,5977],{},[60,5938,5939,5941,5944,5947],{},[75,5940,5667],{},[75,5942,5943],{},"広告最適化・コンテンツ推薦",[75,5945,5946],{},"広告単価・量の直接向上",[75,5948,5949],{},"高（既に数字に表れている）",[60,5951,5952,5954,5957,5960],{},[75,5953,5670],{},[75,5955,5956],{},"AWSインフラ・Trainium",[75,5958,5959],{},"AIクラウドサービス収益",[75,5961,5962],{},"高（需要旺盛、容量制約）",[60,5964,5965,5968,5971,5974],{},[75,5966,5967],{},"Alphabet",[75,5969,5970],{},"Google Cloud/Gemini API",[75,5972,5973],{},"クラウド収益+検索広告防衛",[75,5975,5976],{},"高（急加速中）",[60,5978,5979,5981,5984,5987],{},[75,5980,5676],{},[75,5982,5983],{},"Azure AI/Copilot",[75,5985,5986],{},"クラウド収益+ライセンス",[75,5988,5989],{},"非常に高（5スター評価）",[19,5991,5992],{},"4社ともAI投資のROIが実現しつつある。ただし、Metaは広告事業への集中度が高くクラウド収益源を持たないため、AIの上振れポテンシャルがコアの広告事業改善に限定される。一方、Microsoft・Amazon・Alphabetは「AIインフラを第三者に提供する」というより汎用的なクラウドビジネスモデルを有しており、AIの普及とともに需要が爆発的に拡大する構造だ。",[205,5994],{},[32,5996,5997],{"id":5997},"バリュエーション比較",[54,5999,6000,6014],{},[57,6001,6002],{},[60,6003,6004,6006,6008,6010,6012],{},[63,6005],{},[63,6007,5667],{},[63,6009,5670],{},[63,6011,5967],{},[63,6013,5676],{},[70,6015,6016,6028,6043,6056,6071,6085,6100,6115,6132,6146],{},[60,6017,6018,6020,6022,6024,6026],{},[75,6019,5800],{},[75,6021,5803],{},[75,6023,5806],{},[75,6025,5809],{},[75,6027,5812],{},[60,6029,6030,6033,6035,6038,6041],{},[75,6031,6032],{},"フェアバリュー",[75,6034,5786],{},[75,6036,6037],{},"$280",[75,6039,6040],{},"$433",[75,6042,5795],{},[60,6044,6045,6048,6050,6052,6054],{},[75,6046,6047],{},"対FVE乖離率",[75,6049,5820],{},[75,6051,5823],{},[75,6053,5826],{},[75,6055,5829],{},[60,6057,6058,6060,6063,6066,6068],{},[75,6059,5767],{},[75,6061,6062],{},"★★★★",[75,6064,6065],{},"★★★",[75,6067,6062],{},[75,6069,6070],{},"★★★★★",[60,6072,6073,6076,6079,6081,6083],{},[75,6074,6075],{},"経済的堀",[75,6077,6078],{},"Wide",[75,6080,6078],{},[75,6082,6078],{},[75,6084,6078],{},[60,6086,6087,6090,6093,6096,6098],{},[75,6088,6089],{},"不確実性評価",[75,6091,6092],{},"High",[75,6094,6095],{},"Medium",[75,6097,6095],{},[75,6099,6095],{},[60,6101,6102,6105,6108,6111,6113],{},[75,6103,6104],{},"資本配分評価",[75,6106,6107],{},"Standard",[75,6109,6110],{},"Exemplary",[75,6112,6110],{},[75,6114,6110],{},[60,6116,6117,6120,6123,6126,6129],{},[75,6118,6119],{},"時価総額",[75,6121,6122],{},"$1.70兆",[75,6124,6125],{},"$2.83兆",[75,6127,6128],{},"$4.23兆",[75,6130,6131],{},"$3.15兆",[60,6133,6134,6137,6139,6141,6144],{},[75,6135,6136],{},"設備投資計画（2026）",[75,6138,5753],{},[75,6140,5756],{},[75,6142,6143],{},"大幅増",[75,6145,5762],{},[60,6147,6148,6151,6153,6156,6158],{},[75,6149,6150],{},"5年売上CAGR予想",[75,6152,665],{},[75,6154,6155],{},"11%",[75,6157,5720],{},[75,6159,6160],{},"13%",[205,6162],{},[32,6164,6166],{"id":6165},"シナリオ分析強気メインテールリスク","シナリオ分析：強気・メイン・テールリスク",[54,6168,6169,6186],{},[57,6170,6171],{},[60,6172,6173,6175,6178,6180,6182,6184],{},[63,6174,601],{},[63,6176,6177],{},"想定",[63,6179,5667],{},[63,6181,5670],{},[63,6183,5967],{},[63,6185,5676],{},[70,6187,6188,6210,6232],{},[60,6189,6190,6195,6198,6201,6204,6207],{},[75,6191,6192],{},[23,6193,6194],{},"強気",[75,6196,6197],{},"AI投資の全面回収・独占強化・規制クリア",[75,6199,6200],{},"$1,100-1,200",[75,6202,6203],{},"$340-360",[75,6205,6206],{},"$550-600",[75,6208,6209],{},"$700-750",[60,6211,6212,6217,6220,6223,6226,6229],{},[75,6213,6214],{},[23,6215,6216],{},"メイン",[75,6218,6219],{},"現在のトレンド継続・FVEに向けた収束",[75,6221,6222],{},"$800-900",[75,6224,6225],{},"$270-300",[75,6227,6228],{},"$410-460",[75,6230,6231],{},"$570-630",[60,6233,6234,6239,6242,6245,6248,6251],{},[75,6235,6236],{},[23,6237,6238],{},"テール",[75,6240,6241],{},"規制当局による制限・ROI悪化・景気後退",[75,6243,6244],{},"$400-500",[75,6246,6247],{},"$180-220",[75,6249,6250],{},"$250-300",[75,6252,6253],{},"$360-420",[19,6255,6256,6259],{},[23,6257,6258],{},"強気シナリオ"," の実現確率はMicrosoftとAlphabetで最も高い。MicrosoftはOpenAIパートナーシップと企業向けAIアプリケーションの展開が確実に進んでおり、Alphabetはクラウドと検索の両方でAI優位性を証明した。Metaの強気シナリオはReality Labsの収益化と独禁訴訟の解決が前提となる。",[19,6261,6262,6265],{},[23,6263,6264],{},"テールリスク"," で最も注意が必要なのはMetaだ。独禁訴訟（FTC対Meta）の判決次第では事業分割の可能性がゼロではなく、ここが他3社にはない固有のリスクとなっている。",[205,6267],{},[32,6269,764],{"id":764},[19,6271,6272,6274],{},[23,6273,5667],{}," — 2026年Q2決算でAI広告の成長が継続するか。Reality Labsへの設備投資が更に拡大するか。独禁訴訟の進展。",[19,6276,6277,6279],{},[23,6278,5670],{}," — AWSの成長率が28%以上を維持できるか。Project Kuiper（衛星インターネット）の進捗。Q2の設備投資による利益率への影響。",[19,6281,6282,6284],{},[23,6283,5967],{}," — Google Cloud成長率が60%超を継続できるか。AI OverviewによるSEO構造変化の影響。Gemini APIの単価・利用量動向。",[19,6286,6287,6289],{},[23,6288,5676],{}," — Azureのキャパシティ制約が予告通り2026年後半に解消に向かうか。Copilotの法人展開シート数の推移。OpenAIとのプロフィットシェアリング（2030年まで継続）の動向。",[205,6291],{},[32,6293,6294],{"id":6294},"投資判断",[19,6296,6297,6300],{},[23,6298,6299],{},"Microsoft（MSFT）：強気ロング。"," Morningstar5スター、FVE比29%割安という数字が全てを語っている。企業向けAIの商業化で最も安定した収益構造を持ち、Azureの供給制約解消が下半期の成長加速カタリストとなる公算が高い。",[19,6302,6303,6306],{},[23,6304,6305],{},"Alphabet（GOOGL）：中期ロング。"," Google Cloudの爆発的成長がファンダメンタルズを変えた。FVEが$340→$433に引き上げられたにもかかわらず依然として19%割安。AI検索との両立という懸念が、今四半期の数字で払拭されつつある。",[19,6308,6309,6312],{},[23,6310,6311],{},"Meta（META）：短期ニュートラル・中期ロング。"," 広告AIエンジンの実力は本物だが、設備投資増加と独禁リスクが短期的な株価の上値を抑える。$600以下が積極的なエントリーポイントとして機能するか注視。",[19,6314,6315,6318],{},[23,6316,6317],{},"Amazon（AMZN）：短期ニュートラル。"," 株価が1ヶ月で25%上昇しFVEとほぼ同水準。中長期のAWS成長シナリオは不変だが、今の株価では織り込みが進みすぎている。押し目待ちが合理的。",[19,6320,6321,6322,6325],{},"結論として、2026年Q1決算は ",[23,6323,6324],{},"「AIの黒字化元年」"," を示す転換点となった。4社ともAI投資が本格的にリターンを生み始めているが、ROIの確実性とバリュエーションの組み合わせで見ると、現時点で最も魅力的なのはMicrosoft、次いでAlphabetだ。MetaとAmazonは固有の論点（独禁・織り込み）を慎重に見極めたい局面である。",[19,6327,6328],{},[802,6329,6330],{},"本記事は投資情報の提供を目的としたものであり、特定の金融商品の売買を勧誘するものではありません。投資に関する最終的な判断はご自身の責任のもとで行ってください。株式投資には価格変動リスク・為替リスク・流動性リスクを含む様々なリスクが伴い、投資元本が保証されるものではありません。",[205,6332],{},[10,6334,6335,6339,6350,6354,6357,6368,6374,6377,6379,6383,6540,6543,6545,6549,6553,6560,6567,6577,6581,6584,6587,6598,6602,6609,6612,6619,6622,6626,6629,6640,6647,6653,6655,6659,6662,6733,6744,6746,6750,6900,6902,6906,6981,6988,6994,6996,7000,7005,7010,7015,7020,7022,7026,7032,7038,7044,7050,7057],{"lang":809},[14,6336,6338],{"id":6337},"decoding-big-tech-q1-2026-earnings-a-comparative-analysis-of-meta-amazon-google-and-microsoft","Decoding Big Tech Q1 2026 Earnings: A Comparative Analysis of Meta, Amazon, Google, and Microsoft",[19,6340,6341,6342,6345,6346,6349],{},"I read Big Tech earnings every quarter from a CVC seat — both for company cloud-and-AI-tool decisions and for our investment portfolio. With AI capex now north of ",[23,6343,6344],{},"$600B annually",", the only useful frame, in my experience, is to separate ",[23,6347,6348],{},"\"ad-AI,\" \"cloud-AI,\" and \"ecosystem-AI\""," and grade each company against the right ROI structure. Q1 2026 is the first quarter where those three different ROI structures showed up clearly in the numbers — that's the lens this piece is built around.",[32,6351,6353],{"id":6352},"a-synchronized-print-that-marks-year-one-of-ai-monetization","A Synchronized Print That Marks \"Year One of AI Monetization\"",[19,6355,6356],{},"Let me put the quarter's structure down flat first.",[19,6358,6359,6360,6363,6364,6367],{},"In late April 2026, the US Big Tech cohort delivered Q1 2026 earnings in rapid succession. The question that weighed on the market through 2024 and 2025 — ",[23,6361,6362],{},"\"is the massive AI capex actually converting into revenue?\""," — finally got a meaningful answer. Meta alone raised its 2026 capex plan to $135B; combined with Amazon, Microsoft, and Alphabet, the four-company total is expected to exceed ",[23,6365,6366],{},"$600B in 2026",". Quite literally, a scale of investment without precedent.",[19,6369,6370,6371,849],{},"This quarter, the answer came back in concrete form. Google Cloud is now generating Gemini API revenue at roughly a $150B annual run-rate. Microsoft's AI business reached a $37B annual run-rate, up 123% YoY. Meta is monetizing AI in real time through higher ad conversion and longer time-on-platform. Amazon's growth re-accelerated, driven by AWS AI workloads. The answer to \"can AI actually make money?\" is, this quarter, ",[23,6372,6373],{},"clearly tilting toward yes",[19,6375,6376],{},"But the four companies make money from AI in structurally different ways. Valuation, momentum, and risk profiles diverge meaningfully. Drawing on Morningstar analyst reports and live market data, this piece compares the four across multiple dimensions and asks where the real investment opportunity sits today.",[205,6378],{},[32,6380,6382],{"id":6381},"q1-earnings-side-by-side-kpis","Q1 Earnings — Side-by-Side KPIs",[54,6384,6385,6399],{},[57,6386,6387],{},[60,6388,6389,6391,6393,6395,6397],{},[63,6390,1015],{},[63,6392,5667],{},[63,6394,5670],{},[63,6396,5673],{},[63,6398,5676],{},[70,6400,6401,6418,6434,6449,6466,6483,6499,6514,6527],{},[60,6402,6403,6406,6409,6412,6415],{},[75,6404,6405],{},"Revenue (YoY)",[75,6407,6408],{},"$56.0B (+33%)",[75,6410,6411],{},"$181.5B (+15%)",[75,6413,6414],{},"$110.0B (+22%)",[75,6416,6417],{},"$82.9B (+18%)",[60,6419,6420,6423,6426,6429,6432],{},[75,6421,6422],{},"Operating margin",[75,6424,6425],{},"41% (-90bp)",[75,6427,6428],{},"13.1% (+130bp)",[75,6430,6431],{},"34% (+220bp)",[75,6433,5712],{},[60,6435,6436,6438,6441,6444,6446],{},[75,6437,5717],{},[75,6439,6440],{},"n/d",[75,6442,6443],{},"$8.47 (E/F)",[75,6445,6440],{},[75,6447,6448],{},"$4.27 (vs $4.06 cons.)",[60,6450,6451,6454,6457,6460,6463],{},[75,6452,6453],{},"Cloud / AI KPI",[75,6455,6456],{},"Ad price +12%, vol +19%",[75,6458,6459],{},"AWS +28%, $150B run-rate",[75,6461,6462],{},"Cloud +63%, $20B",[75,6464,6465],{},"Azure +40%, AI $37B run-rate",[60,6467,6468,6471,6474,6477,6480],{},[75,6469,6470],{},"2026 capex plan",[75,6472,6473],{},"$135B",[75,6475,6476],{},"$200B",[75,6478,6479],{},"Sharp increase (n/d)",[75,6481,6482],{},"$190B",[60,6484,6485,6488,6491,6494,6496],{},[75,6486,6487],{},"Morningstar rating",[75,6489,6490],{},"★★★★ Undervalued",[75,6492,6493],{},"★★★ Near fair value",[75,6495,6490],{},[75,6497,6498],{},"★★★★★ Materially undervalued",[60,6500,6501,6504,6506,6509,6512],{},[75,6502,6503],{},"Fair value estimate",[75,6505,5786],{},[75,6507,6508],{},"$280 (raised from $260)",[75,6510,6511],{},"$433 (raised from $340)",[75,6513,5795],{},[60,6515,6516,6519,6521,6523,6525],{},[75,6517,6518],{},"Current price",[75,6520,5803],{},[75,6522,5806],{},[75,6524,5809],{},[75,6526,5812],{},[60,6528,6529,6532,6534,6536,6538],{},[75,6530,6531],{},"Discount to FV",[75,6533,5820],{},[75,6535,5823],{},[75,6537,5826],{},[75,6539,5829],{},[19,6541,6542],{},"Even at this level, the table makes two things intuitive: Microsoft is the most attractive on valuation, and Amazon offers the most balanced risk/reward today.",[205,6544],{},[32,6546,6548],{"id":6547},"company-by-company-deep-dive","Company-by-Company Deep Dive",[122,6550,6552],{"id":6551},"meta-the-ad-ai-engine-has-arrived-capex-becomes-the-question","Meta — The Ad-AI Engine Has Arrived; Capex Becomes the Question",[19,6554,6555,6556,6559],{},"If I had to summarize Meta's quarter in one line: ",[23,6557,6558],{},"the ad-AI engine has finally come of age",". Q1 revenue grew 33% YoY to $56B. Time spent on AI-recommended Reels content rose 10%, Facebook video time +8%, and Instagram conversion rate +1.6%. These are direct fingerprints of AI capex flowing into ad revenue.",[19,6561,6562,6563,6566],{},"The structure is ideal: AI-optimized ad algorithms are pushing both ad price (+12% YoY) and ad volume (+19%) at the same time. Morningstar's emphasis is well-placed — Meta's reach of 3.8B+ daily active users lets the ",[23,6564,6565],{},"scale economics"," of AI investment actually work. The bigger the data corpus, the better the model, the better the ad performance, the higher the ad price. A self-reinforcing flywheel.",[19,6568,6569,6570,6573,6574,849],{},"The pushback came from capex. Guidance was raised from $125B to $135B, and the stock sold off after-hours. Reality Labs (VR/AR) continues to bleed roughly $16B+ in annual operating losses, amplifying investor concern over discipline. Morningstar held its $850 fair value, leaving the current $669 share price 21% undervalued. The fork is whether to read the post-print drawdown as ",[23,6571,6572],{},"excessive worry about AI capex"," or as ",[23,6575,6576],{},"a fair pricing of execution risk",[122,6578,6580],{"id":6579},"amazon-aws-re-accelerates-e-commerce-resilience-confirmed","Amazon — AWS Re-accelerates; E-commerce Resilience Confirmed",[19,6582,6583],{},"Amazon's biggest surprise this quarter was the re-acceleration of AWS. AWS revenue growth hit 28% YoY, on a $150B annual run-rate. Both AI and traditional cloud workloads are showing strong demand. Every segment beat Morningstar's estimates, and Morningstar raised its fair value from $260 to $280.",[19,6585,6586],{},"AWS is only 17% of total revenue but produces the bulk of consolidated operating profit — that structural fact is unchanged. Advertising remains a high-growth, high-margin engine, and operational efficiency gains in e-commerce (robotics, multi-node optimization) flow straight to the operating margin (11.8% → 13.1%).",[19,6588,6589,6590,6593,6594,6597],{},"That said, Morningstar's call is currently ",[23,6591,6592],{},"\"near fair value\""," (★★★, 3 stars). The stock has rebounded 25% over the past month, and most of the good news appears priced in. The 2026 capex plan of $200B will weigh on free cash flow for several years. For long-duration holders, AWS is still a great franchise; as a near-term entry, ",[23,6595,6596],{},"waiting for a pullback"," is the more rational stance.",[122,6599,6601],{"id":6600},"alphabet-google-cloud-re-acceleration-erases-the-ai-loser-label","Alphabet (Google) — Cloud Re-acceleration Erases the \"AI Loser\" Label",[19,6603,6604,6605,6608],{},"A year ago, the market saw Alphabet as the AI-race ",[23,6606,6607],{},"loser",". Concern that OpenAI would erode search depressed the stock, and Morningstar held a cautious tone. This quarter changed that view at the foundation level.",[19,6610,6611],{},"Google Cloud grew 63% YoY, with revenue of $20B. Per Morningstar, Gemini API alone is at roughly a $15B annual run-rate, up sharply from $9B just one quarter earlier. Morningstar's fair value was raised from $340 to $433. The drivers behind the upgrade are growing confidence in TPU monetization (Google's custom silicon) and the accelerating commercial deployment of Gemini.",[19,6613,6614,6615,6618],{},"Search itself is also intact. The rollout of AI Overview and AI Mode has, so far, ",[23,6616,6617],{},"deepened existing-user engagement"," rather than leaking traffic to chatbots — search ad revenue grew 19% YoY. The biggest fear (that AI would eat search) is, for now, contradicted by the numbers.",[19,6620,6621],{},"Consolidated operating margin expanded by 220bp YoY, with margin gains in both Cloud and Services. The current $350 share price against Morningstar's $433 FV implies 19% upside — putting Alphabet alongside Meta as the most attractive valuation pair.",[122,6623,6625],{"id":6624},"microsoft-the-most-solid-ai-monetization-story-the-cleanest-discount","Microsoft — The Most Solid AI Monetization Story; The Cleanest Discount",[19,6627,6628],{},"Among the four, Microsoft was in a class of its own. Revenue grew 18% YoY to $82.9B, beating the high end of guidance, and EPS of $4.27 cleared consensus of $4.06 by a wide margin. Azure grew 40%, with AI alone at a $37B annual run-rate (+123% YoY).",[19,6630,6631,6632,6635,6636,6639],{},"The detail that stood out was Azure's ",[23,6633,6634],{},"capacity constraint",". Demand is outrunning supply, and management explicitly said ",[23,6637,6638],{},"\"supply will remain tight at least through end-2026.\""," On the surface that reads negative, but the flip side is that demand is not in question — every new GW of capacity converts straight to revenue. The $190B 2026 capex plan is exactly the build-out that demand requires.",[19,6641,6642,6643,6646],{},"Morningstar held the $600 fair value (raising the growth outlook while trimming margin assumptions in places) and reaffirmed the ",[23,6644,6645],{},"★★★★★ (5 stars)"," rating against the current $424 share price. That is \"materially undervalued\" — the strongest buy signal among the four. The combination of the OpenAI partnership, Copilot enterprise deployment (15M seats), and the bundled Azure / Microsoft 365 enterprise demand provides a diversification of growth engines that no peer matches.",[19,6648,6649,6650,849],{},"Remaining Performance Obligations (RPO) ballooned 99% YoY to $627B. Roughly 25% of that is recognized as revenue within 12 months, giving Microsoft ",[23,6651,6652],{},"unusually high near-term revenue visibility",[205,6654],{},[32,6656,6658],{"id":6657},"ai-roi-who-is-earning-the-best-return","AI ROI — Who Is Earning the Best Return?",[19,6660,6661],{},"The four companies' \"AI capex → AI return\" structures map out as follows.",[54,6663,6664,6679],{},[57,6665,6666],{},[60,6667,6668,6670,6673,6676],{},[63,6669,1228],{},[63,6671,6672],{},"Primary AI use",[63,6674,6675],{},"ROI mechanism",[63,6677,6678],{},"ROI certainty",[70,6680,6681,6694,6707,6720],{},[60,6682,6683,6685,6688,6691],{},[75,6684,5667],{},[75,6686,6687],{},"Ad optimization & content reco",[75,6689,6690],{},"Direct lift in ad price & volume",[75,6692,6693],{},"High (already in the numbers)",[60,6695,6696,6698,6701,6704],{},[75,6697,5670],{},[75,6699,6700],{},"AWS infra & Trainium",[75,6702,6703],{},"AI cloud service revenue",[75,6705,6706],{},"High (demand-led, capacity-limited)",[60,6708,6709,6711,6714,6717],{},[75,6710,5967],{},[75,6712,6713],{},"Google Cloud / Gemini API",[75,6715,6716],{},"Cloud revenue + search-ad defense",[75,6718,6719],{},"High (re-accelerating)",[60,6721,6722,6724,6727,6730],{},[75,6723,5676],{},[75,6725,6726],{},"Azure AI / Copilot",[75,6728,6729],{},"Cloud revenue + license attach",[75,6731,6732],{},"Very high (Morningstar 5 stars)",[19,6734,6735,6736,6739,6740,6743],{},"All four are converting AI capex into return. But Meta is ",[23,6737,6738],{},"concentrated in advertising"," and lacks a third-party cloud revenue stream, so AI upside is bounded by the core ad business. Microsoft, Amazon, and Alphabet operate the more general-purpose ",[23,6741,6742],{},"\"AI infra-as-a-service\""," model, where adoption tailwinds compound directly into revenue.",[205,6745],{},[32,6747,6749],{"id":6748},"valuation-comparison","Valuation Comparison",[54,6751,6752,6766],{},[57,6753,6754],{},[60,6755,6756,6758,6760,6762,6764],{},[63,6757],{},[63,6759,5667],{},[63,6761,5670],{},[63,6763,5967],{},[63,6765,5676],{},[70,6767,6768,6780,6793,6805,6817,6830,6843,6856,6873,6887],{},[60,6769,6770,6772,6774,6776,6778],{},[75,6771,6518],{},[75,6773,5803],{},[75,6775,5806],{},[75,6777,5809],{},[75,6779,5812],{},[60,6781,6782,6785,6787,6789,6791],{},[75,6783,6784],{},"Fair value",[75,6786,5786],{},[75,6788,6037],{},[75,6790,6040],{},[75,6792,5795],{},[60,6794,6795,6797,6799,6801,6803],{},[75,6796,6531],{},[75,6798,5820],{},[75,6800,5823],{},[75,6802,5826],{},[75,6804,5829],{},[60,6806,6807,6809,6811,6813,6815],{},[75,6808,6487],{},[75,6810,6062],{},[75,6812,6065],{},[75,6814,6062],{},[75,6816,6070],{},[60,6818,6819,6822,6824,6826,6828],{},[75,6820,6821],{},"Economic moat",[75,6823,6078],{},[75,6825,6078],{},[75,6827,6078],{},[75,6829,6078],{},[60,6831,6832,6835,6837,6839,6841],{},[75,6833,6834],{},"Uncertainty rating",[75,6836,6092],{},[75,6838,6095],{},[75,6840,6095],{},[75,6842,6095],{},[60,6844,6845,6848,6850,6852,6854],{},[75,6846,6847],{},"Capital allocation",[75,6849,6107],{},[75,6851,6110],{},[75,6853,6110],{},[75,6855,6110],{},[60,6857,6858,6861,6864,6867,6870],{},[75,6859,6860],{},"Market cap",[75,6862,6863],{},"$1.70T",[75,6865,6866],{},"$2.83T",[75,6868,6869],{},"$4.23T",[75,6871,6872],{},"$3.15T",[60,6874,6875,6878,6880,6882,6885],{},[75,6876,6877],{},"2026 capex",[75,6879,6473],{},[75,6881,6476],{},[75,6883,6884],{},"Sharply up",[75,6886,6482],{},[60,6888,6889,6892,6894,6896,6898],{},[75,6890,6891],{},"5-year revenue CAGR (est.)",[75,6893,665],{},[75,6895,6155],{},[75,6897,6440],{},[75,6899,6160],{},[205,6901],{},[32,6903,6905],{"id":6904},"scenarios-bull-base-tail","Scenarios — Bull, Base, Tail",[54,6907,6908,6925],{},[57,6909,6910],{},[60,6911,6912,6914,6917,6919,6921,6923],{},[63,6913,1389],{},[63,6915,6916],{},"Assumptions",[63,6918,5667],{},[63,6920,5670],{},[63,6922,5967],{},[63,6924,5676],{},[70,6926,6927,6945,6963],{},[60,6928,6929,6934,6937,6939,6941,6943],{},[75,6930,6931],{},[23,6932,6933],{},"Bull",[75,6935,6936],{},"Full AI ROI, monopoly entrenchment, regulatory clearance",[75,6938,6200],{},[75,6940,6203],{},[75,6942,6206],{},[75,6944,6209],{},[60,6946,6947,6952,6955,6957,6959,6961],{},[75,6948,6949],{},[23,6950,6951],{},"Base",[75,6953,6954],{},"Current trends continue; convergence toward FV",[75,6956,6222],{},[75,6958,6225],{},[75,6960,6228],{},[75,6962,6231],{},[60,6964,6965,6970,6973,6975,6977,6979],{},[75,6966,6967],{},[23,6968,6969],{},"Tail",[75,6971,6972],{},"Antitrust action, AI ROI disappoints, recession",[75,6974,6244],{},[75,6976,6247],{},[75,6978,6250],{},[75,6980,6253],{},[19,6982,6983,6984,6987],{},"The ",[23,6985,6986],{},"bull case probability"," is highest for Microsoft and Alphabet. Microsoft's OpenAI partnership and enterprise AI rollout are advancing on schedule. Alphabet has now demonstrated AI strength in both Cloud and Search. Meta's bull case is contingent on Reality Labs eventually monetizing and the antitrust suit resolving favorably.",[19,6989,6983,6990,6993],{},[23,6991,6992],{},"biggest tail risk"," sits at Meta. Depending on the FTC v. Meta ruling, structural breakup is not zero-probability — a name-specific risk that the other three do not share.",[205,6995],{},[32,6997,6999],{"id":6998},"what-to-watch-from-here","What to Watch From Here",[19,7001,7002,7004],{},[23,7003,5667],{}," — Whether Q2 2026 sustains the ad-AI growth trajectory. Whether Reality Labs capex expands further. Antitrust case progress.",[19,7006,7007,7009],{},[23,7008,5670],{}," — Whether AWS holds 28%+ growth. Project Kuiper (satellite internet) execution. The margin impact of the Q2 capex ramp.",[19,7011,7012,7014],{},[23,7013,5967],{}," — Whether Google Cloud sustains 60%+ growth. SEO structural shifts driven by AI Overview. Gemini API price/volume dynamics.",[19,7016,7017,7019],{},[23,7018,5676],{}," — Whether Azure capacity constraints ease in H2 2026 as guided. Copilot enterprise seat trajectory. The OpenAI profit-sharing arrangement (in place through 2030).",[205,7021],{},[32,7023,7025],{"id":7024},"investment-calls","Investment Calls",[19,7027,7028,7031],{},[23,7029,7030],{},"Microsoft (MSFT) — Bullish long."," Morningstar 5 stars and a 29% discount to FV say the rest. The most stable enterprise-AI revenue mix, with capacity-constraint relief in H2 acting as a likely growth catalyst.",[19,7033,7034,7037],{},[23,7035,7036],{},"Alphabet (GOOGL) — Medium-term long."," Google Cloud's explosive growth has changed the fundamentals. Despite the FV upgrade from $340 to $433, the stock still trades at a 19% discount. The \"AI cannibalizes search\" fear is being neutralized by this quarter's data.",[19,7039,7040,7043],{},[23,7041,7042],{},"Meta (META) — Short-term neutral, medium-term long."," The ad-AI engine is real, but rising capex and antitrust overhang will cap the near-term upside. Watching whether sub-$600 acts as a more aggressive entry.",[19,7045,7046,7049],{},[23,7047,7048],{},"Amazon (AMZN) — Short-term neutral."," The stock has run 25% in the past month and trades near FV. The long-run AWS thesis is unchanged, but at current levels much of the good news is in the price. Wait for a pullback.",[19,7051,7052,7053,7056],{},"The bigger picture: Q1 2026 marked the inflection into ",[23,7054,7055],{},"\"year one of AI monetization.\""," All four are turning AI capex into real returns. But weighing ROI certainty against valuation, the most attractive name today is Microsoft, followed by Alphabet. Meta and Amazon each carry name-specific considerations (antitrust, mean reversion of recent gains) that warrant patience.",[19,7058,7059],{},[802,7060,7061],{},"This article is for informational purposes only and is not a recommendation to buy or sell any financial instrument. Please make investment decisions at your own discretion. Equity investing involves price, FX, and liquidity risks; principal is not guaranteed.",{"title":143,"searchDepth":1605,"depth":1605,"links":7063},[7064,7065,7066,7072,7073,7074,7075,7076,7077,7078,7079,7085,7086,7087,7088,7089],{"id":5626,"depth":1605,"text":5627},{"id":5652,"depth":1605,"text":5653},{"id":5837,"depth":1605,"text":5837,"children":7067},[7068,7069,7070,7071],{"id":5840,"depth":1610,"text":5841},{"id":5857,"depth":1610,"text":5858},{"id":5874,"depth":1610,"text":5875},{"id":5890,"depth":1610,"text":5891},{"id":5912,"depth":1605,"text":5913},{"id":5997,"depth":1605,"text":5997},{"id":6165,"depth":1605,"text":6166},{"id":764,"depth":1605,"text":764},{"id":6294,"depth":1605,"text":6294},{"id":6352,"depth":1605,"text":6353},{"id":6381,"depth":1605,"text":6382},{"id":6547,"depth":1605,"text":6548,"children":7080},[7081,7082,7083,7084],{"id":6551,"depth":1610,"text":6552},{"id":6579,"depth":1610,"text":6580},{"id":6600,"depth":1610,"text":6601},{"id":6624,"depth":1610,"text":6625},{"id":6657,"depth":1605,"text":6658},{"id":6748,"depth":1605,"text":6749},{"id":6904,"depth":1605,"text":6905},{"id":6998,"depth":1605,"text":6999},{"id":7024,"depth":1605,"text":7025},"2026年Q1決算で米国Big Tech4社のAI投資が数字に表れ始めた。バリュエーション・モメンタム・リスクプロファイルを多角的に比較し、どこに投資機会があるかを徹底検討する。",{"date":7092,"image":7093,"alt":7094,"tags":7095,"tagsEn":7096,"published":1666},"30th Apr 2026","/blogs-img/blog-big-tech-q1-2026-comparison.png","Meta・Amazon・Google・Microsoftの2026年Q1決算比較サマリ",[4240,1657,4238,4241,5599],[4245,1662,4243,4241,5599],"/blogs/11-big-tech-q1-2026-comparison",{"title":5608,"description":7090},"blogs/11-big-tech-q1-2026-comparison","4SQCaPMWBAs87mtOuLLjJ2uLlNs4kncUBKM8bKDqAd0",{"id":7102,"title":7103,"body":7104,"description":7634,"extension":1651,"meta":7635,"navigation":1666,"ogImage":7637,"path":7650,"seo":7651,"stem":7652,"__hash__":7653},"content/blogs/12-claude-skills-mcp-practical-guide.md","Claudeで動画とウェブを\"全自動\"にしたら制作時間が1/5になった話 | How Claude Cut My Video + Web Workflow to 1/5 the Time",{"type":7,"value":7105,"toc":7618},[7106,7356,7358],[10,7107,7108,7112,7119,7121,7125,7132,7152,7159,7166,7168,7172,7179,7182,7202,7209,7216,7218,7222,7225,7229,7251,7255,7277,7284,7286,7290,7293,7313,7320,7326,7329,7331,7336],{"lang":12},[14,7109,7111],{"id":7110},"claudeで動画とウェブを全自動にしたら制作時間が15になった話","Claudeで動画とウェブを\"全自動\"にしたら制作時間が1/5になった話",[19,7113,7114,7115,7118],{},"私はVCの仕事をしながらYouTube・X・このブログをひとりで回しているのですが、2026年4月にClaude Designが出て以降、",[23,7116,7117],{},"ひとつの動画とランディングページを作るのに毎週使っていた15時間が、3時間まで縮みました","。最初は「またAIの誇大広告か」と疑っていたのが正直なところです。でもSkills・MCP・Claude Designの3つを\"つなぎ方\"として使い始めた瞬間、見える景色が変わりました。今回はその\"つなぎ方\"を、実体験ベースでまるごと共有します。",[205,7120],{},[32,7122,7124],{"id":7123},"ツールが増えるほど時間が消えていくあるある","「ツールが増えるほど時間が消えていく」あるある",[19,7126,7127,7128,7131],{},"ぶっちゃけ、ここ2年で一番ストレスだったのは",[23,7129,7130],{},"ツールの数が増えれば増えるほど自分の作業時間が減らない","ことでした。",[312,7133,7134,7137,7140,7143,7146,7149],{},[315,7135,7136],{},"動画台本はChatGPT",[315,7138,7139],{},"編集はCapCutかDaVinci",[315,7141,7142],{},"サムネはCanva",[315,7144,7145],{},"ランディングはFigma → Webflow",[315,7147,7148],{},"投稿用文言はまた別のAI",[315,7150,7151],{},"リサーチはまた別のAI",[19,7153,7154,7155,7158],{},"これ、わかる人めちゃくちゃいると思うんですよね。「AIで時短！」と言いながら、実際には",[23,7156,7157],{},"ツール間のコピペとフォーマット直しに時間が溶ける","という状態。私自身、AI使い始めた頃は週20時間以上をクリエイティブ作業に使っていて、しかもクオリティは正直バラついていました。",[19,7160,7161,7162,7165],{},"ここを根本から変えたのが、Claudeの",[23,7163,7164],{},"Skills × MCP × Claude Design","の組み合わせです。",[205,7167],{},[32,7169,7171],{"id":7170},"つなぎ方の正体3点セットで何が変わるか","\"つなぎ方\"の正体：3点セットで何が変わるか",[19,7173,7174,7175,7178],{},"Skills・MCP・Claude Designはそれぞれ単体だと「便利な機能」止まりですが、組み合わせると",[23,7176,7177],{},"ひとつの作業フロー","になります。",[19,7180,7181],{},"ざっくり言うとこんな役割分担です。",[312,7183,7184,7190,7196],{},[315,7185,7186,7189],{},[23,7187,7188],{},"Skills","：私のブログの書き方や動画台本のフォーマットを覚えさせるところ",[315,7191,7192,7195],{},[23,7193,7194],{},"MCP","：Notion・Google Drive・YouTube・Figmaなど普段使うツールに繋ぐ\"USB-C\"",[315,7197,7198,7201],{},[23,7199,7200],{},"Claude Design","：プロトタイプ・スライド・LP・サムネのモックアップを会話で作るところ",[19,7203,7204,7205,7208],{},"私の場合、ZYL0ブログ用の「zyl0-blog-writer」というSkillを作って、毎回フォーマットを説明する必要をなくしました。これだけで",[23,7206,7207],{},"1記事あたり40〜50分の時短","です。さらにMCPで自分のNotionと繋いでおくと、「先週のリサーチメモから今週の動画台本作って」と一言投げるだけで、台本のドラフトが返ってきます。",[19,7210,7211,7212,7215],{},"正直、最初の設定（Skill作成 + MCP接続）に半日かかります。",[23,7213,7214],{},"ここで多くの人が脱落するんですが、ここを越えるとリターンが桁違い","です。",[205,7217],{},[32,7219,7221],{"id":7220},"ひとり制作のリアル15時間3時間の中身","ひとり制作のリアル：15時間→3時間の中身",[19,7223,7224],{},"具体的に、私の毎週金曜の動画制作フローを公開します。",[122,7226,7228],{"id":7227},"beforeaiを単体で使っていた頃","Before（AIを\"単体\"で使っていた頃）",[312,7230,7231,7234,7237,7240,7243,7246],{},[315,7232,7233],{},"月：企画ブレスト（ChatGPT）2時間",[315,7235,7236],{},"火：台本作成（ChatGPT）3時間",[315,7238,7239],{},"水：サムネ・LP案（Figma + Canva）4時間",[315,7241,7242],{},"木：撮影＋編集（CapCut）5時間",[315,7244,7245],{},"金：投稿文・公開（複数ツール往復）1時間",[315,7247,7248],{},[23,7249,7250],{},"合計：約15時間／週",[122,7252,7254],{"id":7253},"afterskills-mcp-claude-designをつないだ後","After（Skills + MCP + Claude Designをつないだ後）",[312,7256,7257,7260,7263,7266,7269,7272],{},[315,7258,7259],{},"月：Notionにアイデアメモ（5分）→Skills経由で企画案3つ生成（10分）",[315,7261,7262],{},"火：Claudeに「Skills通りの台本作って」（30分）→自分で編集（30分）",[315,7264,7265],{},"水：Claude Designでサムネ・LPプロトタイプ生成（45分）",[315,7267,7268],{},"木：撮影は変わらず（90分）。編集はAIキャプションで30分短縮",[315,7270,7271],{},"金：MCP経由でX・YouTube・ブログに同時投稿用テキスト生成（10分）",[315,7273,7274],{},[23,7275,7276],{},"合計：約3時間／週（撮影90分含む）",[19,7278,7279,7280,7283],{},"「動画台本→サムネ→LP→投稿文」が",[23,7281,7282],{},"ひとつの会話で繋がる","のが何より大きいです。コピペ地獄から解放されました。",[205,7285],{},[32,7287,7289],{"id":7288},"で結局なにから始めれば","「で、結局なにから始めれば？」",[19,7291,7292],{},"ここまで読んで「やってみたい」と思った人へ、最短ルートはこの3ステップです。",[552,7294,7295,7301,7307],{},[315,7296,7297,7300],{},[23,7298,7299],{},"Claude Pro（月20ドル）に契約する"," — Skills・Claude Designはここから使える",[315,7302,7303,7306],{},[23,7304,7305],{},"自分の\"いつもの作業\"を1つだけSkillにする"," — 私の場合は「ブログ記事のフォーマット」だった。やりたい作業を箇条書きにしてClaudeに渡し、SKILL.mdとして保存してもらうだけ",[315,7308,7309,7312],{},[23,7310,7311],{},"MCPサーバーを1つだけ繋ぐ"," — Notion、Google Drive、YouTubeのいずれか、自分が一番触るものを最初に",[19,7314,7315,7316,7319],{},"最初は完璧を狙わないでOKです。",[23,7317,7318],{},"ひとつのSkillとひとつのMCP接続","だけでも、作業時間は確実に減ります。私の場合は最初の1週間で「あ、これは戻れない」と確信しました。",[19,7321,7322,7323,4299],{},"クリエイターにとっての勝負は、ツールの種類ではなく**ツールの\"つなげ方\"**だと、この半年で本気で思うようになりました。AIに使われるのではなく、自分のワークフローに組み込んで使い倒す側に回れるかどうか。",[23,7324,7325],{},"今日Skillを1個作るだけで、来週の自分の時間がまるっと変わります",[19,7327,7328],{},"もし試したら、ぜひ感想を教えてください。私のXまでDMで送ってもらえると、めちゃくちゃ嬉しいです。",[205,7330],{},[19,7332,7333],{},[23,7334,7335],{},"次号の記事案",[312,7337,7338,7344,7350],{},[315,7339,7340,7343],{},[23,7341,7342],{},"案1：「私のSKILL.md全部見せます」"," — ZYL0ブログの実SKILL.mdをそのまま公開して、自作するときの叩き台として使えるようにする企画。",[315,7345,7346,7349],{},[23,7347,7348],{},"案2：「MCP接続でいちばん\"効いた\"5つのサーバー」"," — 私が実際に毎日使っているMCPサーバーを5つ厳選して、設定手順と効果を共有する。",[315,7351,7352,7355],{},[23,7353,7354],{},"案3：「Claude Designでサムネを30秒で作る最短プロンプト」"," — 動画クリエイター向けに、サムネ専用の\"型プロンプト\"だけを掘り下げる短編。",[205,7357],{},[10,7359,7360,7364,7375,7377,7381,7387,7407,7414,7424,7426,7430,7433,7436,7453,7460,7467,7469,7473,7476,7480,7502,7506,7528,7535,7537,7541,7544,7567,7574,7588,7591,7593,7598],{"lang":809},[14,7361,7363],{"id":7362},"how-claude-cut-my-video-web-workflow-to-15-the-time","How Claude Cut My Video + Web Workflow to 1/5 the Time",[19,7365,7366,7367,7370,7371,7374],{},"I run a VC day job and a one-person YouTube/X/blog operation on the side, and since Claude Design dropped in April 2026, ",[23,7368,7369],{},"the 15 hours I used to spend each week on a single video plus a landing page collapsed to three",". Honestly, my first reaction was \"another AI hype cycle.\" Then I started chaining ",[23,7372,7373],{},"Skills, MCP, and Claude Design"," as one workflow — and the picture changed completely. This is the actual chain I run, with the time numbers attached.",[205,7376],{},[32,7378,7380],{"id":7379},"the-more-tools-less-time-trap","The \"More Tools, Less Time\" Trap",[19,7382,7383,7384,849],{},"The most frustrating thing of the last two years has been that ",[23,7385,7386],{},"adding more tools didn't free up any time",[312,7388,7389,7392,7395,7398,7401,7404],{},[315,7390,7391],{},"Scripts in ChatGPT",[315,7393,7394],{},"Editing in CapCut or DaVinci",[315,7396,7397],{},"Thumbnails in Canva",[315,7399,7400],{},"Landing pages in Figma → Webflow",[315,7402,7403],{},"Social copy in yet another AI",[315,7405,7406],{},"Research in yet another AI",[19,7408,7409,7410,7413],{},"If you're a one-person creator, you know this pain. \"AI saves time!\" — except your time evaporates into ",[23,7411,7412],{},"copy-paste and format-fixing between tools",". Back when I was just bolting AIs onto my workflow, I was still spending 20+ hours a week on creative tasks, and quality was inconsistent.",[19,7415,7416,7417,7419,7420,7423],{},"The thing that finally broke that loop was Claude's ",[23,7418,7164],{}," combination — used as a ",[802,7421,7422],{},"workflow",", not as three features.",[205,7425],{},[32,7427,7429],{"id":7428},"what-chaining-them-actually-means","What \"Chaining Them\" Actually Means",[19,7431,7432],{},"Each piece on its own is just another feature. Together they become a single operating loop.",[19,7434,7435],{},"Roughly:",[312,7437,7438,7443,7448],{},[315,7439,7440,7442],{},[23,7441,7188],{}," — where you teach Claude your blog format, video script template, brand voice",[315,7444,7445,7447],{},[23,7446,7194],{}," — the \"USB-C\" connector to Notion, Google Drive, YouTube, Figma, etc.",[315,7449,7450,7452],{},[23,7451,7200],{}," — generates prototypes, slides, landing pages, thumbnail mockups via conversation",[19,7454,7455,7456,7459],{},"For my blog (ZYL0), I built a Skill called \"zyl0-blog-writer\" that knows my structure, voice, and frontmatter. That alone saves ",[23,7457,7458],{},"40–50 minutes per article"," because I no longer re-explain rules. Then I pointed MCP at my Notion, so \"turn last week's research notes into this week's video script\" gives me a usable draft immediately.",[19,7461,7462,7463,7466],{},"To be honest, the first setup (building the Skill, wiring the MCP servers) takes half a day. ",[23,7464,7465],{},"Most people quit there."," But on the other side of that wall, the leverage is non-linear.",[205,7468],{},[32,7470,7472],{"id":7471},"the-15-3-hour-breakdown","The 15 → 3 Hour Breakdown",[19,7474,7475],{},"Here's my actual Friday-shipping workflow before and after.",[122,7477,7479],{"id":7478},"before-using-ais-as-separate-tools","Before (using AIs as separate tools)",[312,7481,7482,7485,7488,7491,7494,7497],{},[315,7483,7484],{},"Mon: brainstorm in ChatGPT (2 hr)",[315,7486,7487],{},"Tue: write script in ChatGPT (3 hr)",[315,7489,7490],{},"Wed: thumbnail + LP in Figma + Canva (4 hr)",[315,7492,7493],{},"Thu: shoot + edit in CapCut (5 hr)",[315,7495,7496],{},"Fri: write social copy across multiple tools (1 hr)",[315,7498,7499],{},[23,7500,7501],{},"Total: ~15 hours/week",[122,7503,7505],{"id":7504},"after-chained-skills-mcp-claude-design","After (chained Skills + MCP + Claude Design)",[312,7507,7508,7511,7514,7517,7520,7523],{},[315,7509,7510],{},"Mon: drop notes in Notion (5 min) → 3 episode angles via my Skill (10 min)",[315,7512,7513],{},"Tue: \"Write the script using the Skill\" (30 min) → human edit (30 min)",[315,7515,7516],{},"Wed: thumbnail + LP prototype via Claude Design (45 min)",[315,7518,7519],{},"Thu: shoot stays the same (90 min); editing saves ~30 min via AI captions",[315,7521,7522],{},"Fri: cross-platform social copy through MCP (10 min)",[315,7524,7525],{},[23,7526,7527],{},"Total: ~3 hours/week (including 90 min shoot)",[19,7529,7530,7531,7534],{},"The single biggest win is that ",[23,7532,7533],{},"script → thumbnail → LP → social"," all live in one conversation. The copy-paste hellscape goes away.",[205,7536],{},[32,7538,7540],{"id":7539},"ok-where-do-i-start","\"OK, Where Do I Start?\"",[19,7542,7543],{},"If you want to try this yourself, the shortest path is three steps.",[552,7545,7546,7552,7561],{},[315,7547,7548,7551],{},[23,7549,7550],{},"Subscribe to Claude Pro ($20/month)"," — Skills and Claude Design unlock here.",[315,7553,7554,7557,7558,849],{},[23,7555,7556],{},"Convert one of your repeating tasks into a single Skill."," For me it was \"blog post format.\" Just bullet-point what you want and ask Claude to save it as ",[141,7559,7560],{},"SKILL.md",[315,7562,7563,7566],{},[23,7564,7565],{},"Connect one MCP server"," — Notion, Google Drive, or YouTube — whichever you touch most.",[19,7568,7569,7570,7573],{},"Don't aim for perfect on day one. ",[23,7571,7572],{},"Even one Skill + one MCP connection visibly reduces your time."," In my first week I knew there was no going back.",[19,7575,7576,7577,7580,7581,7584,7585],{},"The competitive edge for a creator in 2026 isn't which tools you own — it's ",[23,7578,7579],{},"how you chain them",". Whether you stay a tool user, or move up to being someone who ",[802,7582,7583],{},"operates the chain",", is the question. ",[23,7586,7587],{},"Building one Skill today flips next week of your life.",[19,7589,7590],{},"If you try this, I'd genuinely love to hear how it goes. DM me on X.",[205,7592],{},[19,7594,7595],{},[23,7596,7597],{},"Next Issue Ideas",[312,7599,7600,7606,7612],{},[315,7601,7602,7605],{},[23,7603,7604],{},"Idea 1: \"My actual SKILL.md, fully public\""," — Publish the real SKILL.md I use for ZYL0 so other creators can copy it as a starting template.",[315,7607,7608,7611],{},[23,7609,7610],{},"Idea 2: \"5 MCP servers that actually moved the needle\""," — A short, opinionated list of the MCP servers I run daily, with setup steps and the time-saving impact of each.",[315,7613,7614,7617],{},[23,7615,7616],{},"Idea 3: \"The 30-second thumbnail prompt\""," — A focused micro-post for video creators, just on the prompt template that produces a usable thumbnail in Claude Design in 30 seconds.",{"title":143,"searchDepth":1605,"depth":1605,"links":7619},[7620,7621,7622,7626,7627,7628,7629,7633],{"id":7123,"depth":1605,"text":7124},{"id":7170,"depth":1605,"text":7171},{"id":7220,"depth":1605,"text":7221,"children":7623},[7624,7625],{"id":7227,"depth":1610,"text":7228},{"id":7253,"depth":1610,"text":7254},{"id":7288,"depth":1605,"text":7289},{"id":7379,"depth":1605,"text":7380},{"id":7428,"depth":1605,"text":7429},{"id":7471,"depth":1605,"text":7472,"children":7630},[7631,7632],{"id":7478,"depth":1610,"text":7479},{"id":7504,"depth":1610,"text":7505},{"id":7539,"depth":1605,"text":7540},"Skills・MCP・Claude Designを組み合わせて、企画→デザイン→動画台本→公開までを一本の流れで回したら、制作時間が驚くほど縮んだ。クリエイターが今すぐ試せる\"つなぎ方\"を実体験で共有する。",{"date":7636,"image":7637,"alt":7638,"tags":7639,"tagsEn":7645,"published":1666},"3rd May 2026","/blogs-img/blog-claude-skills-mcp-practical-guide.png","Claude Skills・MCP・Claude Design・動画制作のクリエイター向け実践フロー",[7640,7641,7642,7643,7644],"クリエイター","AI活用","YouTube","動画編集","やってみた",[7646,7647,7642,7648,7649],"Creators","AI Tools","Video Editing","I Tried It","/blogs/12-claude-skills-mcp-practical-guide",{"title":7103,"description":7634},"blogs/12-claude-skills-mcp-practical-guide","hbsmWfU8ZJav4EHA7XXYYbW75LR64DJhzdGONzp2DzA",{"id":7655,"title":7656,"body":7657,"description":9997,"extension":1651,"meta":9998,"navigation":1666,"ogImage":9999,"path":10007,"seo":10008,"stem":10009,"__hash__":10010},"content/blogs/13-claude-skills-mcp-web-video-engineer-guide.md","Claude Skills・MCP・ウェブ制作・動画生成を実務で使い倒す：エンジニア向け実践ガイド2026 | Claude Skills, MCP, Website Generation, and Programmatic Video in Production: A Practical Engineer's Guide 2026",{"type":7,"value":7658,"toc":9943},[7659,8846,8848,9939],[10,7660,7661,7665,7673,7684,7686,7690,7744,7746,7750,7753,7762,7769,7841,7844,7848,7851,7854,7857,7961,7963,7967,7970,7973,7976,8017,8021,8024,8177,8181,8290,8293,8335,8342,8344,8348,8351,8354,8358,8361,8438,8441,8445,8448,8479,8482,8484,8488,8495,8498,8547,8550,8553,8556,8562,8565,8609,8612,8680,8683,8685,8688,8691,8695,8708,8714,8720,8724,8730,8740,8743,8749,8755,8759,8765,8771,8777,8779,8782,8843],{"lang":12},[14,7662,7664],{"id":7663},"claude-skillsmcpウェブ制作動画生成を実務で使い倒すエンジニア向け実践ガイド2026","Claude Skills・MCP・ウェブ制作・動画生成を実務で使い倒す：エンジニア向け実践ガイド2026",[792,7666,7667],{},[19,7668,7669,7672],{},[23,7670,7671],{},"バージョン注記",": 本記事は2026年5月時点の情報を基に執筆している。mcp Python SDK 1.27.0、MCP仕様 2025-11-25リビジョン、Remotion最新版を参照している。MCPの仕様更新は活発なため、公式ドキュメント（modelcontextprotocol.io）で最新情報を確認することを推奨する。",[19,7674,7675,7676,7679,7680,7683],{},"私は半導体プロセス開発からCVCに転身し、AWS（IoT Core・Lambda・DynamoDB・API Gateway）でゼロから河川洪水検知システムを6ヶ月で立ち上げた経験があります。",[23,7677,7678],{},"「動かしてみないと本当の落とし穴は見えない」","——これがエンジニアリング現場で身に染みた感覚で、Skills・MCP・Remotionも同じでした。今回はドキュメントを読むだけでは見えない",[23,7681,7682],{},"ハマりどころと最短実装パス","を、実際に手を動かした上で書きます。",[205,7685],{},[32,7687,7689],{"id":7688},"tldrこの記事でわかること","TL;DR：この記事でわかること",[54,7691,7692,7702],{},[57,7693,7694],{},[60,7695,7696,7699],{},[63,7697,7698],{},"テーマ",[63,7700,7701],{},"学べること",[70,7703,7704,7712,7720,7728,7736],{},[60,7705,7706,7709],{},[75,7707,7708],{},"Claude Skills",[75,7710,7711],{},"SKILL.mdの仕組み・MCP との役割分担・実務での使い方",[60,7713,7714,7717],{},[75,7715,7716],{},"MCP サーバー構築",[75,7718,7719],{},"Python FastMCP で50行以下のサーバーを作る最短手順",[60,7721,7722,7725],{},[75,7723,7724],{},"ウェブサイト生成",[75,7726,7727],{},"Claude Code + Next.js / Nuxt 4 でのサイト生成ワークフロー",[60,7729,7730,7733],{},[75,7731,7732],{},"Remotion 動画制作",[75,7734,7735],{},"コードで動画を書いてClaude Codeでレンダリングする実践手順",[60,7737,7738,7741],{},[75,7739,7740],{},"ハマりポイント",[75,7742,7743],{},"それぞれの罠と回避策",[205,7745],{},[32,7747,7749],{"id":7748},"skills-と-mcp-は何が違うのか役割を正確に理解する","Skills と MCP は何が違うのか——役割を正確に理解する",[19,7751,7752],{},"Claude CodeやClaude.aiを使い込んでいくと、必ずぶつかる疑問がある。「Skills と MCP って結局どう使い分けるの？」だ。ドキュメントを読んでもピンとこない場合、次のメンタルモデルが効く。",[19,7754,7755,7758,7759,4299],{},[23,7756,7757],{},"Skills = エージェントの「手順書」","、",[23,7760,7761],{},"MCP = エージェントの「外部接続線」",[19,7763,7764,7765,7768],{},"Skillsは",[141,7766,7767],{},".claude/skills/","以下に置いたSKILL.mdファイルだ。Claudeがタスクの文脈を判断し、関連するSkillを自動でロードして指示に従う。コードを書かず、マークダウンを書くだけで動く。一方MCPはJSON-RPC 2.0で動くクライアント・サーバープロトコルで、ClaudeをGitHub・データベース・社内APIなどの外部システムに繋ぐ。",[54,7770,7771,7782],{},[57,7772,7773],{},[60,7774,7775,7778,7780],{},[63,7776,7777],{},"比較軸",[63,7779,7188],{},[63,7781,7194],{},[70,7783,7784,7795,7806,7817,7828],{},[60,7785,7786,7789,7792],{},[75,7787,7788],{},"実装言語",[75,7790,7791],{},"Markdown（SKILL.md）",[75,7793,7794],{},"Python / TypeScript / Go など",[60,7796,7797,7800,7803],{},[75,7798,7799],{},"コンテキスト消費",[75,7801,7802],{},"30〜50トークン（必要時のみロード）",[75,7804,7805],{},"50,000トークン超になることも",[60,7807,7808,7811,7814],{},[75,7809,7810],{},"用途",[75,7812,7813],{},"手順・ベストプラクティスの埋め込み",[75,7815,7816],{},"外部システムとのリアルタイム連携",[60,7818,7819,7822,7825],{},[75,7820,7821],{},"横断性",[75,7823,7824],{},"Claude Code / Cursor / Gemini CLI で共通動作",[75,7826,7827],{},"MCP対応クライアント全般で動作",[60,7829,7830,7833,7838],{},[75,7831,7832],{},"設定方法",[75,7834,7835,7837],{},[141,7836,7767],{}," にファイルを置くだけ",[75,7839,7840],{},"サーバーを立ててclaude_desktop_config.jsonに登録",[19,7842,7843],{},"重要な点は「SkillはMCPツールを呼べる」という非対称性だ。Skillで手順を定義し、MCPで外部ツールを接続する——両方組み合わせることで初めてフルオートのエージェントができる。",[122,7845,7847],{"id":7846},"skills-の実務パターン","Skills の実務パターン",[19,7849,7850],{},"2026年3月時点のSkillsエコシステムは、公式Anthropicスキル・検証済みサードパーティ・コミュニティスキルの三層構造になっている。frontend-designスキルが27万インストール超、Remotionスキルが週11.7万インストール、gws（Google Workspace）が公開3日で4,900 GitHubスターを記録した。",[19,7852,7853],{},"実務でSkillが威力を発揮するのは「同じ手順を何度も説明するのが嫌な場面」全般だ。独自DBスキーマを理解するSQLアシスタントSkill・社内コーディング規約を埋め込んだレビューSkill・独自データ形式の変換Skillなどを一度書けば、以降はプロンプト一行で呼び出せる。",[19,7855,7856],{},"SKILL.mdを自分で書く場合の最小構成は以下だ：",[134,7858,7862],{"className":7859,"code":7860,"language":7861,"meta":143,"style":143},"language-markdown shiki shiki-themes dracula","---\nname: my-skill\ndescription: >\n  説明。Claudeがこれを見てスキルをロードするかを判断する。\n  トリガー条件を具体的に書くこと。\n---\n\n# My Skill\n\n## 手順\n\n1. まず〇〇を確認する\n2. 次に△△を実行する\n\n## 落とし穴\n\n- □□の場合は××してはいけない\n","markdown",[141,7863,7864,7872,7877,7882,7888,7894,7899,7905,7911,7916,7922,7927,7933,7939,7944,7950,7955],{"__ignoreMap":143},[7865,7866,7869],"span",{"class":7867,"line":7868},"line",1,[7865,7870,7871],{},"---\n",[7865,7873,7874],{"class":7867,"line":1605},[7865,7875,7876],{},"name: my-skill\n",[7865,7878,7879],{"class":7867,"line":1610},[7865,7880,7881],{},"description: >\n",[7865,7883,7885],{"class":7867,"line":7884},4,[7865,7886,7887],{},"  説明。Claudeがこれを見てスキルをロードするかを判断する。\n",[7865,7889,7891],{"class":7867,"line":7890},5,[7865,7892,7893],{},"  トリガー条件を具体的に書くこと。\n",[7865,7895,7897],{"class":7867,"line":7896},6,[7865,7898,7871],{},[7865,7900,7902],{"class":7867,"line":7901},7,[7865,7903,7904],{"emptyLinePlaceholder":1666},"\n",[7865,7906,7908],{"class":7867,"line":7907},8,[7865,7909,7910],{},"# My Skill\n",[7865,7912,7914],{"class":7867,"line":7913},9,[7865,7915,7904],{"emptyLinePlaceholder":1666},[7865,7917,7919],{"class":7867,"line":7918},10,[7865,7920,7921],{},"## 手順\n",[7865,7923,7925],{"class":7867,"line":7924},11,[7865,7926,7904],{"emptyLinePlaceholder":1666},[7865,7928,7930],{"class":7867,"line":7929},12,[7865,7931,7932],{},"1. まず〇〇を確認する\n",[7865,7934,7936],{"class":7867,"line":7935},13,[7865,7937,7938],{},"2. 次に△△を実行する\n",[7865,7940,7942],{"class":7867,"line":7941},14,[7865,7943,7904],{"emptyLinePlaceholder":1666},[7865,7945,7947],{"class":7867,"line":7946},15,[7865,7948,7949],{},"## 落とし穴\n",[7865,7951,7953],{"class":7867,"line":7952},16,[7865,7954,7904],{"emptyLinePlaceholder":1666},[7865,7956,7958],{"class":7867,"line":7957},17,[7865,7959,7960],{},"- □□の場合は××してはいけない\n",[205,7962],{},[32,7964,7966],{"id":7965},"mcpサーバーを50行で作るpython-fastmcp-最短実装","MCPサーバーを50行で作る——Python FastMCP 最短実装",[19,7968,7969],{},"MCPの普及速度は前例がない。2024年11月の公開時に月200万SDK DLだったものが、OpenAI採用（2025年3月）・Microsoft統合（7月）・AWS対応（11月）を経て、2026年3月には9,700万DL/月、公開サーバー1万超に達した。ReactはこのDL数に3年かかったが、MCPは16ヶ月でやってのけた。",[19,7971,7972],{},"MCPが解決する問題はシンプルだ。社内CRM・請求システム・機能フラグサービスをAIエージェントから叩くには、従来クライアントAIごとに別の統合コードを書く必要があった。MCP対応サーバーを一つ作れば、Claude・ChatGPT・Cursor・Gemini全クライアントで動く。",[122,7974,7975],{"id":7975},"セットアップ",[134,7977,7981],{"className":7978,"code":7979,"language":7980,"meta":143,"style":143},"language-bash shiki shiki-themes dracula","# uv 推奨（pip でも可）\nuv add mcp\n# または\npip install mcp\n","bash",[141,7982,7983,7989,8002,8007],{"__ignoreMap":143},[7865,7984,7985],{"class":7867,"line":7868},[7865,7986,7988],{"class":7987},"shSDL","# uv 推奨（pip でも可）\n",[7865,7990,7991,7995,7999],{"class":7867,"line":1605},[7865,7992,7994],{"class":7993},"sAOxA","uv",[7865,7996,7998],{"class":7997},"s-mGx"," add",[7865,8000,8001],{"class":7997}," mcp\n",[7865,8003,8004],{"class":7867,"line":1610},[7865,8005,8006],{"class":7987},"# または\n",[7865,8008,8009,8012,8015],{"class":7867,"line":7884},[7865,8010,8011],{"class":7993},"pip",[7865,8013,8014],{"class":7997}," install",[7865,8016,8001],{"class":7997},[122,8018,8020],{"id":8019},"最小構成サーバーpython-fastmcp","最小構成サーバー（Python FastMCP）",[19,8022,8023],{},"以下は実際に動くMCPサーバーの最小実装だ。Tools・Resources・Promptsの三要素を全て含む：",[134,8025,8029],{"className":8026,"code":8027,"language":8028,"meta":143,"style":143},"language-python shiki shiki-themes dracula","from mcp.server.fastmcp import FastMCP\n\nmcp = FastMCP(\"my-service\", json_response=True)\n\n# Tool: 副作用のある操作（POSTに相当）\n@mcp.tool()\ndef check_stock(sku: str) -> dict:\n    \"\"\"指定SKUの在庫数を返す\"\"\"\n    # 実際はDBクエリ等に置き換える\n    return {\"sku\": sku, \"quantity\": 42, \"warehouse\": \"tokyo\"}\n\n# Resource: 読み取り専用データ（GETに相当）\n@mcp.resource(\"products://catalog\")\ndef get_catalog() -> str:\n    \"\"\"商品カタログの一覧を返す\"\"\"\n    return \"SKU-001: Widget A\\nSKU-002: Widget B\"\n\n# Prompt: 再利用可能なプロンプトテンプレート\n@mcp.prompt()\ndef stock_check_prompt(sku: str) -> str:\n    \"\"\"在庫確認のためのプロンプトを生成\"\"\"\n    return f\"SKU {sku} の在庫を確認し、少ない場合は補充を提案してください。\"\n\nif __name__ == \"__main__\":\n    # ローカル開発: stdio\n    # mcp.run()\n    # リモート/本番: Streamable HTTP\n    mcp.run(transport=\"streamable-http\", host=\"0.0.0.0\", port=3050)\n","python",[141,8030,8031,8036,8040,8045,8049,8054,8059,8064,8069,8074,8079,8083,8088,8093,8098,8103,8108,8112,8118,8124,8130,8136,8142,8147,8153,8159,8165,8171],{"__ignoreMap":143},[7865,8032,8033],{"class":7867,"line":7868},[7865,8034,8035],{},"from mcp.server.fastmcp import FastMCP\n",[7865,8037,8038],{"class":7867,"line":1605},[7865,8039,7904],{"emptyLinePlaceholder":1666},[7865,8041,8042],{"class":7867,"line":1610},[7865,8043,8044],{},"mcp = FastMCP(\"my-service\", json_response=True)\n",[7865,8046,8047],{"class":7867,"line":7884},[7865,8048,7904],{"emptyLinePlaceholder":1666},[7865,8050,8051],{"class":7867,"line":7890},[7865,8052,8053],{},"# Tool: 副作用のある操作（POSTに相当）\n",[7865,8055,8056],{"class":7867,"line":7896},[7865,8057,8058],{},"@mcp.tool()\n",[7865,8060,8061],{"class":7867,"line":7901},[7865,8062,8063],{},"def check_stock(sku: str) -> dict:\n",[7865,8065,8066],{"class":7867,"line":7907},[7865,8067,8068],{},"    \"\"\"指定SKUの在庫数を返す\"\"\"\n",[7865,8070,8071],{"class":7867,"line":7913},[7865,8072,8073],{},"    # 実際はDBクエリ等に置き換える\n",[7865,8075,8076],{"class":7867,"line":7918},[7865,8077,8078],{},"    return {\"sku\": sku, \"quantity\": 42, \"warehouse\": \"tokyo\"}\n",[7865,8080,8081],{"class":7867,"line":7924},[7865,8082,7904],{"emptyLinePlaceholder":1666},[7865,8084,8085],{"class":7867,"line":7929},[7865,8086,8087],{},"# Resource: 読み取り専用データ（GETに相当）\n",[7865,8089,8090],{"class":7867,"line":7935},[7865,8091,8092],{},"@mcp.resource(\"products://catalog\")\n",[7865,8094,8095],{"class":7867,"line":7941},[7865,8096,8097],{},"def get_catalog() -> str:\n",[7865,8099,8100],{"class":7867,"line":7946},[7865,8101,8102],{},"    \"\"\"商品カタログの一覧を返す\"\"\"\n",[7865,8104,8105],{"class":7867,"line":7952},[7865,8106,8107],{},"    return \"SKU-001: Widget A\\nSKU-002: Widget B\"\n",[7865,8109,8110],{"class":7867,"line":7957},[7865,8111,7904],{"emptyLinePlaceholder":1666},[7865,8113,8115],{"class":7867,"line":8114},18,[7865,8116,8117],{},"# Prompt: 再利用可能なプロンプトテンプレート\n",[7865,8119,8121],{"class":7867,"line":8120},19,[7865,8122,8123],{},"@mcp.prompt()\n",[7865,8125,8127],{"class":7867,"line":8126},20,[7865,8128,8129],{},"def stock_check_prompt(sku: str) -> str:\n",[7865,8131,8133],{"class":7867,"line":8132},21,[7865,8134,8135],{},"    \"\"\"在庫確認のためのプロンプトを生成\"\"\"\n",[7865,8137,8139],{"class":7867,"line":8138},22,[7865,8140,8141],{},"    return f\"SKU {sku} の在庫を確認し、少ない場合は補充を提案してください。\"\n",[7865,8143,8145],{"class":7867,"line":8144},23,[7865,8146,7904],{"emptyLinePlaceholder":1666},[7865,8148,8150],{"class":7867,"line":8149},24,[7865,8151,8152],{},"if __name__ == \"__main__\":\n",[7865,8154,8156],{"class":7867,"line":8155},25,[7865,8157,8158],{},"    # ローカル開発: stdio\n",[7865,8160,8162],{"class":7867,"line":8161},26,[7865,8163,8164],{},"    # mcp.run()\n",[7865,8166,8168],{"class":7867,"line":8167},27,[7865,8169,8170],{},"    # リモート/本番: Streamable HTTP\n",[7865,8172,8174],{"class":7867,"line":8173},28,[7865,8175,8176],{},"    mcp.run(transport=\"streamable-http\", host=\"0.0.0.0\", port=3050)\n",[122,8178,8180],{"id":8179},"claude-desktop-への登録","Claude Desktop への登録",[134,8182,8186],{"className":8183,"code":8184,"language":8185,"meta":143,"style":143},"language-json shiki shiki-themes dracula","{\n  \"mcpServers\": {\n    \"my-service\": {\n      \"command\": \"python\",\n      \"args\": [\"/path/to/server.py\"]\n    }\n  }\n}\n","json",[141,8187,8188,8194,8214,8228,8251,8275,8280,8285],{"__ignoreMap":143},[7865,8189,8190],{"class":7867,"line":7868},[7865,8191,8193],{"class":8192},"sCdxs","{\n",[7865,8195,8196,8200,8204,8207,8211],{"class":7867,"line":1605},[7865,8197,8199],{"class":8198},"sY8FZ","  \"",[7865,8201,8203],{"class":8202},"sLL85","mcpServers",[7865,8205,8206],{"class":8198},"\"",[7865,8208,8210],{"class":8209},"s0Tla",":",[7865,8212,8213],{"class":8192}," {\n",[7865,8215,8216,8219,8222,8224,8226],{"class":7867,"line":1610},[7865,8217,8218],{"class":8198},"    \"",[7865,8220,8221],{"class":8202},"my-service",[7865,8223,8206],{"class":8198},[7865,8225,8210],{"class":8209},[7865,8227,8213],{"class":8192},[7865,8229,8230,8233,8236,8238,8240,8244,8246,8248],{"class":7867,"line":7884},[7865,8231,8232],{"class":8198},"      \"",[7865,8234,8235],{"class":8202},"command",[7865,8237,8206],{"class":8198},[7865,8239,8210],{"class":8209},[7865,8241,8243],{"class":8242},"seVfx"," \"",[7865,8245,8028],{"class":7997},[7865,8247,8206],{"class":8242},[7865,8249,8250],{"class":8192},",\n",[7865,8252,8253,8255,8258,8260,8262,8265,8267,8270,8272],{"class":7867,"line":7890},[7865,8254,8232],{"class":8198},[7865,8256,8257],{"class":8202},"args",[7865,8259,8206],{"class":8198},[7865,8261,8210],{"class":8209},[7865,8263,8264],{"class":8192}," [",[7865,8266,8206],{"class":8242},[7865,8268,8269],{"class":7997},"/path/to/server.py",[7865,8271,8206],{"class":8242},[7865,8273,8274],{"class":8192},"]\n",[7865,8276,8277],{"class":7867,"line":7896},[7865,8278,8279],{"class":8192},"    }\n",[7865,8281,8282],{"class":7867,"line":7901},[7865,8283,8284],{"class":8192},"  }\n",[7865,8286,8287],{"class":7867,"line":7907},[7865,8288,8289],{"class":8192},"}\n",[122,8291,8292],{"id":8292},"トランスポートの選び方",[54,8294,8295,8307],{},[57,8296,8297],{},[60,8298,8299,8302,8304],{},[63,8300,8301],{},"トランスポート",[63,8303,7810],{},[63,8305,8306],{},"注意",[70,8308,8309,8322],{},[60,8310,8311,8316,8319],{},[75,8312,8313],{},[141,8314,8315],{},"stdio",[75,8317,8318],{},"ローカル・Claude Desktop連携・開発",[75,8320,8321],{},"ネットワーク設定不要",[60,8323,8324,8329,8332],{},[75,8325,8326],{},[141,8327,8328],{},"streamable-http",[75,8330,8331],{},"リモート・チーム共有・クラウドデプロイ",[75,8333,8334],{},"2025年3月仕様で標準化。SSE は非推奨",[19,8336,8337,8338,8341],{},"TypeScriptで書く場合は",[141,8339,8340],{},"@modelcontextprotocol/sdk","とZodを使う。Python FastMCPと同じ三要素（Tool / Resource / Prompt）をサポートし、型安全性が高い。どちらの言語でも「ワイヤプロトコルはJSON-RPC 2.0で共通」なので、PythonサーバーにTypeScriptクライアントを繋ぐことも普通に動く。",[205,8343],{},[32,8345,8347],{"id":8346},"claude-code-でウェブサイトを生成するclaudemdとskillの組み合わせ","Claude Code でウェブサイトを生成する——CLAUDE.mdとSkillの組み合わせ",[19,8349,8350],{},"2026年初頭時点でClaude Codeは全GitHubコミットの4%を生成しており、VS Codeへの日次インストールは2,900万に達した。「バイブコーディング」と呼ばれるこのワークフローは、エンジニア経験ゼロの人でもNext.jsアプリを作れるレベルに成熟している。",[19,8352,8353],{},"ただしClaudeに「サイト作って」と言うだけでは、Interフォント・パープルグラデーション・白背景というお約束の「AI生成デザイン」が出てくる。これを回避するのがCLAUDE.mdとSkillの組み合わせだ。",[122,8355,8357],{"id":8356},"claudemdの位置づけ","CLAUDE.mdの位置づけ",[19,8359,8360],{},"プロジェクトルートに置くCLAUDE.mdはClaude Codeがセッション開始時に自動読み込みする設定ファイルだ。ここに「プロジェクトのコンテキスト」を書く：",[134,8362,8364],{"className":7859,"code":8363,"language":7861,"meta":143,"style":143},"# Project: MyApp\n\n## スタック\n- Nuxt 4 (Nuxt Content 3, Vue 3 Composition API)\n- Tailwind CSS v4, shadcn/ui\n- Vercel デプロイ\n\n## ルール\n- コミット前に必ずビルドを通す\n- TypeScript strict モード必須\n- MDX コンポーネントは使わない（Nuxt Content の CommonMark のみ）\n\n## 禁止事項\n- console.log を残したままコミットしない\n- any 型の使用\n",[141,8365,8366,8371,8375,8380,8385,8390,8395,8399,8404,8409,8414,8419,8423,8428,8433],{"__ignoreMap":143},[7865,8367,8368],{"class":7867,"line":7868},[7865,8369,8370],{},"# Project: MyApp\n",[7865,8372,8373],{"class":7867,"line":1605},[7865,8374,7904],{"emptyLinePlaceholder":1666},[7865,8376,8377],{"class":7867,"line":1610},[7865,8378,8379],{},"## スタック\n",[7865,8381,8382],{"class":7867,"line":7884},[7865,8383,8384],{},"- Nuxt 4 (Nuxt Content 3, Vue 3 Composition API)\n",[7865,8386,8387],{"class":7867,"line":7890},[7865,8388,8389],{},"- Tailwind CSS v4, shadcn/ui\n",[7865,8391,8392],{"class":7867,"line":7896},[7865,8393,8394],{},"- Vercel デプロイ\n",[7865,8396,8397],{"class":7867,"line":7901},[7865,8398,7904],{"emptyLinePlaceholder":1666},[7865,8400,8401],{"class":7867,"line":7907},[7865,8402,8403],{},"## ルール\n",[7865,8405,8406],{"class":7867,"line":7913},[7865,8407,8408],{},"- コミット前に必ずビルドを通す\n",[7865,8410,8411],{"class":7867,"line":7918},[7865,8412,8413],{},"- TypeScript strict モード必須\n",[7865,8415,8416],{"class":7867,"line":7924},[7865,8417,8418],{},"- MDX コンポーネントは使わない（Nuxt Content の CommonMark のみ）\n",[7865,8420,8421],{"class":7867,"line":7929},[7865,8422,7904],{"emptyLinePlaceholder":1666},[7865,8424,8425],{"class":7867,"line":7935},[7865,8426,8427],{},"## 禁止事項\n",[7865,8429,8430],{"class":7867,"line":7941},[7865,8431,8432],{},"- console.log を残したままコミットしない\n",[7865,8434,8435],{"class":7867,"line":7946},[7865,8436,8437],{},"- any 型の使用\n",[19,8439,8440],{},"CLAUDE.mdとSkillの役割分担は明確だ。CLAUDE.mdは「このプロジェクトの文脈」、Skillは「このタスクのやり方」。両方あることで、プロジェクトを知りつつ特定作業を正しくこなすエージェントができる。",[122,8442,8444],{"id":8443},"nextjs-nuxt-でのサイト生成ワークフロー","Next.js / Nuxt でのサイト生成ワークフロー",[19,8446,8447],{},"実際に動くフローは以下になる：",[552,8449,8450,8456,8465,8473],{},[315,8451,8452,8455],{},[23,8453,8454],{},"CLAUDE.mdを作成",": スタック・命名規則・禁止事項を記述",[315,8457,8458,8461,8462],{},[23,8459,8460],{},"公式または自作Skillを追加",": ",[141,8463,8464],{},"npx skills add frontend-design",[315,8466,8467,8461,8470],{},[23,8468,8469],{},"プランモードで要件確認",[141,8471,8472],{},"claude --plan \"〇〇なランディングページを作りたい\"",[315,8474,8475,8478],{},[23,8476,8477],{},"イテレーティブに実装",": 一度に全部頼まず、セクション単位で進める",[19,8480,8481],{},"重要なのはステップ3のプランモードだ。いきなり実装させると「間違った問題を解くコード」が生成されることがある。まずClaudeに既存のコードベースを探索させ、変更箇所・影響範囲・エッジケースを特定したプランを出させる。プランを確認してからイエスと言うだけで、実装ミスの大半を防げる。",[205,8483],{},[32,8485,8487],{"id":8486},"claude-code-remotion-で動画をコードで書く","Claude Code + Remotion で動画をコードで書く",[19,8489,8490,8491,8494],{},"「動画をコードで管理する」という発想は、フロントエンドエンジニアには自然に受け入れられる。Remotionはまさにそれを実現するReactベースの動画フレームワークで、",[141,8492,8493],{},"useCurrentFrame()","でフレーム番号を取得しアニメーションをコードで記述する。Claude Codeとの組み合わせは「プロンプトで動画を書く」ワークフローを実現する。",[122,8496,7975],{"id":8497},"セットアップ-1",[134,8499,8501],{"className":7978,"code":8500,"language":7980,"meta":143,"style":143},"# Remotionプロジェクトを作成\nnpx create-video@latest my-video-project\ncd my-video-project\n\n# Agent Skillsをインストール（Remotionチームが提供するベストプラクティス集）\nnpx skills add remotion-dev/skills\n",[141,8502,8503,8508,8519,8526,8530,8535],{"__ignoreMap":143},[7865,8504,8505],{"class":7867,"line":7868},[7865,8506,8507],{"class":7987},"# Remotionプロジェクトを作成\n",[7865,8509,8510,8513,8516],{"class":7867,"line":1605},[7865,8511,8512],{"class":7993},"npx",[7865,8514,8515],{"class":7997}," create-video@latest",[7865,8517,8518],{"class":7997}," my-video-project\n",[7865,8520,8521,8524],{"class":7867,"line":1610},[7865,8522,8523],{"class":8202},"cd",[7865,8525,8518],{"class":7997},[7865,8527,8528],{"class":7867,"line":7884},[7865,8529,7904],{"emptyLinePlaceholder":1666},[7865,8531,8532],{"class":7867,"line":7890},[7865,8533,8534],{"class":7987},"# Agent Skillsをインストール（Remotionチームが提供するベストプラクティス集）\n",[7865,8536,8537,8539,8542,8544],{"class":7867,"line":7896},[7865,8538,8512],{"class":7993},[7865,8540,8541],{"class":7997}," skills",[7865,8543,7998],{"class":7997},[7865,8545,8546],{"class":7997}," remotion-dev/skills\n",[19,8548,8549],{},"SkillsをインストールするとClaude Codeがリモーションの規約を理解し、正しいコンポーネント構造でコードを生成できるようになる。38種類以上のカテゴリがあり、アニメーション・3D・字幕・音声・チャート・トランジションをカバーする。",[122,8551,8552],{"id":8552},"実際のプロンプト例",[19,8554,8555],{},"良いプロンプトの書き方にはコツがある。具体的な数字・解像度・秒数を最初に宣言し、シーンの境界を明示する：",[134,8557,8560],{"className":8558,"code":8559,"language":139,"meta":143},[137],"1920x1080px、30fps、10秒の動画を作って。\n\n0-2秒: ロゴフェードイン（中央配置、白背景）\n2-6秒: 3つの機能を順番にスライドイン（左から右）\n        - \"高速\" + 稲妻アイコン\n        - \"安全\" + シールドアイコン\n        - \"簡単\" + チェックアイコン\n6-9秒: CTAボタンが拡大して登場「今すぐ試す」\n9-10秒: フェードアウト\n\nカラー: bg #0a0a0a, accent #7c3aed, text white\n",[141,8561,8559],{"__ignoreMap":143},[19,8563,8564],{},"生成されたコードはRemotionスタジオでリアルタイムプレビューし、問題なければレンダリングする：",[134,8566,8568],{"className":7978,"code":8567,"language":7980,"meta":143,"style":143},"# プレビュー\nnpx remotion studio\n\n# レンダリング（MP4）\nnpx remotion render MyComposition out/video.mp4\n",[141,8569,8570,8575,8585,8589,8594],{"__ignoreMap":143},[7865,8571,8572],{"class":7867,"line":7868},[7865,8573,8574],{"class":7987},"# プレビュー\n",[7865,8576,8577,8579,8582],{"class":7867,"line":1605},[7865,8578,8512],{"class":7993},[7865,8580,8581],{"class":7997}," remotion",[7865,8583,8584],{"class":7997}," studio\n",[7865,8586,8587],{"class":7867,"line":1610},[7865,8588,7904],{"emptyLinePlaceholder":1666},[7865,8590,8591],{"class":7867,"line":7884},[7865,8592,8593],{"class":7987},"# レンダリング（MP4）\n",[7865,8595,8596,8598,8600,8603,8606],{"class":7867,"line":7890},[7865,8597,8512],{"class":7993},[7865,8599,8581],{"class":7997},[7865,8601,8602],{"class":7997}," render",[7865,8604,8605],{"class":7997}," MyComposition",[7865,8607,8608],{"class":7997}," out/video.mp4\n",[122,8610,8611],{"id":8611},"ユースケース別の向き不向き",[54,8613,8614,8627],{},[57,8615,8616],{},[60,8617,8618,8621,8624],{},[63,8619,8620],{},"ユースケース",[63,8622,8623],{},"向いている",[63,8625,8626],{},"向いていない",[70,8628,8629,8639,8649,8659,8670],{},[60,8630,8631,8634,8637],{},[75,8632,8633],{},"製品デモ動画",[75,8635,8636],{},"◎ データと連動した動的更新が容易",[75,8638,250],{},[60,8640,8641,8644,8647],{},[75,8642,8643],{},"データビジュアライゼーション",[75,8645,8646],{},"◎ コードで数値を動的に渡せる",[75,8648,250],{},[60,8650,8651,8654,8657],{},[75,8652,8653],{},"SNS縦型クリップ",[75,8655,8656],{},"◎ 9:16を指定するだけ",[75,8658,250],{},[60,8660,8661,8664,8667],{},[75,8662,8663],{},"人物が話す動画",[75,8665,8666],{},"△ HeyGen等の専用ツールが優勢",[75,8668,8669],{},"複雑なリップシンク",[60,8671,8672,8675,8678],{},[75,8673,8674],{},"長尺映像（10分超）",[75,8676,8677],{},"△ シーン管理が煩雑になる",[75,8679,250],{},[19,8681,8682],{},"ローカルレンダリングをするのでクラウド動画生成ツールのウォーターマーク・クレジット制限は一切関係ない。「ビルドアーティファクトとしての動画」という考え方で、製品変更のたびに動画をCI/CDで自動更新するパイプラインを組む事例もすでに登場している。",[205,8684],{},[32,8686,8687],{"id":8687},"ハマりポイントと回避策",[19,8689,8690],{},"実際に試してハマった点を正直に書く。",[122,8692,8694],{"id":8693},"mcp-関連","MCP 関連",[19,8696,8697,8700,8701,8703,8704,8707],{},[23,8698,8699],{},"SSEトランスポートを使っていてエラーが出る",": 2025年3月のMCP仕様アップデートでSSE（Server-Sent Events）トランスポートは非推奨になった。リモート接続には必ず",[141,8702,8328],{},"を使う。FastMCPなら",[141,8705,8706],{},"mcp.run(transport=\"streamable-http\")","に変えるだけで移行できる。",[19,8709,8710,8713],{},[23,8711,8712],{},"MCP サーバーを複数入れたら起動が遅くなった",": 5サーバー・58ツール構成だとコンテキスト消費が55,000トークンを超えることがある。AnthropicのTool Search機能（オンデマンドロード）を有効にすると約85%削減できる。モノリシックなサーバーを避け、用途別に小さく分割すること。",[19,8715,8716,8719],{},[23,8717,8718],{},"セキュリティ",": OWASP MCP Top 10（2025年公開）に従いサーバーを審査する。特にプロンプトインジェクション（ツールのdescriptionに悪意ある指示を埋め込む攻撃）に注意。コミュニティのサーバーは使用前にコードレビュー必須。",[122,8721,8723],{"id":8722},"skills-関連","Skills 関連",[19,8725,8726,8729],{},[23,8727,8728],{},"スキルが期待通りにトリガーされない",": descriptionフィールドが曖昧だとClaudeがスキルをロードしない。「〇〇というキーワードが出たら必ずこのSkillを使え」という強い表現を入れること。",[19,8731,8732,8735,8736,8739],{},[23,8733,8734],{},"コンテキスト上限付近でSkillが無視される",": セッションが長くなりコンテキストの70〜85%超えると精度が落ちる。",[141,8737,8738],{},"/compact","でコンテキストを圧縮するか、新しいセッションで継続する。",[122,8741,8742],{"id":8742},"ウェブサイト生成関連",[19,8744,8745,8748],{},[23,8746,8747],{},"一度に全部作らせようとした",": 「ECサイトを全部作って」は失敗率が高い。コンポーネント単位・ページ単位でイテレーティブに進めること。プランモードで方向性を確認してから実装させる。",[19,8750,8751,8754],{},[23,8752,8753],{},"生成コードをそのままコミットした",": ACM 2025年の研究で、LLMのコードは人間比で論理エラーが1.75倍多い可能性が示されている。テストを通してからコミットを徹底する。",[122,8756,8758],{"id":8757},"remotion-関連","Remotion 関連",[19,8760,8761,8764],{},[23,8762,8763],{},"最初から複雑な動画を作ろうとした",": 5秒のサンプルから始め、背景色→テキスト→アニメーション→複数シーンの順に積み上げること。",[19,8766,8767,8770],{},[23,8768,8769],{},"曖昧な指示を出した",": 「80px」「30フレーム（1秒）」のように数値で指定するとClaude Codeの精度が上がる。「大きいフォント」「ゆっくりフェード」は精度が下がる。",[19,8772,8773,8776],{},[23,8774,8775],{},"シーン間でキャラクターの外見が変わる",": Remotionは静的アセットを直接コードで参照するため、画像ファイルを固定すれば解決する。複雑な人物動画はHeyGen等の専用ツールを併用する。",[205,8778],{},[32,8780,8781],{"id":8781},"まとめと次のステップ",[54,8783,8784,8797],{},[57,8785,8786],{},[60,8787,8788,8791,8794],{},[63,8789,8790],{},"やること",[63,8792,8793],{},"優先度",[63,8795,8796],{},"所要時間",[70,8798,8799,8810,8820,8833],{},[60,8800,8801,8804,8807],{},[75,8802,8803],{},"MCPサーバーを一つ作る（社内API連携）",[75,8805,8806],{},"高",[75,8808,8809],{},"1〜2時間",[60,8811,8812,8815,8817],{},[75,8813,8814],{},"プロジェクトにCLAUDE.mdを追加",[75,8816,8806],{},[75,8818,8819],{},"30分",[60,8821,8822,8827,8830],{},[75,8823,8824,8826],{},[141,8825,8464],{}," を試す",[75,8828,8829],{},"中",[75,8831,8832],{},"10分",[60,8834,8835,8838,8840],{},[75,8836,8837],{},"Remotionで30秒のデモ動画を作る",[75,8839,8829],{},[75,8841,8842],{},"2〜3時間",[19,8844,8845],{},"Claude Codeがコミット全体の4%を生成し、Spotifyのエンジニアが2025年12月以降手でコードを書いていないという現在、これらのツールへの習熟は「あると便利」から「ないと遅い」に変わりつつある。まず一つのMCPサーバーかSkillから始め、「毎回説明するのが面倒なこと」を自動化するところから入るのが最短距離だ。",[205,8847],{},[10,8849,8850,8854,8862,8873,8875,8879,8932,8934,8938,8941,8950,8956,9031,9038,9042,9045,9048,9051,9131,9133,9137,9140,9143,9146,9177,9181,9184,9311,9315,9397,9401,9442,9448,9450,9454,9457,9460,9464,9467,9541,9544,9548,9551,9580,9583,9585,9589,9595,9598,9641,9644,9648,9651,9657,9660,9699,9703,9770,9773,9775,9779,9782,9785,9796,9802,9812,9815,9824,9833,9837,9843,9849,9853,9859,9865,9871,9873,9877,9936],{"lang":809},[14,8851,8853],{"id":8852},"claude-skills-mcp-website-generation-and-programmatic-video-in-production-a-practical-engineers-guide-2026","Claude Skills, MCP, Website Generation, and Programmatic Video in Production: A Practical Engineer's Guide 2026",[792,8855,8856],{},[19,8857,8858,8861],{},[23,8859,8860],{},"Version note",": This article is based on information as of May 2026. It references mcp Python SDK 1.27.0, MCP spec revision 2025-11-25, and the latest Remotion release. The MCP spec evolves rapidly, so verify the latest information at modelcontextprotocol.io.",[19,8863,8864,8865,8868,8869,8872],{},"I came up through semiconductor process development before moving to corporate VC, and I built a flood-detection IoT system from scratch on AWS (IoT Core, Lambda, DynamoDB, API Gateway) in about six months. The single most reliable lesson from that work: ",[23,8866,8867],{},"the real gotchas only show up when you run the code",". Skills, MCP, and Remotion behave the same way. So this piece is the ",[23,8870,8871],{},"shortest implementation path plus the actual gotchas"," I hit running everything end-to-end.",[205,8874],{},[32,8876,8878],{"id":8877},"tldr-what-youll-learn","TL;DR — What You'll Learn",[54,8880,8881,8891],{},[57,8882,8883],{},[60,8884,8885,8888],{},[63,8886,8887],{},"Topic",[63,8889,8890],{},"What You'll Pick Up",[70,8892,8893,8900,8908,8916,8924],{},[60,8894,8895,8897],{},[75,8896,7708],{},[75,8898,8899],{},"How SKILL.md works, its split with MCP, day-to-day patterns",[60,8901,8902,8905],{},[75,8903,8904],{},"Building MCP servers",[75,8906,8907],{},"The shortest path to a Python FastMCP server in under 50 lines",[60,8909,8910,8913],{},[75,8911,8912],{},"Website generation",[75,8914,8915],{},"Site generation workflow with Claude Code + Next.js / Nuxt 4",[60,8917,8918,8921],{},[75,8919,8920],{},"Remotion video",[75,8922,8923],{},"Writing video as code and rendering it through Claude Code",[60,8925,8926,8929],{},[75,8927,8928],{},"Pitfalls",[75,8930,8931],{},"The traps in each area and how to dodge them",[205,8933],{},[32,8935,8937],{"id":8936},"skills-vs-mcp-getting-the-mental-model-right","Skills vs MCP — Getting the Mental Model Right",[19,8939,8940],{},"If you spend any real time with Claude Code or Claude.ai, you'll hit the same question: \"How do I actually decide between Skills and MCP?\" Reading the docs doesn't always make it click. The mental model that does:",[19,8942,8943,8946,8947],{},[23,8944,8945],{},"Skills = the agent's \"playbook\"",", ",[23,8948,8949],{},"MCP = the agent's \"external connections.\"",[19,8951,8952,8953,8955],{},"Skills are SKILL.md files placed under ",[141,8954,7767],{},". Claude reads the task context, automatically loads the relevant skill, and follows its instructions. You write Markdown — no code. MCP, in contrast, is a client-server protocol over JSON-RPC 2.0 that connects Claude to external systems like GitHub, databases, or internal APIs.",[54,8957,8958,8969],{},[57,8959,8960],{},[60,8961,8962,8965,8967],{},[63,8963,8964],{},"Dimension",[63,8966,7188],{},[63,8968,7194],{},[70,8970,8971,8982,8993,9004,9015],{},[60,8972,8973,8976,8979],{},[75,8974,8975],{},"Implementation language",[75,8977,8978],{},"Markdown (SKILL.md)",[75,8980,8981],{},"Python / TypeScript / Go, etc.",[60,8983,8984,8987,8990],{},[75,8985,8986],{},"Context consumption",[75,8988,8989],{},"30–50 tokens (loaded only when needed)",[75,8991,8992],{},"Can exceed 50,000 tokens",[60,8994,8995,8998,9001],{},[75,8996,8997],{},"Use case",[75,8999,9000],{},"Embedding procedures and best practices",[75,9002,9003],{},"Real-time integration with external systems",[60,9005,9006,9009,9012],{},[75,9007,9008],{},"Cross-tool",[75,9010,9011],{},"Works in Claude Code / Cursor / Gemini CLI",[75,9013,9014],{},"Works in any MCP-capable client",[60,9016,9017,9020,9025],{},[75,9018,9019],{},"Setup",[75,9021,9022,9023],{},"Drop a file under ",[141,9024,7767],{},[75,9026,9027,9028],{},"Run a server and register it in ",[141,9029,9030],{},"claude_desktop_config.json",[19,9032,9033,9034,9037],{},"The key asymmetry is: ",[23,9035,9036],{},"a Skill can call MCP tools, but not the other way around."," Define procedures with Skills, connect external tools with MCP — putting them together is what unlocks fully automated agents.",[122,9039,9041],{"id":9040},"real-world-skills-patterns","Real-World Skills Patterns",[19,9043,9044],{},"As of March 2026, the Skills ecosystem is structured in three layers: official Anthropic skills, vetted third-party skills, and community skills. The frontend-design skill has crossed 270K installs, the Remotion skill is doing 117K installs per week, and gws (Google Workspace) hit 4,900 GitHub stars within three days of release.",[19,9046,9047],{},"Skills shine in any situation where \"I'm tired of explaining the same procedure over and over.\" Write a SQL assistant skill that knows your custom DB schema once, embed your team's coding conventions into a review skill, build a converter skill for your proprietary data format — after that, a single one-line prompt invokes them.",[19,9049,9050],{},"The minimum SKILL.md, when you write your own:",[134,9052,9054],{"className":7859,"code":9053,"language":7861,"meta":143,"style":143},"---\nname: my-skill\ndescription: >\n  Description here. Claude reads this to decide whether to load the skill.\n  Spell out the trigger conditions concretely.\n---\n\n# My Skill\n\n## Steps\n\n1. First, check X.\n2. Then, run Y.\n\n## Pitfalls\n\n- When Z, do not do W.\n",[141,9055,9056,9060,9064,9068,9073,9078,9082,9086,9090,9094,9099,9103,9108,9113,9117,9122,9126],{"__ignoreMap":143},[7865,9057,9058],{"class":7867,"line":7868},[7865,9059,7871],{},[7865,9061,9062],{"class":7867,"line":1605},[7865,9063,7876],{},[7865,9065,9066],{"class":7867,"line":1610},[7865,9067,7881],{},[7865,9069,9070],{"class":7867,"line":7884},[7865,9071,9072],{},"  Description here. Claude reads this to decide whether to load the skill.\n",[7865,9074,9075],{"class":7867,"line":7890},[7865,9076,9077],{},"  Spell out the trigger conditions concretely.\n",[7865,9079,9080],{"class":7867,"line":7896},[7865,9081,7871],{},[7865,9083,9084],{"class":7867,"line":7901},[7865,9085,7904],{"emptyLinePlaceholder":1666},[7865,9087,9088],{"class":7867,"line":7907},[7865,9089,7910],{},[7865,9091,9092],{"class":7867,"line":7913},[7865,9093,7904],{"emptyLinePlaceholder":1666},[7865,9095,9096],{"class":7867,"line":7918},[7865,9097,9098],{},"## Steps\n",[7865,9100,9101],{"class":7867,"line":7924},[7865,9102,7904],{"emptyLinePlaceholder":1666},[7865,9104,9105],{"class":7867,"line":7929},[7865,9106,9107],{},"1. First, check X.\n",[7865,9109,9110],{"class":7867,"line":7935},[7865,9111,9112],{},"2. Then, run Y.\n",[7865,9114,9115],{"class":7867,"line":7941},[7865,9116,7904],{"emptyLinePlaceholder":1666},[7865,9118,9119],{"class":7867,"line":7946},[7865,9120,9121],{},"## Pitfalls\n",[7865,9123,9124],{"class":7867,"line":7952},[7865,9125,7904],{"emptyLinePlaceholder":1666},[7865,9127,9128],{"class":7867,"line":7957},[7865,9129,9130],{},"- When Z, do not do W.\n",[205,9132],{},[32,9134,9136],{"id":9135},"building-an-mcp-server-in-50-lines-the-python-fastmcp-shortcut","Building an MCP Server in 50 Lines — The Python FastMCP Shortcut",[19,9138,9139],{},"MCP's adoption curve is unprecedented. From 2 million SDK downloads/month at launch (November 2024), it crossed 22 million when OpenAI adopted (March 2025), 45 million when Microsoft integrated (July), 68 million when AWS came on (November), and reached 97 million per month with over 10,000 public servers by March 2026. React took three years to reach those download numbers; MCP did it in 16 months.",[19,9141,9142],{},"The problem MCP solves is simple: previously, hooking your internal CRM, billing system, or feature-flag service into AI agents meant writing separate integration code for every AI client. Build one MCP-compatible server, and it works from Claude, ChatGPT, Cursor, and Gemini.",[122,9144,9019],{"id":9145},"setup",[134,9147,9149],{"className":7978,"code":9148,"language":7980,"meta":143,"style":143},"# uv recommended (pip works too)\nuv add mcp\n# or\npip install mcp\n",[141,9150,9151,9156,9164,9169],{"__ignoreMap":143},[7865,9152,9153],{"class":7867,"line":7868},[7865,9154,9155],{"class":7987},"# uv recommended (pip works too)\n",[7865,9157,9158,9160,9162],{"class":7867,"line":1605},[7865,9159,7994],{"class":7993},[7865,9161,7998],{"class":7997},[7865,9163,8001],{"class":7997},[7865,9165,9166],{"class":7867,"line":1610},[7865,9167,9168],{"class":7987},"# or\n",[7865,9170,9171,9173,9175],{"class":7867,"line":7884},[7865,9172,8011],{"class":7993},[7865,9174,8014],{"class":7997},[7865,9176,8001],{"class":7997},[122,9178,9180],{"id":9179},"minimal-server-python-fastmcp","Minimal Server (Python FastMCP)",[19,9182,9183],{},"Below is a minimal MCP server that actually runs. It includes all three primitives — Tools, Resources, and Prompts:",[134,9185,9187],{"className":8026,"code":9186,"language":8028,"meta":143,"style":143},"from mcp.server.fastmcp import FastMCP\n\nmcp = FastMCP(\"my-service\", json_response=True)\n\n# Tool: side-effecting operations (analogous to POST)\n@mcp.tool()\ndef check_stock(sku: str) -> dict:\n    \"\"\"Return the inventory count for the given SKU.\"\"\"\n    # In practice, replace with a DB query, etc.\n    return {\"sku\": sku, \"quantity\": 42, \"warehouse\": \"tokyo\"}\n\n# Resource: read-only data (analogous to GET)\n@mcp.resource(\"products://catalog\")\ndef get_catalog() -> str:\n    \"\"\"Return the product catalog listing.\"\"\"\n    return \"SKU-001: Widget A\\nSKU-002: Widget B\"\n\n# Prompt: a reusable prompt template\n@mcp.prompt()\ndef stock_check_prompt(sku: str) -> str:\n    \"\"\"Generate a prompt for checking inventory.\"\"\"\n    return f\"Check the stock for SKU {sku} and propose a restock if it is low.\"\n\nif __name__ == \"__main__\":\n    # Local dev: stdio\n    # mcp.run()\n    # Remote/production: Streamable HTTP\n    mcp.run(transport=\"streamable-http\", host=\"0.0.0.0\", port=3050)\n",[141,9188,9189,9193,9197,9201,9205,9210,9214,9218,9223,9228,9232,9236,9241,9245,9249,9254,9258,9262,9267,9271,9275,9280,9285,9289,9293,9298,9302,9307],{"__ignoreMap":143},[7865,9190,9191],{"class":7867,"line":7868},[7865,9192,8035],{},[7865,9194,9195],{"class":7867,"line":1605},[7865,9196,7904],{"emptyLinePlaceholder":1666},[7865,9198,9199],{"class":7867,"line":1610},[7865,9200,8044],{},[7865,9202,9203],{"class":7867,"line":7884},[7865,9204,7904],{"emptyLinePlaceholder":1666},[7865,9206,9207],{"class":7867,"line":7890},[7865,9208,9209],{},"# Tool: side-effecting operations (analogous to POST)\n",[7865,9211,9212],{"class":7867,"line":7896},[7865,9213,8058],{},[7865,9215,9216],{"class":7867,"line":7901},[7865,9217,8063],{},[7865,9219,9220],{"class":7867,"line":7907},[7865,9221,9222],{},"    \"\"\"Return the inventory count for the given SKU.\"\"\"\n",[7865,9224,9225],{"class":7867,"line":7913},[7865,9226,9227],{},"    # In practice, replace with a DB query, etc.\n",[7865,9229,9230],{"class":7867,"line":7918},[7865,9231,8078],{},[7865,9233,9234],{"class":7867,"line":7924},[7865,9235,7904],{"emptyLinePlaceholder":1666},[7865,9237,9238],{"class":7867,"line":7929},[7865,9239,9240],{},"# Resource: read-only data (analogous to GET)\n",[7865,9242,9243],{"class":7867,"line":7935},[7865,9244,8092],{},[7865,9246,9247],{"class":7867,"line":7941},[7865,9248,8097],{},[7865,9250,9251],{"class":7867,"line":7946},[7865,9252,9253],{},"    \"\"\"Return the product catalog listing.\"\"\"\n",[7865,9255,9256],{"class":7867,"line":7952},[7865,9257,8107],{},[7865,9259,9260],{"class":7867,"line":7957},[7865,9261,7904],{"emptyLinePlaceholder":1666},[7865,9263,9264],{"class":7867,"line":8114},[7865,9265,9266],{},"# Prompt: a reusable prompt template\n",[7865,9268,9269],{"class":7867,"line":8120},[7865,9270,8123],{},[7865,9272,9273],{"class":7867,"line":8126},[7865,9274,8129],{},[7865,9276,9277],{"class":7867,"line":8132},[7865,9278,9279],{},"    \"\"\"Generate a prompt for checking inventory.\"\"\"\n",[7865,9281,9282],{"class":7867,"line":8138},[7865,9283,9284],{},"    return f\"Check the stock for SKU {sku} and propose a restock if it is low.\"\n",[7865,9286,9287],{"class":7867,"line":8144},[7865,9288,7904],{"emptyLinePlaceholder":1666},[7865,9290,9291],{"class":7867,"line":8149},[7865,9292,8152],{},[7865,9294,9295],{"class":7867,"line":8155},[7865,9296,9297],{},"    # Local dev: stdio\n",[7865,9299,9300],{"class":7867,"line":8161},[7865,9301,8164],{},[7865,9303,9304],{"class":7867,"line":8167},[7865,9305,9306],{},"    # Remote/production: Streamable HTTP\n",[7865,9308,9309],{"class":7867,"line":8173},[7865,9310,8176],{},[122,9312,9314],{"id":9313},"registering-with-claude-desktop","Registering With Claude Desktop",[134,9316,9317],{"className":8183,"code":8184,"language":8185,"meta":143,"style":143},[141,9318,9319,9323,9335,9347,9365,9385,9389,9393],{"__ignoreMap":143},[7865,9320,9321],{"class":7867,"line":7868},[7865,9322,8193],{"class":8192},[7865,9324,9325,9327,9329,9331,9333],{"class":7867,"line":1605},[7865,9326,8199],{"class":8198},[7865,9328,8203],{"class":8202},[7865,9330,8206],{"class":8198},[7865,9332,8210],{"class":8209},[7865,9334,8213],{"class":8192},[7865,9336,9337,9339,9341,9343,9345],{"class":7867,"line":1610},[7865,9338,8218],{"class":8198},[7865,9340,8221],{"class":8202},[7865,9342,8206],{"class":8198},[7865,9344,8210],{"class":8209},[7865,9346,8213],{"class":8192},[7865,9348,9349,9351,9353,9355,9357,9359,9361,9363],{"class":7867,"line":7884},[7865,9350,8232],{"class":8198},[7865,9352,8235],{"class":8202},[7865,9354,8206],{"class":8198},[7865,9356,8210],{"class":8209},[7865,9358,8243],{"class":8242},[7865,9360,8028],{"class":7997},[7865,9362,8206],{"class":8242},[7865,9364,8250],{"class":8192},[7865,9366,9367,9369,9371,9373,9375,9377,9379,9381,9383],{"class":7867,"line":7890},[7865,9368,8232],{"class":8198},[7865,9370,8257],{"class":8202},[7865,9372,8206],{"class":8198},[7865,9374,8210],{"class":8209},[7865,9376,8264],{"class":8192},[7865,9378,8206],{"class":8242},[7865,9380,8269],{"class":7997},[7865,9382,8206],{"class":8242},[7865,9384,8274],{"class":8192},[7865,9386,9387],{"class":7867,"line":7896},[7865,9388,8279],{"class":8192},[7865,9390,9391],{"class":7867,"line":7901},[7865,9392,8284],{"class":8192},[7865,9394,9395],{"class":7867,"line":7907},[7865,9396,8289],{"class":8192},[122,9398,9400],{"id":9399},"choosing-a-transport","Choosing a Transport",[54,9402,9403,9416],{},[57,9404,9405],{},[60,9406,9407,9410,9413],{},[63,9408,9409],{},"Transport",[63,9411,9412],{},"Use Case",[63,9414,9415],{},"Notes",[70,9417,9418,9430],{},[60,9419,9420,9424,9427],{},[75,9421,9422],{},[141,9423,8315],{},[75,9425,9426],{},"Local, Claude Desktop integration, development",[75,9428,9429],{},"No network setup required",[60,9431,9432,9436,9439],{},[75,9433,9434],{},[141,9435,8328],{},[75,9437,9438],{},"Remote, team-shared, cloud-deployed",[75,9440,9441],{},"Standardized in the March 2025 spec. SSE is deprecated",[19,9443,9444,9445,9447],{},"For TypeScript, use ",[141,9446,8340],{}," with Zod. It supports the same three primitives (Tool / Resource / Prompt) as Python FastMCP, with strong type safety. In either language, \"the wire protocol is JSON-RPC 2.0,\" so connecting a TypeScript client to a Python server just works.",[205,9449],{},[32,9451,9453],{"id":9452},"generating-websites-with-claude-code-combining-claudemd-and-skills","Generating Websites With Claude Code — Combining CLAUDE.md and Skills",[19,9455,9456],{},"As of early 2026, Claude Code generates 4% of all GitHub commits, with daily VS Code installs reaching 29 million. The \"vibe coding\" workflow has matured to the point where someone with zero engineering experience can ship a Next.js app.",[19,9458,9459],{},"That said, telling Claude \"make me a site\" still produces the familiar AI-generated look — Inter font, purple gradient, white background. The way to escape that is the combination of CLAUDE.md and Skills.",[122,9461,9463],{"id":9462},"where-claudemd-fits","Where CLAUDE.md Fits",[19,9465,9466],{},"A CLAUDE.md placed at the project root is auto-loaded by Claude Code at session start. Use it to capture \"what this project is\":",[134,9468,9470],{"className":7859,"code":9469,"language":7861,"meta":143,"style":143},"# Project: MyApp\n\n## Stack\n- Nuxt 4 (Nuxt Content 3, Vue 3 Composition API)\n- Tailwind CSS v4, shadcn/ui\n- Deployed on Vercel\n\n## Rules\n- Always make sure the build passes before committing.\n- TypeScript strict mode is required.\n- Don't use MDX components (Nuxt Content's CommonMark only).\n\n## Don't\n- Leave console.log calls in committed code.\n- Use the any type.\n",[141,9471,9472,9476,9480,9485,9489,9493,9498,9502,9507,9512,9517,9522,9526,9531,9536],{"__ignoreMap":143},[7865,9473,9474],{"class":7867,"line":7868},[7865,9475,8370],{},[7865,9477,9478],{"class":7867,"line":1605},[7865,9479,7904],{"emptyLinePlaceholder":1666},[7865,9481,9482],{"class":7867,"line":1610},[7865,9483,9484],{},"## Stack\n",[7865,9486,9487],{"class":7867,"line":7884},[7865,9488,8384],{},[7865,9490,9491],{"class":7867,"line":7890},[7865,9492,8389],{},[7865,9494,9495],{"class":7867,"line":7896},[7865,9496,9497],{},"- Deployed on Vercel\n",[7865,9499,9500],{"class":7867,"line":7901},[7865,9501,7904],{"emptyLinePlaceholder":1666},[7865,9503,9504],{"class":7867,"line":7907},[7865,9505,9506],{},"## Rules\n",[7865,9508,9509],{"class":7867,"line":7913},[7865,9510,9511],{},"- Always make sure the build passes before committing.\n",[7865,9513,9514],{"class":7867,"line":7918},[7865,9515,9516],{},"- TypeScript strict mode is required.\n",[7865,9518,9519],{"class":7867,"line":7924},[7865,9520,9521],{},"- Don't use MDX components (Nuxt Content's CommonMark only).\n",[7865,9523,9524],{"class":7867,"line":7929},[7865,9525,7904],{"emptyLinePlaceholder":1666},[7865,9527,9528],{"class":7867,"line":7935},[7865,9529,9530],{},"## Don't\n",[7865,9532,9533],{"class":7867,"line":7941},[7865,9534,9535],{},"- Leave console.log calls in committed code.\n",[7865,9537,9538],{"class":7867,"line":7946},[7865,9539,9540],{},"- Use the any type.\n",[19,9542,9543],{},"The split between CLAUDE.md and Skills is clean: CLAUDE.md is \"the context of this project,\" Skills are \"how to do this task.\" Together, you get an agent that understands the project and executes specific tasks correctly.",[122,9545,9547],{"id":9546},"site-generation-workflow-with-nextjs-nuxt","Site Generation Workflow With Next.js / Nuxt",[19,9549,9550],{},"A workflow that actually works in practice:",[552,9552,9553,9559,9566,9574],{},[315,9554,9555,9558],{},[23,9556,9557],{},"Create CLAUDE.md",": stack, naming conventions, things to avoid",[315,9560,9561,8461,9564],{},[23,9562,9563],{},"Add an official or custom Skill",[141,9565,8464],{},[315,9567,9568,8461,9571],{},[23,9569,9570],{},"Sanity-check requirements in plan mode",[141,9572,9573],{},"claude --plan \"I want a landing page that looks like X\"",[315,9575,9576,9579],{},[23,9577,9578],{},"Implement iteratively",": don't ask for the whole thing at once — go section by section",[19,9581,9582],{},"Step 3 is the critical one. Asking for an implementation directly sometimes produces \"code that solves the wrong problem.\" Have Claude explore the existing codebase first, identify the change locations, impact range, and edge cases, and surface them as a plan. Once the plan looks right, just say yes — that alone prevents most implementation mistakes.",[205,9584],{},[32,9586,9588],{"id":9587},"writing-video-as-code-with-claude-code-remotion","Writing Video as Code With Claude Code + Remotion",[19,9590,9591,9592,9594],{},"The idea of \"managing video in code\" lands naturally for front-end engineers. Remotion is exactly that — a React-based video framework where ",[141,9593,8493],{}," returns the current frame number and you describe animation in code. Combined with Claude Code, it produces a \"write video by prompt\" workflow.",[122,9596,9019],{"id":9597},"setup-1",[134,9599,9601],{"className":7978,"code":9600,"language":7980,"meta":143,"style":143},"# Create a Remotion project\nnpx create-video@latest my-video-project\ncd my-video-project\n\n# Install Agent Skills (best-practices bundle from the Remotion team)\nnpx skills add remotion-dev/skills\n",[141,9602,9603,9608,9616,9622,9626,9631],{"__ignoreMap":143},[7865,9604,9605],{"class":7867,"line":7868},[7865,9606,9607],{"class":7987},"# Create a Remotion project\n",[7865,9609,9610,9612,9614],{"class":7867,"line":1605},[7865,9611,8512],{"class":7993},[7865,9613,8515],{"class":7997},[7865,9615,8518],{"class":7997},[7865,9617,9618,9620],{"class":7867,"line":1610},[7865,9619,8523],{"class":8202},[7865,9621,8518],{"class":7997},[7865,9623,9624],{"class":7867,"line":7884},[7865,9625,7904],{"emptyLinePlaceholder":1666},[7865,9627,9628],{"class":7867,"line":7890},[7865,9629,9630],{"class":7987},"# Install Agent Skills (best-practices bundle from the Remotion team)\n",[7865,9632,9633,9635,9637,9639],{"class":7867,"line":7896},[7865,9634,8512],{"class":7993},[7865,9636,8541],{"class":7997},[7865,9638,7998],{"class":7997},[7865,9640,8546],{"class":7997},[19,9642,9643],{},"Once the skills are in place, Claude Code understands Remotion conventions and generates code with the right component structure. There are 38+ categories covering animation, 3D, captions, audio, charts, and transitions.",[122,9645,9647],{"id":9646},"example-prompt","Example Prompt",[19,9649,9650],{},"There's a knack to writing good prompts: declare the concrete numbers — resolution, duration — up front, and make scene boundaries explicit:",[134,9652,9655],{"className":9653,"code":9654,"language":139,"meta":143},[137],"Create a 1920x1080px, 30fps, 10-second video.\n\n0–2s: Logo fades in (centered, white background)\n2–6s: Three features slide in one by one (left to right)\n        - \"Fast\" + lightning icon\n        - \"Safe\" + shield icon\n        - \"Easy\" + check icon\n6–9s: CTA button scales up: \"Try it now\"\n9–10s: Fade out\n\nColors: bg #0a0a0a, accent #7c3aed, text white\n",[141,9656,9654],{"__ignoreMap":143},[19,9658,9659],{},"Preview the generated code live in Remotion Studio, and once it looks right, render:",[134,9661,9663],{"className":7978,"code":9662,"language":7980,"meta":143,"style":143},"# Preview\nnpx remotion studio\n\n# Render (MP4)\nnpx remotion render MyComposition out/video.mp4\n",[141,9664,9665,9670,9678,9682,9687],{"__ignoreMap":143},[7865,9666,9667],{"class":7867,"line":7868},[7865,9668,9669],{"class":7987},"# Preview\n",[7865,9671,9672,9674,9676],{"class":7867,"line":1605},[7865,9673,8512],{"class":7993},[7865,9675,8581],{"class":7997},[7865,9677,8584],{"class":7997},[7865,9679,9680],{"class":7867,"line":1610},[7865,9681,7904],{"emptyLinePlaceholder":1666},[7865,9683,9684],{"class":7867,"line":7884},[7865,9685,9686],{"class":7987},"# Render (MP4)\n",[7865,9688,9689,9691,9693,9695,9697],{"class":7867,"line":7890},[7865,9690,8512],{"class":7993},[7865,9692,8581],{"class":7997},[7865,9694,8602],{"class":7997},[7865,9696,8605],{"class":7997},[7865,9698,8608],{"class":7997},[122,9700,9702],{"id":9701},"what-remotion-is-and-isnt-good-at","What Remotion Is and Isn't Good At",[54,9704,9705,9717],{},[57,9706,9707],{},[60,9708,9709,9711,9714],{},[63,9710,9412],{},[63,9712,9713],{},"Good Fit",[63,9715,9716],{},"Not a Good Fit",[70,9718,9719,9729,9739,9749,9760],{},[60,9720,9721,9724,9727],{},[75,9722,9723],{},"Product demo videos",[75,9725,9726],{},"◎ Easy to wire up to data and update dynamically",[75,9728,250],{},[60,9730,9731,9734,9737],{},[75,9732,9733],{},"Data visualization",[75,9735,9736],{},"◎ Pass numbers in dynamically through code",[75,9738,250],{},[60,9740,9741,9744,9747],{},[75,9742,9743],{},"Vertical social clips",[75,9745,9746],{},"◎ Just specify 9:16",[75,9748,250],{},[60,9750,9751,9754,9757],{},[75,9752,9753],{},"Talking-head videos",[75,9755,9756],{},"△ Specialized tools like HeyGen win",[75,9758,9759],{},"Complex lip-sync",[60,9761,9762,9765,9768],{},[75,9763,9764],{},"Long-form footage (10+ min)",[75,9766,9767],{},"△ Scene management gets unwieldy",[75,9769,250],{},[19,9771,9772],{},"Because rendering happens locally, watermarks and credit caps from cloud video generators don't apply at all. Treating \"video as a build artifact\" — auto-rebuilding the video in CI/CD whenever the product changes — is already showing up in real-world pipelines.",[205,9774],{},[32,9776,9778],{"id":9777},"pitfalls-and-how-to-avoid-them","Pitfalls and How to Avoid Them",[19,9780,9781],{},"Here, honestly, are the things that tripped me up.",[122,9783,7194],{"id":9784},"mcp",[19,9786,9787,9790,9791,9793,9794,849],{},[23,9788,9789],{},"Errors when using SSE transport."," The March 2025 MCP spec update deprecated SSE (Server-Sent Events). For remote connections, always use ",[141,9792,8328],{},". With FastMCP, switching is just ",[141,9795,8706],{},[19,9797,9798,9801],{},[23,9799,9800],{},"Multiple MCP servers slowing startup."," A 5-server, 58-tool setup can push context consumption past 55,000 tokens. Anthropic's Tool Search feature (on-demand loading) cuts that by about 85%. Avoid monolithic servers — split by purpose into smaller ones.",[19,9803,9804,9807,9808,9811],{},[23,9805,9806],{},"Security."," Audit your servers against the OWASP MCP Top 10 (published in 2025). Watch especially for prompt injection (malicious instructions hidden in a tool's ",[141,9809,9810],{},"description","). Code-review community servers before using them.",[122,9813,7188],{"id":9814},"skills",[19,9816,9817,9820,9821,9823],{},[23,9818,9819],{},"The skill doesn't trigger when expected."," A vague ",[141,9822,9810],{}," field means Claude won't load the skill. Use strong language like \"whenever the user mentions X, use this skill.\"",[19,9825,9826,9829,9830,9832],{},[23,9827,9828],{},"Skills get ignored near the context limit."," Once a session crosses 70–85% of the context window, accuracy drops. Use ",[141,9831,8738],{}," to compress context, or continue in a fresh session.",[122,9834,9836],{"id":9835},"website-generation","Website Generation",[19,9838,9839,9842],{},[23,9840,9841],{},"\"Build the whole thing at once.\""," \"Build me an entire e-commerce site\" has a very high failure rate. Iterate component by component, page by page. Confirm direction in plan mode before implementation.",[19,9844,9845,9848],{},[23,9846,9847],{},"Committing generated code as-is."," A 2025 ACM study suggests LLM-written code has roughly 1.75× more logic errors than human-written code. Make sure tests pass before committing.",[122,9850,9852],{"id":9851},"remotion","Remotion",[19,9854,9855,9858],{},[23,9856,9857],{},"Trying to build something complex from the start."," Start with a 5-second sample, then layer up: background color → text → animation → multiple scenes.",[19,9860,9861,9864],{},[23,9862,9863],{},"Vague directions."," \"80px,\" \"30 frames (1s),\" and other concrete numbers boost Claude Code's accuracy. \"Big font\" and \"slow fade\" hurt it.",[19,9866,9867,9870],{},[23,9868,9869],{},"Character appearance changes between scenes."," Remotion references static assets directly in code, so pin the image files. For complex character video, pair with a specialized tool like HeyGen.",[205,9872],{},[32,9874,9876],{"id":9875},"wrap-up-and-next-steps","Wrap-up and Next Steps",[54,9878,9879,9892],{},[57,9880,9881],{},[60,9882,9883,9886,9889],{},[63,9884,9885],{},"Action",[63,9887,9888],{},"Priority",[63,9890,9891],{},"Time Required",[70,9893,9894,9904,9914,9926],{},[60,9895,9896,9899,9901],{},[75,9897,9898],{},"Build one MCP server (internal API integration)",[75,9900,6092],{},[75,9902,9903],{},"1–2 hours",[60,9905,9906,9909,9911],{},[75,9907,9908],{},"Add CLAUDE.md to your project",[75,9910,6092],{},[75,9912,9913],{},"30 minutes",[60,9915,9916,9921,9923],{},[75,9917,9918,9919],{},"Try ",[141,9920,8464],{},[75,9922,6095],{},[75,9924,9925],{},"10 minutes",[60,9927,9928,9931,9933],{},[75,9929,9930],{},"Build a 30-second demo video in Remotion",[75,9932,6095],{},[75,9934,9935],{},"2–3 hours",[19,9937,9938],{},"With Claude Code generating 4% of all commits and Spotify's engineers reportedly writing no code by hand since December 2025, fluency with these tools is shifting from \"nice to have\" to \"you'll fall behind without it.\" Start with one MCP server or one Skill — automating \"the thing I'm tired of explaining every time\" is the shortest path in.",[9940,9941,9942],"style",{},"html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html pre.shiki code .shSDL, html code.shiki .shSDL{--shiki-default:#6272A4}html pre.shiki code .sAOxA, html code.shiki .sAOxA{--shiki-default:#50FA7B}html pre.shiki code .s-mGx, html code.shiki .s-mGx{--shiki-default:#F1FA8C}html pre.shiki code .sCdxs, html code.shiki .sCdxs{--shiki-default:#F8F8F2}html pre.shiki code .sY8FZ, html code.shiki .sY8FZ{--shiki-default:#8BE9FE}html pre.shiki code .sLL85, html code.shiki .sLL85{--shiki-default:#8BE9FD}html pre.shiki code .s0Tla, html code.shiki .s0Tla{--shiki-default:#FF79C6}html pre.shiki code .seVfx, html code.shiki .seVfx{--shiki-default:#E9F284}",{"title":143,"searchDepth":1605,"depth":1605,"links":9944},[9945,9946,9949,9955,9959,9964,9970,9971,9972,9975,9981,9985,9990,9996],{"id":7688,"depth":1605,"text":7689},{"id":7748,"depth":1605,"text":7749,"children":9947},[9948],{"id":7846,"depth":1610,"text":7847},{"id":7965,"depth":1605,"text":7966,"children":9950},[9951,9952,9953,9954],{"id":7975,"depth":1610,"text":7975},{"id":8019,"depth":1610,"text":8020},{"id":8179,"depth":1610,"text":8180},{"id":8292,"depth":1610,"text":8292},{"id":8346,"depth":1605,"text":8347,"children":9956},[9957,9958],{"id":8356,"depth":1610,"text":8357},{"id":8443,"depth":1610,"text":8444},{"id":8486,"depth":1605,"text":8487,"children":9960},[9961,9962,9963],{"id":8497,"depth":1610,"text":7975},{"id":8552,"depth":1610,"text":8552},{"id":8611,"depth":1610,"text":8611},{"id":8687,"depth":1605,"text":8687,"children":9965},[9966,9967,9968,9969],{"id":8693,"depth":1610,"text":8694},{"id":8722,"depth":1610,"text":8723},{"id":8742,"depth":1610,"text":8742},{"id":8757,"depth":1610,"text":8758},{"id":8781,"depth":1605,"text":8781},{"id":8877,"depth":1605,"text":8878},{"id":8936,"depth":1605,"text":8937,"children":9973},[9974],{"id":9040,"depth":1610,"text":9041},{"id":9135,"depth":1605,"text":9136,"children":9976},[9977,9978,9979,9980],{"id":9145,"depth":1610,"text":9019},{"id":9179,"depth":1610,"text":9180},{"id":9313,"depth":1610,"text":9314},{"id":9399,"depth":1610,"text":9400},{"id":9452,"depth":1605,"text":9453,"children":9982},[9983,9984],{"id":9462,"depth":1610,"text":9463},{"id":9546,"depth":1610,"text":9547},{"id":9587,"depth":1605,"text":9588,"children":9986},[9987,9988,9989],{"id":9597,"depth":1610,"text":9019},{"id":9646,"depth":1610,"text":9647},{"id":9701,"depth":1610,"text":9702},{"id":9777,"depth":1605,"text":9778,"children":9991},[9992,9993,9994,9995],{"id":9784,"depth":1610,"text":7194},{"id":9814,"depth":1610,"text":7188},{"id":9835,"depth":1610,"text":9836},{"id":9851,"depth":1610,"text":9852},{"id":9875,"depth":1605,"text":9876},"Claude Codeのスキルシステム・MCPサーバー構築・ウェブサイト生成・Remotion動画制作まで、エンジニアが実際に業務で使える知識を一冊にまとめた。コードと具体的な数字で語る実践ガイド。",{"date":7636,"image":9999,"alt":10000,"tags":10001,"tagsEn":10005,"published":1666},"/blogs-img/blog-claude-skills-mcp-web-video-engineer-guide.png","Claude Skills・MCP・Remotionによるエンジニア向け実践ガイド",[10002,7194,10003,9852,10004],"Claude Code","AI Engineering","チュートリアル",[10002,7194,10003,9852,10006],"Tutorial","/blogs/13-claude-skills-mcp-web-video-engineer-guide",{"title":7656,"description":9997},"blogs/13-claude-skills-mcp-web-video-engineer-guide","BtjILs7916jX3WSFJ1C6cKBy_BxG6i2otBjsPReM87A",{"id":10012,"title":10013,"body":10014,"description":10625,"extension":1651,"meta":10626,"navigation":1666,"ogImage":10628,"path":10640,"seo":10641,"stem":10642,"__hash__":10643},"content/blogs/14-cosmology-frontier-2026.md","宇宙の終わり、実は分かってないって知ってた？JWSTとDESIが揺さぶる「宇宙の常識」を3つの仮説で解説 | Wait, We Don't Actually Know How the Universe Ends? JWST and DESI Are Cracking Cosmology — 3 Wild Theories Explained",{"type":7,"value":10015,"toc":10607},[10016,10307,10309],[10,10017,10018,10022,10025,10029,10032,10035,10046,10049,10051,10055,10069,10072,10079,10099,10102,10109,10116,10119,10121,10125,10132,10139,10150,10153,10157,10164,10175,10179,10186,10197,10201,10207,10222,10229,10231,10234,10237,10264,10271,10274,10281,10283,10287],{"lang":12},[14,10019,10021],{"id":10020},"宇宙の終わり実は分かってないって知ってたjwstとdesiが揺さぶる宇宙の常識を3つの仮説で解説","宇宙の終わり、実は分かってないって知ってた？JWSTとDESIが揺さぶる「宇宙の常識」を3つの仮説で解説",[19,10023,10024],{},"私、もともと半導体プロセス開発の出身で、「観測データが理論より先に動く瞬間」を現場で何度も見てきた人間です。今回の宇宙論の話は、まさにその\"瞬間\"が起きてる気がして、最新論文50本読み込んで素人向けに整理してみました。",[32,10026,10028],{"id":10027},"え宇宙の常識ってそんなに揺らいでるの","え、宇宙の常識ってそんなに揺らいでるの？",[19,10030,10031],{},"「宇宙はビッグバンで始まって、今もずっと膨張してる」——学校でこう習った人、多いよね。",[19,10033,10034],{},"実はこの当たり前、ここ数年でちょっと怪しくなってきてる。",[19,10036,10037,10038,10041,10042,10045],{},"きっかけは2つの新しい観測装置。",[23,10039,10040],{},"JWST（ジェイムズ・ウェッブ宇宙望遠鏡）"," と ",[23,10043,10044],{},"DESI（ダーク・エネルギー分光器）","。前者は「銀河を異常に鮮明に写す巨大カメラ」、後者は「宇宙の3D地図を作る分光装置」みたいなもの。この2つが、これまで完璧に近いと思われていた「宇宙の標準モデル」に小さな、でも無視できないヒビを入れている。",[19,10047,10048],{},"最新の宇宙論論文を50本読み込んで、何が起きてるのかをできる限り素人向けに整理してみた。",[205,10050],{},[32,10052,10054],{"id":10053},"宇宙の取扱説明書が合わなくなってきた","「宇宙の取扱説明書」が合わなくなってきた",[19,10056,10057,10058,10061,10062,10065,10066,4299],{},"宇宙論には ",[23,10059,10060],{},"ΛCDM（ラムダ・シーディーエム）"," という標準モデルがある。要するに、宇宙の挙動を ",[23,10063,10064],{},"「通常の物質 + 暗黒物質 + 暗黒エネルギー」"," の3つで説明する ",[23,10067,10068],{},"「宇宙の取扱説明書」",[19,10070,10071],{},"ここ20年、この説明書はめちゃくちゃよく当たってきた。なのに最近、いくつか「あれ？」が出てる。",[19,10073,10074,10075,10078],{},"代表的なのが ",[23,10076,10077],{},"「H0張力（ハッブル定数の食い違い）」","。H0っていうのは「宇宙が今どれくらいの速さで膨張してるか」を表す数字。これを2つの方法で測ると——",[312,10080,10081,10090],{},[315,10082,10083,10086,10087],{},[23,10084,10085],{},"方法A","：ビッグバンの残光（CMB＝宇宙背景放射）から逆算 → だいたい ",[23,10088,10089],{},"67〜68",[315,10091,10092,10095,10096],{},[23,10093,10094],{},"方法B","：近くの銀河の超新星から直接測定 → だいたい ",[23,10097,10098],{},"73",[19,10100,10101],{},"数字、合ってないよね？",[19,10103,10104,10105,10108],{},"たとえると、",[23,10106,10107],{},"同じ部屋の気温を温度計2つで測ったら20度と23度","だった、みたいな話。誤差じゃ説明できないレベルで違う。",[19,10110,10111,10112,10115],{},"しかもこの差、観測精度が上がるたびに",[23,10113,10114],{},"消えるどころかむしろ濃くなってる","。だから「ただの偶然」では片付けにくくなってきた。",[19,10117,10118],{},"これに加えて、JWSTが見せた風景がまた騒ぎを大きくしている。",[205,10120],{},[32,10122,10124],{"id":10123},"jwstが見せた宇宙の早熟児と3つのワクワクする仮説","JWSTが見せた「宇宙の早熟児」と、3つのワクワクする仮説",[19,10126,10127,10128,10131],{},"JWSTは「宇宙誕生からたった3〜4億年後」の銀河まで撮れる。そこに写っていたのは、",[23,10129,10130],{},"「明るすぎる」「巨大すぎる」「成熟しすぎている」銀河","たち。",[19,10133,10134,10135,10138],{},"人間でいうと ",[23,10136,10137],{},"「生まれて3ヶ月の赤ちゃんがプロサッカー選手だった」"," ような違和感。",[19,10140,10141,10142,10145,10146,10149],{},"しかも、宇宙が透明になった「再電離」という現象の主役も、これまで考えられていた ",[23,10143,10144],{},"巨大ブラックホール（AGN）"," ではなく、",[23,10147,10148],{},"小さくて目立たない矮小銀河の集団"," だった可能性が高くなってきた。常識がひっくり返ってる。",[19,10151,10152],{},"これらをまとめて説明できる仮説が、論文界隈で3つ出てきている。素人向けにかなり噛み砕くと、こんな感じ。",[122,10154,10156],{"id":10155},"仮説宇宙はリサイクル品だった磁化メモリー反跳宇宙","仮説①：宇宙はリサイクル品だった？「磁化メモリー反跳宇宙」",[19,10158,10159,10160,10163],{},"ビッグバンが宇宙の最初じゃなく、",[23,10161,10162],{},"前の宇宙の磁場の記憶を引き継いで反跳した","——という仮説。",[312,10165,10166,10169,10172],{},[315,10167,10168],{},"面白さポイント：宇宙が「中古品」かもしれない",[315,10170,10171],{},"弱点：反跳の仕組み自体は別途説明が必要",[315,10173,10174],{},"検証：CMBの偏光データに「ねじれた光のパターン」が見つかれば一気に有力候補に",[122,10176,10178],{"id":10177},"仮説宇宙のぼっこりが早熟銀河を作った非可換クランプ熱化宇宙","仮説②：宇宙の「ぼっこり」が早熟銀河を作った？「非可換クランプ熱化宇宙」",[19,10180,10181,10182,10185],{},"生まれたての宇宙のごく狭い領域に ",[23,10183,10184],{},"極端な揺らぎが集中して","、それが初期銀河の早熟化と未来の真空崩壊の種になった——という説。",[312,10187,10188,10191,10194],{},[315,10189,10190],{},"面白さポイント：JWSTで見える「早すぎる銀河」が一気に説明できる",[315,10192,10193],{},"弱点：原始ブラックホールを作りすぎる危険があり、観測制約とギリギリ",[315,10195,10196],{},"検証：宇宙が出した「特定の周波数だけ強い重力波の山」が見つかるかどうか",[122,10198,10200],{"id":10199},"仮説暗黒物質と暗黒エネルギーがコソコソ会話してた相互作用暗黒セクター緩慢終焉","仮説③：暗黒物質と暗黒エネルギーがコソコソ会話してた？「相互作用暗黒セクター緩慢終焉」",[19,10202,10203,10206],{},[23,10204,10205],{},"暗黒物質と暗黒エネルギーが微妙に交信してる"," ことで、最近の宇宙でだけ膨張史がズレる——という説。",[312,10208,10209,10212,10215],{},[315,10210,10211],{},"面白さポイント：H0の食い違いと「銀河構造の伸び悩み問題（S8）」を一発で解決する可能性",[315,10213,10214],{},"弱点：宇宙の誕生自体には踏み込まない",[315,10216,10217,10218,10221],{},"終焉：ビッグリップでも熱的死でもなく、",[23,10219,10220],{},"「成長が止まった静かな宇宙」"," に向かう",[19,10223,10224,10225,10228],{},"特に③は「現在のデータに最も近い」と評価されている。",[23,10226,10227],{},"「宇宙の終わりは爆発でも凍結でもなく、ただ静かに音だけが残る」"," って、ちょっとロマンあるよね。",[205,10230],{},[32,10232,10233],{"id":10233},"答えは生きてるうちに出るかもしれない",[19,10235,10236],{},"ここがいちばん面白いところ。",[19,10238,10239,10240,10243,10244,10247,10248,10251,10252,10255,10256,10259,10260,10263],{},"これらの仮説は ",[23,10241,10242],{},"今後5〜10年で決着がつく可能性が高い","。ヨーロッパの ",[23,10245,10246],{},"Euclid","、アメリカの ",[23,10249,10250],{},"Rubin / Roman","、CMB専用の ",[23,10253,10254],{},"CMB-S4 / LiteBIRD","、電波の ",[23,10257,10258],{},"SKA","、宇宙重力波の ",[23,10261,10262],{},"LISA"," ——次世代観測装置がほぼ同時期に動き出している。",[19,10265,10266,10267,10270],{},"つまり、",[23,10268,10269],{},"自分が生きてる間に「宇宙の終わり方」の方向くらいは出る","。これ、人類史で初めての世代。",[19,10272,10273],{},"宇宙のニュースを見るたびに「また何か騒いでるな」って思ってた人、これからはちょっと違って見えるはず。「あ、今この3つの仮説のどれが正しいか、世界中で勝負してるんだな」って。",[19,10275,10276,10277,10280],{},"面白かったら ",[23,10278,10279],{},"X でシェア"," してくれると嬉しい。コメントで「どの仮説が好き？」を教えてくれたら、もっと深掘り記事を書きます。",[205,10282],{},[19,10284,10285],{},[23,10286,7335],{},[312,10288,10289,10295,10301],{},[315,10290,10291,10294],{},[23,10292,10293],{},"案1：ダークマターの正体って結局何？候補理論を素人向けに3つ並べて比較"," — WIMP・アクシオン・原始ブラックホールという3つの主要候補を、強み・弱み・「どこまで分かってきたか」で図解。今回触れた「暗黒物質と暗黒エネルギーの会話」をもっと深掘りする続編。",[315,10296,10297,10300],{},[23,10298,10299],{},"案2：JWSTが見せた「ありえない銀河」TOP5"," — 本文で軽く触れた「早すぎる銀河」を実際の発見例ベースで5つ厳選。観測画像と「なぜ常識外か」を並べてビジュアル中心に語る。",[315,10302,10303,10306],{},[23,10304,10305],{},"案3：宇宙の終わり方3パターンを可視化：Big Rip vs 熱的死 vs 静かな終焉"," — 仮説①〜③それぞれが導く「終わりの形」を時間軸に沿って図解。読者が「結局どれが好き？」を選びたくなる構成。",[205,10308],{},[10,10310,10311,10315,10322,10326,10329,10332,10343,10346,10348,10352,10362,10365,10372,10395,10398,10401,10408,10411,10413,10417,10428,10431,10446,10449,10453,10460,10471,10475,10482,10493,10497,10504,10518,10525,10527,10531,10534,10564,10571,10574,10581,10583,10587],{"lang":809},[14,10312,10314],{"id":10313},"wait-we-dont-actually-know-how-the-universe-ends-jwst-and-desi-are-cracking-cosmology-3-wild-theories-explained","Wait, We Don't Actually Know How the Universe Ends? JWST and DESI Are Cracking Cosmology — 3 Wild Theories Explained",[19,10316,10317,10318,10321],{},"I came up through semiconductor process development, where I saw, more than once, ",[23,10319,10320],{},"observation outrun theory"," in real time. Cosmology in 2026 has that same flavor — so I read 50 of the latest papers and tried to translate it into something a non-physicist can actually follow.",[32,10323,10325],{"id":10324},"hold-on-the-universes-settled-science-is-cracking","Hold on — the universe's \"settled science\" is cracking?",[19,10327,10328],{},"You probably learned in school that the universe started with the Big Bang and has been expanding ever since.",[19,10330,10331],{},"Turns out — that tidy story is starting to crack.",[19,10333,10334,10335,10338,10339,10342],{},"The trigger is two new instruments: ",[23,10336,10337],{},"JWST (the James Webb Space Telescope)"," and ",[23,10340,10341],{},"DESI (the Dark Energy Spectroscopic Instrument)",". JWST is basically a giant camera that captures galaxies in stunning sharpness. DESI is a machine that builds a 3D map of the universe by spectroscopically scanning millions of galaxies. Together, they've put small but undeniable cracks in what was, until recently, the closest thing physics had to a perfect description of the cosmos.",[19,10344,10345],{},"I read through 50 of the latest cosmology papers and tried to translate what's actually going on, in plain English.",[205,10347],{},[32,10349,10351],{"id":10350},"the-user-manual-of-the-universe-doesnt-add-up-anymore","The \"user manual of the universe\" doesn't add up anymore",[19,10353,10354,10355,10358,10359,849],{},"Cosmologists have a standard model called ",[23,10356,10357],{},"ΛCDM (Lambda Cold Dark Matter)",". In plain terms, it's the universe's \"user manual\" — a recipe that explains everything using ",[23,10360,10361],{},"regular matter + dark matter + dark energy",[19,10363,10364],{},"For about 20 years, this manual has been freakishly accurate. But lately, a few \"wait, what?\" moments have shown up.",[19,10366,10367,10368,10371],{},"The most famous one is the ",[23,10369,10370],{},"H0 tension",". H0 is the number that tells us how fast the universe is expanding right now. Measure it two ways and you get:",[312,10373,10374,10387],{},[315,10375,10376,10379,10380,10383,10384],{},[23,10377,10378],{},"Method A",": Run the math backward from the leftover glow of the Big Bang (the ",[23,10381,10382],{},"CMB",") → about ",[23,10385,10386],{},"67–68",[315,10388,10389,10392,10393],{},[23,10390,10391],{},"Method B",": Look at supernovae in nearby galaxies directly → about ",[23,10394,10098],{},[19,10396,10397],{},"Those numbers don't match.",[19,10399,10400],{},"It's like measuring the temperature of the same room with two thermometers and getting 68°F and 73°F. Not a measurement error — a real disagreement.",[19,10402,10403,10404,10407],{},"And here's the kicker: the more precise our instruments get, the ",[23,10405,10406],{},"stronger"," the disagreement becomes. That makes \"it's just statistical noise\" harder and harder to defend.",[19,10409,10410],{},"Then JWST showed up and made things louder.",[205,10412],{},[32,10414,10416],{"id":10415},"jwsts-prodigy-galaxies-and-3-wild-new-theories","JWST's \"prodigy galaxies\" and 3 wild new theories",[19,10418,10419,10420,10423,10424,10427],{},"JWST can image galaxies from when the universe was only ",[23,10421,10422],{},"300–400 million years old",". And what it's seeing? Galaxies that are ",[23,10425,10426],{},"too bright, too big, too mature"," for that age.",[19,10429,10430],{},"It's like finding a 3-month-old baby playing professional soccer.",[19,10432,10433,10434,10437,10438,10441,10442,10445],{},"On top of that, the cosmic event called ",[23,10435,10436],{},"reionization"," — when the universe became transparent — was probably driven not by giant black holes (",[23,10439,10440],{},"AGN","), as we used to think, but by hordes of ",[23,10443,10444],{},"small, faint dwarf galaxies",". The textbook is being rewritten.",[19,10447,10448],{},"To make sense of all this, three big theories are now seriously on the table. In plain English:",[122,10450,10452],{"id":10451},"theory-1-the-universe-is-recycled-magnetized-memory-bouncing-universe","Theory 1: The universe is recycled? — \"Magnetized Memory Bouncing Universe\"",[19,10454,10455,10456,10459],{},"What if the Big Bang wasn't really the beginning, and our universe ",[23,10457,10458],{},"bounced from a previous one, inheriting its magnetic field memory","?",[312,10461,10462,10465,10468],{},[315,10463,10464],{},"Why it's cool: the universe might be a \"second-hand\" cosmos",[315,10466,10467],{},"Weakness: the bounce mechanism itself still needs more theory work",[315,10469,10470],{},"Test: a \"twisted\" pattern in the CMB polarization would be smoking-gun evidence",[122,10472,10474],{"id":10473},"theory-2-a-lump-in-the-early-universe-non-abelian-clumpy-reheating-universe","Theory 2: A \"lump\" in the early universe? — \"Non-Abelian Clumpy Reheating Universe\"",[19,10476,10477,10478,10481],{},"In a tiny patch of the very early universe, ",[23,10479,10480],{},"fluctuations got concentrated"," and made everything later go faster — explaining both the early prodigy galaxies and a possible far-future vacuum collapse.",[312,10483,10484,10487,10490],{},[315,10485,10486],{},"Why it's cool: JWST's \"too-fast\" galaxies fall right into place",[315,10488,10489],{},"Weakness: it risks producing too many primordial black holes — close to current limits",[315,10491,10492],{},"Test: a narrow \"bump\" of gravitational waves at a specific frequency",[122,10494,10496],{"id":10495},"theory-3-dark-matter-and-dark-energy-whispering-to-each-other-interacting-dark-sector-slow-ending-universe","Theory 3: Dark matter and dark energy whispering to each other? — \"Interacting Dark Sector, Slow Ending Universe\"",[19,10498,10499,10500,10503],{},"What if ",[23,10501,10502],{},"dark matter and dark energy are quietly exchanging energy",", and that's why the recent universe behaves slightly off-script?",[312,10505,10506,10509,10512],{},[315,10507,10508],{},"Why it's cool: solves both H0 tension and the \"growth-of-structure\" puzzle (S8) in one move",[315,10510,10511],{},"Weakness: doesn't say much about the universe's actual birth",[315,10513,10514,10515],{},"Ending: not Big Rip, not heat death — the universe just ",[23,10516,10517],{},"goes quiet, with structure growth frozen and only a faint \"dark acoustic\" sound left behind",[19,10519,10520,10521,10524],{},"Theory 3 is currently the closest fit to the data. And honestly — ",[23,10522,10523],{},"\"the universe ends not with a bang or a freeze, but in silence, with only a faint dark sound\""," — that's poetic.",[205,10526],{},[32,10528,10530],{"id":10529},"the-answer-might-come-in-our-lifetime","The answer might come in our lifetime",[19,10532,10533],{},"Here's the best part.",[19,10535,10536,10537,10540,10541,10543,10544,10338,10547,10550,10551,10338,10554,10557,10558,10560,10561,10563],{},"These theories will likely be ",[23,10538,10539],{},"decided in the next 5–10 years",". Europe's ",[23,10542,10246],{},", the US's ",[23,10545,10546],{},"Rubin",[23,10548,10549],{},"Roman"," observatories, the CMB-focused ",[23,10552,10553],{},"CMB-S4",[23,10555,10556],{},"LiteBIRD",", the radio telescope ",[23,10559,10258],{},", and the space gravitational-wave detector ",[23,10562,10262],{}," — they're all coming online roughly at once.",[19,10565,10566,10567,10570],{},"In other words: ",[23,10568,10569],{},"we'll likely live to see at least the direction of the answer to \"how does the universe end?\""," That's a first in human history.",[19,10572,10573],{},"So next time you scroll past a \"physics is broken\" headline, you'll know what's really going on: three theories slugging it out, and the entire planet's instruments racing to break the tie.",[19,10575,10576,10577,10580],{},"If this clicked for you, ",[23,10578,10579],{},"share it on X"," and drop a comment with \"which theory is your favorite.\" That's how I decide which deep dive to write next.",[205,10582],{},[19,10584,10585],{},[23,10586,7597],{},[312,10588,10589,10595,10601],{},[315,10590,10591,10594],{},[23,10592,10593],{},"Idea 1: \"What Even Is Dark Matter? 3 Candidate Theories Compared in Plain English\""," — A side-by-side breakdown of WIMPs, axions, and primordial black holes — strengths, weaknesses, and how much the field has actually narrowed down. A natural deep dive on the \"dark sector whispering\" angle from this post.",[315,10596,10597,10600],{},[23,10598,10599],{},"Idea 2: \"5 'Impossible Galaxies' JWST Has Found\""," — The 5 most striking real-world examples of \"too-fast\" galaxies, with images and a one-line \"why this breaks the textbook\" each. Visual-first follow-up to today's post.",[315,10602,10603,10606],{},[23,10604,10605],{},"Idea 3: \"How the Universe Could End, Visualized: Big Rip vs Heat Death vs the Quiet Ending\""," — Each of today's three theories implies a different final state. A timeline-style visual explainer that makes readers pick a favorite.",{"title":143,"searchDepth":1605,"depth":1605,"links":10608},[10609,10610,10611,10616,10617,10618,10619,10624],{"id":10027,"depth":1605,"text":10028},{"id":10053,"depth":1605,"text":10054},{"id":10123,"depth":1605,"text":10124,"children":10612},[10613,10614,10615],{"id":10155,"depth":1610,"text":10156},{"id":10177,"depth":1610,"text":10178},{"id":10199,"depth":1610,"text":10200},{"id":10233,"depth":1605,"text":10233},{"id":10324,"depth":1605,"text":10325},{"id":10350,"depth":1605,"text":10351},{"id":10415,"depth":1605,"text":10416,"children":10620},[10621,10622,10623],{"id":10451,"depth":1610,"text":10452},{"id":10473,"depth":1610,"text":10474},{"id":10495,"depth":1610,"text":10496},{"id":10529,"depth":1605,"text":10530},"「宇宙はビッグバンで始まってずっと膨張してる」って習ったよね。でも最新観測でその常識にヒビが入り始めてる。宇宙の終わりを巡る3つのワクワクする仮説を、文系でも分かる言葉で。",{"date":10627,"image":10628,"alt":10629,"tags":10630,"tagsEn":10635,"published":1666},"5th May 2026","/blogs-img/blog-cosmology-frontier-2026.png","宇宙の終焉と最新宇宙論の3仮説を解説するサムネイル画像",[10631,10632,10633,7644,10634],"宇宙","サイエンス","雑学","体験談",[10636,10637,10638,7649,10639],"Cosmology","Science","Trivia","Personal Experience","/blogs/14-cosmology-frontier-2026",{"title":10013,"description":10625},"blogs/14-cosmology-frontier-2026","Eujlt2R-AobYdJnMH0pjzsVGl1rIIvgm-HCpDsIDEsI",{"id":4,"title":5,"body":10645,"description":1650,"extension":1651,"meta":11831,"navigation":1666,"ogImage":1654,"path":1667,"seo":11834,"stem":1669,"__hash__":1670},{"type":7,"value":10646,"toc":11787},[10647,11209,11211],[10,10648,10649,10651,10657,10659,10661,10665,10669,10717,10719,10723,10728,10730,10732,10774,10776,10778,10780,10782,10836,10838,10844,10846,10848,10862,10864,10866,10868,10870,10934,10936,11012,11014,11019,11025,11027,11029,11047,11049,11051,11055,11057,11127,11129,11133,11135,11137,11139,11143,11145,11147,11165,11167,11169,11171,11173,11177,11179,11181,11199,11203,11205],{"lang":12},[14,10650,17],{"id":16},[19,10652,21,10653,30],{},[23,10654,25,10655,29],{},[23,10656,28],{},[32,10658,35],{"id":34},[19,10660,38],{},[19,10662,41,10663,45],{},[23,10664,44],{},[19,10666,48,10667,52],{},[23,10668,51],{},[54,10670,10671,10679],{},[57,10672,10673],{},[60,10674,10675,10677],{},[63,10676,65],{},[63,10678,68],{},[70,10680,10681,10687,10693,10699,10705,10711],{},[60,10682,10683,10685],{},[75,10684,77],{},[75,10686,80],{},[60,10688,10689,10691],{},[75,10690,85],{},[75,10692,88],{},[60,10694,10695,10697],{},[75,10696,93],{},[75,10698,96],{},[60,10700,10701,10703],{},[75,10702,101],{},[75,10704,104],{},[60,10706,10707,10709],{},[75,10708,109],{},[75,10710,112],{},[60,10712,10713,10715],{},[75,10714,117],{},[75,10716,120],{},[122,10718,125],{"id":124},[19,10720,128,10721,132],{},[23,10722,131],{},[134,10724,10726],{"className":10725,"code":138,"language":139},[137],[141,10727,138],{"__ignoreMap":143},[19,10729,146],{},[122,10731,149],{"id":149},[54,10733,10734,10742],{},[57,10735,10736],{},[60,10737,10738,10740],{},[63,10739,158],{},[63,10741,161],{},[70,10743,10744,10750,10756,10762,10768],{},[60,10745,10746,10748],{},[75,10747,168],{},[75,10749,171],{},[60,10751,10752,10754],{},[75,10753,176],{},[75,10755,179],{},[60,10757,10758,10760],{},[75,10759,184],{},[75,10761,187],{},[60,10763,10764,10766],{},[75,10765,192],{},[75,10767,195],{},[60,10769,10770,10772],{},[75,10771,200],{},[75,10773,203],{},[205,10775],{},[32,10777,210],{"id":209},[19,10779,213],{},[122,10781,216],{"id":216},[54,10783,10784,10798],{},[57,10785,10786],{},[60,10787,10788,10790,10792,10794,10796],{},[63,10789,225],{},[63,10791,228],{},[63,10793,231],{},[63,10795,234],{},[63,10797,237],{},[70,10799,10800,10812,10824],{},[60,10801,10802,10804,10806,10808,10810],{},[75,10803,244],{},[75,10805,247],{},[75,10807,250],{},[75,10809,253],{},[75,10811,256],{},[60,10813,10814,10816,10818,10820,10822],{},[75,10815,261],{},[75,10817,264],{},[75,10819,267],{},[75,10821,250],{},[75,10823,272],{},[60,10825,10826,10828,10830,10832,10834],{},[75,10827,277],{},[75,10829,280],{},[75,10831,283],{},[75,10833,286],{},[75,10835,289],{},[122,10837,292],{"id":292},[19,10839,295,10840,299,10842,303],{},[23,10841,298],{},[23,10843,302],{},[122,10845,307],{"id":306},[19,10847,310],{},[312,10849,10850,10854,10858],{},[315,10851,10852,320],{},[23,10853,319],{},[315,10855,10856],{},[23,10857,325],{},[315,10859,10860],{},[23,10861,330],{},[19,10863,333],{},[205,10865],{},[32,10867,338],{"id":338},[19,10869,341],{},[54,10871,10872,10882],{},[57,10873,10874],{},[60,10875,10876,10878,10880],{},[63,10877,350],{},[63,10879,68],{},[63,10881,355],{},[70,10883,10884,10894,10904,10914,10924],{},[60,10885,10886,10890,10892],{},[75,10887,10888],{},[23,10889,364],{},[75,10891,367],{},[75,10893,370],{},[60,10895,10896,10900,10902],{},[75,10897,10898],{},[23,10899,377],{},[75,10901,380],{},[75,10903,383],{},[60,10905,10906,10910,10912],{},[75,10907,10908],{},[23,10909,390],{},[75,10911,393],{},[75,10913,396],{},[60,10915,10916,10920,10922],{},[75,10917,10918],{},[23,10919,403],{},[75,10921,406],{},[75,10923,409],{},[60,10925,10926,10930,10932],{},[75,10927,10928],{},[23,10929,416],{},[75,10931,419],{},[75,10933,422],{},[122,10935,425],{"id":425},[54,10937,10938,10950],{},[57,10939,10940],{},[60,10941,10942,10944,10946,10948],{},[63,10943,434],{},[63,10945,437],{},[63,10947,109],{},[63,10949,442],{},[70,10951,10952,10964,10976,10988,11000],{},[60,10953,10954,10958,10960,10962],{},[75,10955,10956],{},[23,10957,451],{},[75,10959,454],{},[75,10961,457],{},[75,10963,460],{},[60,10965,10966,10970,10972,10974],{},[75,10967,10968],{},[23,10969,467],{},[75,10971,470],{},[75,10973,473],{},[75,10975,476],{},[60,10977,10978,10982,10984,10986],{},[75,10979,10980],{},[23,10981,483],{},[75,10983,486],{},[75,10985,489],{},[75,10987,492],{},[60,10989,10990,10994,10996,10998],{},[75,10991,10992],{},[23,10993,499],{},[75,10995,502],{},[75,10997,505],{},[75,10999,508],{},[60,11001,11002,11006,11008,11010],{},[75,11003,11004],{},[23,11005,515],{},[75,11007,518],{},[75,11009,489],{},[75,11011,523],{},[122,11013,527],{"id":526},[134,11015,11017],{"className":11016,"code":531,"language":139},[137],[141,11018,531],{"__ignoreMap":143},[19,11020,536,11021,544],{},[23,11022,539,11023,543],{},[23,11024,542],{},[122,11026,547],{"id":547},[19,11028,550],{},[552,11030,11031,11035,11039,11043],{},[315,11032,11033,559],{},[23,11034,558],{},[315,11036,11037,565],{},[23,11038,564],{},[315,11040,11041,571],{},[23,11042,570],{},[315,11044,11045,577],{},[23,11046,576],{},[205,11048],{},[32,11050,582],{"id":582},[19,11052,585,11053,589],{},[23,11054,588],{},[122,11056,592],{"id":592},[54,11058,11059,11077],{},[57,11060,11061],{},[60,11062,11063,11065,11067,11069,11071,11073,11075],{},[63,11064,601],{},[63,11066,604],{},[63,11068,364],{},[63,11070,377],{},[63,11072,390],{},[63,11074,403],{},[63,11076,416],{},[70,11078,11079,11095,11111],{},[60,11080,11081,11083,11085,11087,11089,11091,11093],{},[75,11082,621],{},[75,11084,624],{},[75,11086,627],{},[75,11088,630],{},[75,11090,633],{},[75,11092,627],{},[75,11094,638],{},[60,11096,11097,11099,11101,11103,11105,11107,11109],{},[75,11098,643],{},[75,11100,646],{},[75,11102,627],{},[75,11104,630],{},[75,11106,630],{},[75,11108,627],{},[75,11110,657],{},[60,11112,11113,11115,11117,11119,11121,11123,11125],{},[75,11114,662],{},[75,11116,665],{},[75,11118,668],{},[75,11120,638],{},[75,11122,668],{},[75,11124,675],{},[75,11126,638],{},[679,11128,682],{"id":681},[19,11130,685,11131,689],{},[23,11132,688],{},[679,11134,693],{"id":692},[19,11136,696],{},[679,11138,700],{"id":699},[19,11140,703,11141,707],{},[23,11142,706],{},[122,11144,710],{"id":710},[19,11146,713],{},[312,11148,11149,11153,11157,11161],{},[315,11150,11151,721],{},[23,11152,720],{},[315,11154,11155,727],{},[23,11156,726],{},[315,11158,11159,733],{},[23,11160,732],{},[315,11162,11163,739],{},[23,11164,738],{},[19,11166,742],{},[205,11168],{},[32,11170,748],{"id":747},[19,11172,751],{},[19,11174,754,11175,758],{},[23,11176,757],{},[19,11178,761],{},[122,11180,764],{"id":764},[312,11182,11183,11187,11191,11195],{},[315,11184,11185,772],{},[23,11186,771],{},[315,11188,11189,778],{},[23,11190,777],{},[315,11192,11193,784],{},[23,11194,783],{},[315,11196,11197,790],{},[23,11198,789],{},[792,11200,11201],{},[19,11202,796],{},[205,11204],{},[19,11206,11207],{},[802,11208,804],{},[205,11210],{},[10,11212,11213,11215,11221,11223,11225,11229,11235,11283,11285,11291,11296,11298,11300,11342,11344,11346,11348,11350,11404,11406,11414,11416,11418,11432,11434,11436,11438,11442,11506,11508,11584,11586,11591,11595,11597,11599,11617,11619,11621,11625,11627,11697,11699,11703,11705,11709,11711,11715,11717,11735,11739,11741,11743,11749,11753,11757,11759,11777,11781,11783],{"lang":809},[14,11214,813],{"id":812},[19,11216,816,11217,820,11219,824],{},[23,11218,819],{},[23,11220,823],{},[32,11222,828],{"id":827},[19,11224,831],{},[19,11226,834,11227,838],{},[23,11228,837],{},[19,11230,841,11231,845,11233,849],{},[23,11232,844],{},[23,11234,848],{},[54,11236,11237,11245],{},[57,11238,11239],{},[60,11240,11241,11243],{},[63,11242,858],{},[63,11244,861],{},[70,11246,11247,11253,11259,11265,11271,11277],{},[60,11248,11249,11251],{},[75,11250,868],{},[75,11252,871],{},[60,11254,11255,11257],{},[75,11256,876],{},[75,11258,879],{},[60,11260,11261,11263],{},[75,11262,884],{},[75,11264,887],{},[60,11266,11267,11269],{},[75,11268,892],{},[75,11270,895],{},[60,11272,11273,11275],{},[75,11274,900],{},[75,11276,903],{},[60,11278,11279,11281],{},[75,11280,908],{},[75,11282,911],{},[122,11284,915],{"id":914},[19,11286,918,11287,922,11289,926],{},[23,11288,921],{},[23,11290,925],{},[134,11292,11294],{"className":11293,"code":930,"language":139},[137],[141,11295,930],{"__ignoreMap":143},[19,11297,935],{},[122,11299,939],{"id":938},[54,11301,11302,11310],{},[57,11303,11304],{},[60,11305,11306,11308],{},[63,11307,948],{},[63,11309,951],{},[70,11311,11312,11318,11324,11330,11336],{},[60,11313,11314,11316],{},[75,11315,958],{},[75,11317,961],{},[60,11319,11320,11322],{},[75,11321,966],{},[75,11323,969],{},[60,11325,11326,11328],{},[75,11327,974],{},[75,11329,977],{},[60,11331,11332,11334],{},[75,11333,982],{},[75,11335,985],{},[60,11337,11338,11340],{},[75,11339,990],{},[75,11341,993],{},[205,11343],{},[32,11345,999],{"id":998},[19,11347,1002],{},[122,11349,1006],{"id":1005},[54,11351,11352,11366],{},[57,11353,11354],{},[60,11355,11356,11358,11360,11362,11364],{},[63,11357,1015],{},[63,11359,1018],{},[63,11361,1021],{},[63,11363,1024],{},[63,11365,237],{},[70,11367,11368,11380,11392],{},[60,11369,11370,11372,11374,11376,11378],{},[75,11371,1033],{},[75,11373,1036],{},[75,11375,250],{},[75,11377,1041],{},[75,11379,256],{},[60,11381,11382,11384,11386,11388,11390],{},[75,11383,1048],{},[75,11385,1051],{},[75,11387,1054],{},[75,11389,250],{},[75,11391,272],{},[60,11393,11394,11396,11398,11400,11402],{},[75,11395,1063],{},[75,11397,1066],{},[75,11399,1069],{},[75,11401,1072],{},[75,11403,289],{},[122,11405,1078],{"id":1077},[19,11407,1081,11408,1085,11410,1089,11412,1093],{},[23,11409,1084],{},[23,11411,1088],{},[23,11413,1092],{},[122,11415,1097],{"id":1096},[19,11417,1100],{},[312,11419,11420,11424,11428],{},[315,11421,11422,1108],{},[23,11423,1107],{},[315,11425,11426],{},[23,11427,1113],{},[315,11429,11430],{},[23,11431,1118],{},[19,11433,1121],{},[205,11435],{},[32,11437,1127],{"id":1126},[19,11439,1130,11440,1134],{},[23,11441,1133],{},[54,11443,11444,11454],{},[57,11445,11446],{},[60,11447,11448,11450,11452],{},[63,11449,1143],{},[63,11451,1146],{},[63,11453,951],{},[70,11455,11456,11466,11476,11486,11496],{},[60,11457,11458,11462,11464],{},[75,11459,11460],{},[23,11461,1157],{},[75,11463,1160],{},[75,11465,1163],{},[60,11467,11468,11472,11474],{},[75,11469,11470],{},[23,11471,1170],{},[75,11473,1173],{},[75,11475,1176],{},[60,11477,11478,11482,11484],{},[75,11479,11480],{},[23,11481,1183],{},[75,11483,1186],{},[75,11485,1189],{},[60,11487,11488,11492,11494],{},[75,11489,11490],{},[23,11491,1196],{},[75,11493,1199],{},[75,11495,1202],{},[60,11497,11498,11502,11504],{},[75,11499,11500],{},[23,11501,1209],{},[75,11503,1212],{},[75,11505,1215],{},[122,11507,1219],{"id":1218},[54,11509,11510,11522],{},[57,11511,11512],{},[60,11513,11514,11516,11518,11520],{},[63,11515,1228],{},[63,11517,1231],{},[63,11519,900],{},[63,11521,1236],{},[70,11523,11524,11536,11548,11560,11572],{},[60,11525,11526,11530,11532,11534],{},[75,11527,11528],{},[23,11529,451],{},[75,11531,1247],{},[75,11533,1250],{},[75,11535,1253],{},[60,11537,11538,11542,11544,11546],{},[75,11539,11540],{},[23,11541,1260],{},[75,11543,1263],{},[75,11545,1266],{},[75,11547,1269],{},[60,11549,11550,11554,11556,11558],{},[75,11551,11552],{},[23,11553,483],{},[75,11555,1278],{},[75,11557,1281],{},[75,11559,1284],{},[60,11561,11562,11566,11568,11570],{},[75,11563,11564],{},[23,11565,499],{},[75,11567,1293],{},[75,11569,1296],{},[75,11571,1299],{},[60,11573,11574,11578,11580,11582],{},[75,11575,11576],{},[23,11577,515],{},[75,11579,1308],{},[75,11581,1281],{},[75,11583,1313],{},[122,11585,1317],{"id":1316},[134,11587,11589],{"className":11588,"code":1321,"language":139},[137],[141,11590,1321],{"__ignoreMap":143},[19,11592,1326,11593,1330],{},[23,11594,1329],{},[122,11596,1334],{"id":1333},[19,11598,1337],{},[552,11600,11601,11605,11609,11613],{},[315,11602,11603,1345],{},[23,11604,1344],{},[315,11606,11607,1351],{},[23,11608,1350],{},[315,11610,11611,1357],{},[23,11612,1356],{},[315,11614,11615,1363],{},[23,11616,1362],{},[205,11618],{},[32,11620,1369],{"id":1368},[19,11622,1372,11623,1376],{},[23,11624,1375],{},[122,11626,1380],{"id":1379},[54,11628,11629,11647],{},[57,11630,11631],{},[60,11632,11633,11635,11637,11639,11641,11643,11645],{},[63,11634,1389],{},[63,11636,1392],{},[63,11638,1157],{},[63,11640,1170],{},[63,11642,1399],{},[63,11644,1196],{},[63,11646,1209],{},[70,11648,11649,11665,11681],{},[60,11650,11651,11653,11655,11657,11659,11661,11663],{},[75,11652,1410],{},[75,11654,624],{},[75,11656,1415],{},[75,11658,1418],{},[75,11660,1421],{},[75,11662,1415],{},[75,11664,1426],{},[60,11666,11667,11669,11671,11673,11675,11677,11679],{},[75,11668,1431],{},[75,11670,646],{},[75,11672,1415],{},[75,11674,1418],{},[75,11676,1418],{},[75,11678,1415],{},[75,11680,1444],{},[60,11682,11683,11685,11687,11689,11691,11693,11695],{},[75,11684,1449],{},[75,11686,665],{},[75,11688,1454],{},[75,11690,1426],{},[75,11692,1454],{},[75,11694,1461],{},[75,11696,1426],{},[679,11698,1467],{"id":1466},[19,11700,1470,11701,1474],{},[23,11702,1473],{},[679,11704,1478],{"id":1477},[19,11706,1481,11707,1485],{},[23,11708,1484],{},[679,11710,1489],{"id":1488},[19,11712,1492,11713,1496],{},[23,11714,1495],{},[122,11716,1500],{"id":1499},[312,11718,11719,11723,11727,11731],{},[315,11720,11721,1508],{},[23,11722,1507],{},[315,11724,11725,1514],{},[23,11726,1513],{},[315,11728,11729,1520],{},[23,11730,1519],{},[315,11732,11733,1526],{},[23,11734,1525],{},[19,11736,1529,11737,1533],{},[23,11738,1532],{},[205,11740],{},[32,11742,1539],{"id":1538},[19,11744,1542,11745,1546,11747,849],{},[23,11746,1545],{},[23,11748,1549],{},[19,11750,1552,11751,849],{},[23,11752,1555],{},[19,11754,1558,11755,849],{},[23,11756,1561],{},[122,11758,1565],{"id":1564},[312,11760,11761,11765,11769,11773],{},[315,11762,11763,1573],{},[23,11764,1572],{},[315,11766,11767,1579],{},[23,11768,1578],{},[315,11770,11771,1585],{},[23,11772,1584],{},[315,11774,11775,1591],{},[23,11776,1590],{},[792,11778,11779],{},[19,11780,1596],{},[205,11782],{},[19,11784,11785],{},[802,11786,1603],{},{"title":143,"searchDepth":1605,"depth":1605,"links":11788},[11789,11793,11798,11803,11807,11810,11814,11819,11824,11828],{"id":34,"depth":1605,"text":35,"children":11790},[11791,11792],{"id":124,"depth":1610,"text":125},{"id":149,"depth":1610,"text":149},{"id":209,"depth":1605,"text":210,"children":11794},[11795,11796,11797],{"id":216,"depth":1610,"text":216},{"id":292,"depth":1610,"text":292},{"id":306,"depth":1610,"text":307},{"id":338,"depth":1605,"text":338,"children":11799},[11800,11801,11802],{"id":425,"depth":1610,"text":425},{"id":526,"depth":1610,"text":527},{"id":547,"depth":1610,"text":547},{"id":582,"depth":1605,"text":582,"children":11804},[11805,11806],{"id":592,"depth":1610,"text":592},{"id":710,"depth":1610,"text":710},{"id":747,"depth":1605,"text":748,"children":11808},[11809],{"id":764,"depth":1610,"text":764},{"id":827,"depth":1605,"text":828,"children":11811},[11812,11813],{"id":914,"depth":1610,"text":915},{"id":938,"depth":1610,"text":939},{"id":998,"depth":1605,"text":999,"children":11815},[11816,11817,11818],{"id":1005,"depth":1610,"text":1006},{"id":1077,"depth":1610,"text":1078},{"id":1096,"depth":1610,"text":1097},{"id":1126,"depth":1605,"text":1127,"children":11820},[11821,11822,11823],{"id":1218,"depth":1610,"text":1219},{"id":1316,"depth":1610,"text":1317},{"id":1333,"depth":1610,"text":1334},{"id":1368,"depth":1605,"text":1369,"children":11825},[11826,11827],{"id":1379,"depth":1610,"text":1380},{"id":1499,"depth":1610,"text":1500},{"id":1538,"depth":1605,"text":1539,"children":11829},[11830],{"id":1564,"depth":1610,"text":1565},{"date":1653,"image":1654,"alt":1655,"tags":11832,"tagsEn":11833,"published":1666},[44,1657,1658,1659,1660],[837,1662,1663,1664,1665],{"title":5,"description":1650},{"id":11836,"title":11837,"body":11838,"description":13433,"extension":1651,"meta":13434,"navigation":1666,"ogImage":13436,"path":13441,"seo":13442,"stem":13443,"__hash__":13444},"content/blogs/3-iran-coal-stocks.md","イラン戦争が呼び起こす石炭株の「逆回転」 | Iran War Sparks a Reversal in Coal Stocks",{"type":7,"value":11839,"toc":13389},[11840,12608,12610],[10,11841,11842,11846,11853,11856,11859,11862,11865,11869,11965,11972,11976,11979,11994,11997,11999,12002,12009,12013,12024,12027,12031,12038,12042,12053,12143,12147,12254,12261,12263,12266,12269,12273,12276,12283,12287,12294,12297,12303,12309,12315,12447,12454,12456,12459,12466,12468,12529,12533,12540,12547,12551,12558,12561,12563,12595,12602,12604],{"lang":12},[14,11843,11845],{"id":11844},"イラン戦争が呼び起こす石炭株の逆回転","イラン戦争が呼び起こす石炭株の「逆回転」",[19,11847,11848,11849,11852],{},"私はCVCで海洋・LNG領域を担当しており、EU ETS・THETIS-MRVのフリートデータを独自分析して投資提案に落とし込んできた人間です。LNG船の用船料、欧州ガス貯蔵、アジアの石炭価格——この3つは普段から同じダッシュボードで眺めているのですが、2026年3月の動きはここ数年で最も「全部が同時に動いた」週でした。だからこそ、感情ではなく",[23,11850,11851],{},"運賃と備蓄と契約","の3点で、今回の石炭株「逆回転」を整理してみたいと思っています。",[32,11854,11855],{"id":11855},"ホルムズ海峡封鎖がもたらしたエネルギーの断層",[19,11857,11858],{},"最初に、相場の前提になっている事実関係をフラットに置いておきます。",[19,11860,11861],{},"2026年2月28日、米国・イスラエル連合軍はイランに対する大規模空爆「Operation Epic Fury」を開始しました。最高指導者ハメネイ師を含む政権中枢が複数殺害され、イランは報復としてホルムズ海峡を事実上の封鎖状態に。3月2日以降、商業船舶の通航はほぼ停止し、約2,000隻・2万人の船員が海峡で足止めされています。",[19,11863,11864],{},"この海峡は、世界の石油輸送量の約20%（日量1,780万バレル）と、カタール・UAEからのLNG輸出（世界全体の約20%）が通過する地球規模のチョークポイント。IEAのビロル事務局長はこの事態を**「史上最大のエネルギー安全保障上の危機」**と表現していて、私の海洋データの感覚から言っても、これはレトリックではないと思います。",[122,11866,11868],{"id":11867},"原油ガス価格の急騰","原油・ガス価格の急騰",[54,11870,11871,11886],{},[57,11872,11873],{},[60,11874,11875,11877,11880,11883],{},[63,11876,225],{},[63,11878,11879],{},"開戦前（2/27）",[63,11881,11882],{},"直近（3/28）",[63,11884,11885],{},"変動率",[70,11887,11888,11904,11920,11935,11951],{},[60,11889,11890,11893,11896,11899],{},[75,11891,11892],{},"Brent原油",[75,11894,11895],{},"$73/bbl",[75,11897,11898],{},"$112.57/bbl",[75,11900,11901],{},[23,11902,11903],{},"+54%",[60,11905,11906,11909,11912,11915],{},[75,11907,11908],{},"WTI原油",[75,11910,11911],{},"$68/bbl",[75,11913,11914],{},"$99.64/bbl",[75,11916,11917],{},[23,11918,11919],{},"+47%",[60,11921,11922,11925,11928,11931],{},[75,11923,11924],{},"Newcastle石炭",[75,11926,11927],{},"$116/t",[75,11929,11930],{},"$141.65/t",[75,11932,11933],{},[23,11934,3066],{},[60,11936,11937,11940,11943,11946],{},[75,11938,11939],{},"Dutch TTF ガス",[75,11941,11942],{},"~€32/MWh",[75,11944,11945],{},"€60超/MWh",[75,11947,11948],{},[23,11949,11950],{},"+88%",[60,11952,11953,11956,11959,11962],{},[75,11954,11955],{},"S&P 500",[75,11957,11958],{},"高値圏",[75,11960,11961],{},"高値比▲7%",[75,11963,11964],{},"調整局面",[19,11966,11967,11968,11971],{},"Brent原油は2月27日の$73水準から3月28日には**$112.57",[23,11969,11970],{},"まで急騰し、2022年7月以来の最高値を記録。Goldman Sachsは現在の価格に$14〜$18の地政学リスクプレミアムが乗っていると推定しています。3月2日にカタール防衛省がイランのドローンによるカタールのガス施設攻撃を発表し、QatarEnergyが全ガス生産を停止——Dutch TTFは","€60/MWh超**に跳ね上がりました。欧州のガス貯蔵は2026年2月末時点で460億立方メートルと、2025年の600bcm、2024年の770bcmを大幅に下回る歴史的低水準にあります。",[122,11973,11975],{"id":11974},"_2022年のロシアウクライナ侵攻との似て非なる構造","2022年のロシア・ウクライナ侵攻との「似て非なる構造」",[19,11977,11978],{},"ここは私が特に強調しておきたいポイントです。2022年のロシア・ウクライナ侵攻後のショックと表面的には似ていますが、構造はかなり違います。",[19,11980,11981,11982,11985,11986,11989,11990,11993],{},"2022年はロシアの輸出エネルギー全体の約5%が影響を受けたのに対し、ホルムズ海峡の閉鎖は",[23,11983,11984],{},"世界のエネルギーフローのはるかに大きな割合","を直接打撃しています。さらに、カタールのRas Laffanガス施設への物理的損傷は修復に",[23,11987,11988],{},"3〜5年","を要するとアナリストは見積もっており、推定",[23,11991,11992],{},"1,280万トン/年","のLNG供給がオフライン。停戦してもLNGの供給は戻らない、という構造的な含意があります。",[19,11995,11996],{},"CVC的な解釈で言うと、これは「価格ショック」というより「契約フォースマジュールの連鎖」で、長期供給契約の再交渉が走り出している、というのが私の見方です。THETIS-MRVのフリートパターンは、すでに迂回・係留待機の動きが鮮明になっています。",[205,11998],{},[32,12000,12001],{"id":12001},"アジアを覆う石炭回帰の波",[19,12003,12004,12005,12008],{},"LNGが消えた穴を埋める手段は、原子力・再エネ・",[23,12006,12007],{},"石炭","しか残りません。原子力は立ち上げに時間がかかり、再エネはグリッド整備が間に合わない。だから石炭への回帰が瞬間的に効きやすい——というのが、今アジアで起きていることです。",[122,12010,12012],{"id":12011},"韓国エネルギーモデルの根幹が揺らぐ","韓国：エネルギー・モデルの根幹が揺らぐ",[19,12014,12015,12016,12023],{},"韓国はエネルギー資源の",[23,12017,12018,12019,12022],{},"94%",[23,12020,12021],{},"を輸入に依存し、原油の70%、LNGの20%が中東経由。LNG輸入のうち約15%はカタール産で、Ras Laffan攻撃後カタールは長期ガス契約の一部にフォースマジュールを宣言。共に民主党は中東危機経済対策タスクフォースを設置し、即座に","石炭火力発電の稼働上限（80%キャップ）を完全撤廃","、原子力の稼働率を80%へ引き上げ、定期メンテを前倒し完了させると発表しました。",[19,12025,12026],{},"韓国の電力構成では石炭が約33%、原子力が約31%、LNG火力が約25%。LNGの代替が必要になった瞬間、消費を石炭と原子力に振るしか選択肢がない構造です。",[122,12028,12030],{"id":12029},"インド猛暑と需要の爆発","インド：猛暑と需要の爆発",[19,12032,12033,12034,12037],{},"インドは世界第2位の石炭消費・生産国で、夏場のピーク電力需要は",[23,12035,12036],{},"270GW","に達する見込み（スペインの総発電容量のほぼ2倍）。政府は石炭の増産を加速し、約3ヶ月分の備蓄を確保しています。",[122,12039,12041],{"id":12040},"インドネシア東南アジア輸出規制と価格転嫁","インドネシア・東南アジア：輸出規制と価格転嫁",[19,12043,12044,12045,12048,12049,12052],{},"世界最大の石炭輸出国であるインドネシアは、2026年の生産目標を約",[23,12046,12047],{},"6億トン","（前年8億トンから大幅削減）に設定し、輸出制限を強化。フィリピンはエネルギー供給の危機的低下を理由に",[23,12050,12051],{},"国家非常事態を宣言","。ベトナムはインドネシアからの供給が不安定化したため、米国・ラオスからの輸入を検討し始めました。",[54,12054,12055,12071],{},[57,12056,12057],{},[60,12058,12059,12062,12065,12068],{},[63,12060,12061],{},"国・地域",[63,12063,12064],{},"主な対応",[63,12066,12067],{},"石炭への依存度",[63,12069,12070],{},"特記事項",[70,12072,12073,12087,12101,12115,12129],{},[60,12074,12075,12078,12081,12084],{},[75,12076,12077],{},"韓国",[75,12079,12080],{},"80%稼働上限撤廃、原子力80%へ",[75,12082,12083],{},"電力の33%",[75,12085,12086],{},"補正予算編成へ",[60,12088,12089,12092,12095,12098],{},[75,12090,12091],{},"インド",[75,12093,12094],{},"石炭増産加速",[75,12096,12097],{},"電力の70%超",[75,12099,12100],{},"夏場270GWピーク対応",[60,12102,12103,12106,12109,12112],{},[75,12104,12105],{},"インドネシア",[75,12107,12108],{},"国内優先、輸出制限",[75,12110,12111],{},"世界最大の輸出国",[75,12113,12114],{},"生産目標を8→6億トンに削減",[60,12116,12117,12120,12123,12126],{},[75,12118,12119],{},"フィリピン",[75,12121,12122],{},"国家非常事態宣言",[75,12124,12125],{},"約50%",[75,12127,12128],{},"LNG依存度が急速に上昇中だった",[60,12130,12131,12134,12137,12140],{},[75,12132,12133],{},"ベトナム",[75,12135,12136],{},"米国・ラオスからの輸入検討",[75,12138,12139],{},"約30%",[75,12141,12142],{},"インドネシア供給が不安定化",[122,12144,12146],{"id":12145},"newcastle石炭価格の軌跡","Newcastle石炭価格の軌跡",[54,12148,12149,12162],{},[57,12150,12151],{},[60,12152,12153,12156,12159],{},[63,12154,12155],{},"時期",[63,12157,12158],{},"Newcastle石炭（USD/t）",[63,12160,12161],{},"背景",[70,12163,12164,12175,12186,12197,12208,12219,12230,12241],{},[60,12165,12166,12169,12172],{},[75,12167,12168],{},"2026年1月下旬",[75,12170,12171],{},"~$111",[75,12173,12174],{},"中国の追加石炭火力100基稼働計画",[60,12176,12177,12180,12183],{},[75,12178,12179],{},"2月上旬",[75,12181,12182],{},"~$115",[75,12184,12185],{},"インドネシア減産報道",[60,12187,12188,12191,12194],{},[75,12189,12190],{},"2/24（開戦前）",[75,12192,12193],{},"~$116",[75,12195,12196],{},"米イラン外交交渉決裂の兆し",[60,12198,12199,12202,12205],{},[75,12200,12201],{},"3/3（開戦直後）",[75,12203,12204],{},"~$125",[75,12206,12207],{},"ホルムズ海峡封鎖",[60,12209,12210,12213,12216],{},[75,12211,12212],{},"3/6",[75,12214,12215],{},"$132",[75,12217,12218],{},"カタールLNG施設被弾",[60,12220,12221,12224,12227],{},[75,12222,12223],{},"3/17",[75,12225,12226],{},"$135",[75,12228,12229],{},"韓国が石炭上限撤廃",[60,12231,12232,12235,12238],{},[75,12233,12234],{},"3/24",[75,12236,12237],{},"$140",[75,12239,12240],{},"アジア各国の石炭回帰報道が集中",[60,12242,12243,12246,12251],{},[75,12244,12245],{},"3/27（直近）",[75,12247,12248],{},[23,12249,12250],{},"$141.65",[75,12252,12253],{},"52週高値更新",[19,12255,12256,12257,12260],{},"GMOのLucas White氏は、ホルムズ封鎖が数ヶ月続けば石炭価格がさらに",[23,12258,12259],{},"46%上昇","する可能性を指摘しています。",[205,12262],{},[32,12264,12265],{"id":12265},"注目すべき石炭関連銘柄",[19,12267,12268],{},"エネルギー危機の「最大受益者」の一角として、石炭関連株に改めて注目が集まっています。私がCVCのDD感覚で見たとき、ここは**「Seaborne比率」「冶金炭エクスポージャ」「契約vs スポット比率」**の3点が銘柄選定の本筋だと思っています。",[122,12270,12272],{"id":12271},"peabody-energynyse-btu","Peabody Energy（NYSE: BTU）",[19,12274,12275],{},"米国最大の民間石炭生産企業。米国内のPRB（Powder River Basin）とイリノイ盆地に加え、豪州NSWとクイーンズランドにSeaborne事業（海上出荷向け一般炭・原料炭）を展開していて、アジア需要増の恩恵を直接受ける構造です。3月24日の株価は**+7.9%**急騰し$38.19で引け。",[19,12277,12278,12279,12282],{},"特に注目しているのは",[23,12280,12281],{},"Centurion鉱山","。$750M投資の同鉱山は2026年2月にロングウォール生産を開始（当初予定から2ヶ月前倒し）し、2026年に350万トン、2028年までに470万トンのプレミアムHCC（Hard Coking Coal）出荷を計画。冶金炭は製鉄に不可欠で、一般炭よりも高単価・高マージンの成長ドライバーになります。2026年の出荷ガイダンスはSeaborne一般炭1,250万トン、冶金炭1,080万トン。手元キャッシュ$575M、流動性$900M超と財務体質も安定しています。",[122,12284,12286],{"id":12285},"whitehaven-coalasx-whcax","Whitehaven Coal（ASX: WHC.AX）",[19,12288,12289,12290,12293],{},"豪州NSWを拠点とし、高品質・低不純物の一般炭と原料炭を生産。販売先は日本、韓国、中国、インド、台湾、東南アジア、欧州と広い。Q2 2026では生産量が前四半期比",[23,12291,12292],{},"21%増の1,100万トン","、Narrabri鉱山は**+48%**増産。トン当たりコストはAUD135まで低下し、純負債もAUD1億削減。3月26日の株価はAUD8.915（52週高値AUD9.60圏）。",[122,12295,12296],{"id":12296},"その他の注目銘柄",[19,12298,12299,12302],{},[23,12300,12301],{},"Warrior Met Coal（NYSE: HCC）"," — アラバマ州でプレミアムHCCに特化する純粋な冶金炭プレイヤー。製鉄向け原料炭は一般炭とは異なる需給構造を持ち、アジア製鉄需要が底堅い限り高い価格実現力を維持できます。",[19,12304,12305,12308],{},[23,12306,12307],{},"Alliance Resource Partners（NASDAQ: ARLP）"," — イリノイ盆地・アパラチアで操業する石炭MLP。高配当（2025年は普通ユニットあたり$4.21の分配金）が特徴。",[19,12310,12311,12314],{},[23,12312,12313],{},"Natural Resource Partners（NYSE: NRP）"," — ロイヤリティ・モデル。自ら採掘せず鉱業権のリース料を受け取るので操業リスクが低い。2025年通年のFCFは$169M、負債を$109M返済と財務規律も堅い。",[54,12316,12317,12338],{},[57,12318,12319],{},[60,12320,12321,12324,12327,12330,12332,12335],{},[63,12322,12323],{},"Ticker",[63,12325,12326],{},"銘柄",[63,12328,12329],{},"市場",[63,12331,6119],{},[63,12333,12334],{},"事業特性",[63,12336,12337],{},"開戦後騰落率",[70,12339,12340,12362,12384,12404,12426],{},[60,12341,12342,12347,12350,12353,12356,12359],{},[75,12343,12344],{},[23,12345,12346],{},"BTU",[75,12348,12349],{},"Peabody Energy",[75,12351,12352],{},"NYSE",[75,12354,12355],{},"~$4.0B",[75,12357,12358],{},"米国最大手。豪州Seaborne＋PRB",[75,12360,12361],{},"+17%↑",[60,12363,12364,12369,12372,12375,12378,12381],{},[75,12365,12366],{},[23,12367,12368],{},"WHC.AX",[75,12370,12371],{},"Whitehaven Coal",[75,12373,12374],{},"ASX",[75,12376,12377],{},"~AUD9B",[75,12379,12380],{},"高品質NSW炭。アジア直販",[75,12382,12383],{},"+22%↑",[60,12385,12386,12391,12394,12396,12398,12401],{},[75,12387,12388],{},[23,12389,12390],{},"HCC",[75,12392,12393],{},"Warrior Met Coal",[75,12395,12352],{},[75,12397,1250],{},[75,12399,12400],{},"アラバマ産プレミアムHCC特化",[75,12402,12403],{},"+12%↑",[60,12405,12406,12411,12414,12417,12420,12423],{},[75,12407,12408],{},[23,12409,12410],{},"ARLP",[75,12412,12413],{},"Alliance Resource",[75,12415,12416],{},"NASDAQ",[75,12418,12419],{},"~$3.2B",[75,12421,12422],{},"高配当MLP。米国内需安定型",[75,12424,12425],{},"+8%↑",[60,12427,12428,12433,12436,12438,12441,12444],{},[75,12429,12430],{},[23,12431,12432],{},"NRP",[75,12434,12435],{},"Natural Resource Partners",[75,12437,12352],{},[75,12439,12440],{},"~$1.3B",[75,12442,12443],{},"ロイヤリティ・モデル。低リスク",[75,12445,12446],{},"+10%↑",[19,12448,12449,12450,12453],{},"米国内の一般炭需要はEIAの予測通り減少傾向にありますが、",[23,12451,12452],{},"冶金炭の輸出は2026年に増加見込み","。BTU（Centurion）、HCC、WHCがこの恩恵を最も直接的に享受する構図です。",[205,12455],{},[32,12457,12458],{"id":12458},"シナリオ別投資戦略と今後の展望",[19,12460,12461,12462,12465],{},"私の中ではここがいちばん大事なところで、石炭株は",[23,12463,12464],{},"タクティカル・トレード","として持つべきもの——というのが結論です。",[122,12467,4797],{"id":4797},[54,12469,12470,12485],{},[57,12471,12472],{},[60,12473,12474,12476,12479,12482],{},[63,12475,601],{},[63,12477,12478],{},"Newcastle石炭予想レンジ",[63,12480,12481],{},"Brent原油見通し",[63,12483,12484],{},"石炭株への含意",[70,12486,12487,12501,12515],{},[60,12488,12489,12492,12495,12498],{},[75,12490,12491],{},"ベース：4月中に停戦合意",[75,12493,12494],{},"$120〜$140/t",[75,12496,12497],{},"$80〜$90/bbl",[75,12499,12500],{},"現水準から利確の動き。急落リスクあり",[60,12502,12503,12506,12509,12512],{},[75,12504,12505],{},"メイン：紛争数ヶ月継続",[75,12507,12508],{},"$145〜$200/t",[75,12510,12511],{},"$100〜$120/bbl",[75,12513,12514],{},"BTU/WHCに追加上昇余地。+30〜40%",[60,12516,12517,12520,12523,12526],{},[75,12518,12519],{},"テール：年内解決せず",[75,12521,12522],{},"$180〜$260/t",[75,12524,12525],{},"$120〜$150/bbl",[75,12527,12528],{},"1970年代型エネルギー危機の再現。スタグフレーション",[122,12530,12532],{"id":12531},"短期エネルギーセクター内の相対ロング","短期：エネルギーセクター内の相対ロング",[19,12534,12535,12536,12539],{},"ただしマクロ環境には注意が必要です。Brent原油は$112を超え、S&P 500は高値から**▲7%",[23,12537,12538],{},"、Nasdaqは高値から","▲10%超**の調整入り。エネルギーインフレ長期化懸念から、中央銀行の利下げ期待は後退し、金利の「higher for longer」が現実味を帯びてきました。",[19,12541,12542,12543,12546],{},"このような環境下では、石炭株を市場全体に対するアウトライトのロングとして持つよりも、",[23,12544,12545],{},"エネルギーセクター内の相対ロング","——再エネ株や需要減退懸念のある業種のショートとペアにする戦略が筋が良いと見ています。CVCで投資委員会を経験した感覚では、こういう局面ほど「ヘッジしないネイキッドロング」が後で効いてきます。",[122,12548,12550],{"id":12549},"中長期逆説的な再エネシフトの触媒","中長期：逆説的な再エネシフトの触媒",[19,12552,12553,12554,12557],{},"逆説的ですが、この危機はまさに",[23,12555,12556],{},"再エネ・電化へのシフトを加速させる触媒","になりつつあります。IEAのビロル氏は「太陽光はもはやロマンチックな話ではなくビジネスだ」と発言。Ember社のリサーチマネージャーも「エレクトロテックにより各国は輸入燃料を完全に排除するツールを手にした」と指摘しています。",[19,12559,12560],{},"2022年のロシア・ウクライナ侵攻後、欧州のソーラー導入は倍増しました。それが金利上昇とともに減速した経緯がありますが、今回の危機は「エネルギー安全保障＝国産エネルギー＝再エネ」という再定義をさらに強固にします。私が普段見ているLNGメタンスリップ→海運脱炭素のコリドーも、同じ文脈で投資urgencyが上がる、というのが正直な感触です。",[122,12562,764],{"id":764},[312,12564,12565,12571,12577,12583,12589],{},[315,12566,12567,12570],{},[23,12568,12569],{},"4月6日のデッドライン"," — トランプ大統領がイランに求めたホルムズ海峡再開期限。期限切れ後の米国の対応が最大の焦点",[315,12572,12573,12576],{},[23,12574,12575],{},"カタールRas Laffan施設の復旧見通し"," — 修復3〜5年の見方が確定するか。LNG供給の長期構造を左右",[315,12578,12579,12582],{},[23,12580,12581],{},"IEA緊急備蓄放出の規模"," — IEA加盟国は4億バレルの戦略備蓄放出で合意済み。取り崩しペースが価格のダウンサイドを決める",[315,12584,12585,12588],{},[23,12586,12587],{},"中国の動向"," — 人民元建て「通行料」システムの拡大はエネルギーフロー分断のシグナル",[315,12590,12591,12594],{},[23,12592,12593],{},"インドネシアの輸出政策"," — 国内優先方針強化はNewcastle価格に直接効く",[19,12596,12597,12598,12601],{},"結論として、",[23,12599,12600],{},"短期ロング・中期ニュートラル","。石炭株はイラン危機の「タクティカル・トレード」であり、構造的な長期保有対象ではありません。しかし、この危機が浮き彫りにした「エネルギー輸入依存のリスク」は、分散型エネルギー・蓄電・船舶脱炭素の投資urgencyを確実に高めます。危機の中にこそ、次の構造変化への投資機会が潜んでいます。",[205,12603],{},[19,12605,12606],{},[802,12607,4907],{},[205,12609],{},[10,12611,12612,12616,12626,12630,12633,12636,12643,12647,12731,12742,12746,12752,12767,12774,12776,12780,12787,12791,12801,12804,12808,12815,12819,12830,12919,12923,13021,13028,13030,13034,13048,13052,13059,13066,13070,13081,13085,13091,13097,13103,13214,13225,13227,13231,13237,13240,13301,13305,13316,13323,13327,13334,13341,13343,13375,13382,13384],{"lang":809},[14,12613,12615],{"id":12614},"coal-stocks-reverse-rotation-triggered-by-the-iran-war","Coal Stocks' \"Reverse Rotation\" Triggered by the Iran War",[19,12617,12618,12619,12622,12623,849],{},"I cover maritime and LNG inside a corporate VC, building independent analyses from EU ETS, and THETIS-MRV fleet data and translating them into investment proposals. LNG charter rates, European gas storage, and Asian coal prices live in the same dashboard for me — and the week of late March 2026 is the first time in years that ",[23,12620,12621],{},"all three moved together"," at this magnitude. So in this piece I want to read the coal-stock \"reverse rotation\" not by the headlines, but by ",[23,12624,12625],{},"freight, storage, and contracts",[32,12627,12629],{"id":12628},"the-energy-fault-line-created-by-the-hormuz-blockade","The Energy Fault Line Created by the Hormuz Blockade",[19,12631,12632],{},"Let me first put the underlying facts down flat.",[19,12634,12635],{},"On February 28, 2026, US-Israeli coalition forces launched \"Operation Epic Fury,\" a large-scale strike on Iran. Multiple senior regime figures, including Supreme Leader Khamenei, were killed, and Iran retaliated by effectively blockading the Strait of Hormuz. Since March 2, commercial shipping has been largely halted; ~2,000 vessels and 20,000 crew are stranded.",[19,12637,12638,12639,12642],{},"The strait is a global energy chokepoint that handles roughly 20% of world oil shipments (17.8 mb/d) and ~20% of global LNG (Qatar + UAE). IEA Executive Director Birol called it ",[23,12640,12641],{},"\"the greatest energy security crisis in history\""," — and from where I sit watching the maritime data, that's not rhetoric.",[122,12644,12646],{"id":12645},"crude-and-gas-in-numbers","Crude and gas, in numbers",[54,12648,12649,12664],{},[57,12650,12651],{},[60,12652,12653,12655,12658,12661],{},[63,12654,3955],{},[63,12656,12657],{},"Pre-war (2/27)",[63,12659,12660],{},"Recent (3/28)",[63,12662,12663],{},"Change",[70,12665,12666,12679,12691,12704,12718],{},[60,12667,12668,12671,12673,12675],{},[75,12669,12670],{},"Brent crude",[75,12672,11895],{},[75,12674,11898],{},[75,12676,12677],{},[23,12678,11903],{},[60,12680,12681,12683,12685,12687],{},[75,12682,3979],{},[75,12684,11911],{},[75,12686,11914],{},[75,12688,12689],{},[23,12690,11919],{},[60,12692,12693,12696,12698,12700],{},[75,12694,12695],{},"Newcastle coal",[75,12697,11927],{},[75,12699,11930],{},[75,12701,12702],{},[23,12703,3066],{},[60,12705,12706,12709,12711,12714],{},[75,12707,12708],{},"Dutch TTF gas",[75,12710,11942],{},[75,12712,12713],{},">€60/MWh",[75,12715,12716],{},[23,12717,11950],{},[60,12719,12720,12722,12725,12728],{},[75,12721,11955],{},[75,12723,12724],{},"near highs",[75,12726,12727],{},"▲7% from high",[75,12729,12730],{},"correction",[19,12732,12733,12734,12737,12738,12741],{},"Brent rallied from ~$73 on Feb 27 to ",[23,12735,12736],{},"$112.57"," on Mar 28 — the highest since July 2022 (+51%+). Goldman estimates a $14–$18 geopolitical risk premium is in the price. On March 2, Qatar's defense ministry confirmed Iranian drone strikes on Qatari gas facilities; QatarEnergy halted all gas production. Dutch TTF surged past ",[23,12739,12740],{},"€60/MWh",". European gas storage at end-February stood at 46 bcm — well below 60 bcm in 2025 and 77 bcm in 2024.",[122,12743,12745],{"id":12744},"the-similar-but-different-structure-vs-the-2022-shock","The \"similar but different\" structure vs. the 2022 shock",[19,12747,12748,12749,849],{},"This is the framing I'd most want to highlight. The shape rhymes with 2022's Russia-Ukraine LNG crisis, but the ",[23,12750,12751],{},"structure is materially different",[19,12753,12754,12755,12758,12759,12762,12763,12766],{},"In 2022, ~5% of Russian energy exports were affected. The Hormuz closure hits a much larger share of ",[802,12756,12757],{},"global"," energy flows directly. On top of that, physical damage to Qatar's Ras Laffan facility is estimated to need ",[23,12760,12761],{},"3–5 years"," of repair, and an estimated ",[23,12764,12765],{},"12.8 mtpa"," of LNG supply is offline. Even after a ceasefire, LNG supply doesn't simply come back — that's the structural twist.",[19,12768,12769,12770,12773],{},"In CVC terms, this isn't a \"price shock\" so much as a ",[23,12771,12772],{},"cascade of contractual force majeures",". Long-term supply contracts are being renegotiated, and the THETIS-MRV fleet patterns I track are already showing diversion and offshore-waiting behavior.",[205,12775],{},[32,12777,12779],{"id":12778},"the-coal-return-wave-sweeping-asia","The Coal Return Wave Sweeping Asia",[19,12781,12782,12783,12786],{},"The hole left by missing LNG can only be filled by nuclear, renewables, or ",[23,12784,12785],{},"coal",". Nuclear takes too long to spin up. Renewables can't get grid integration done fast enough. So coal is the instantaneous lever — and that's exactly what's happening across Asia right now.",[122,12788,12790],{"id":12789},"korea-the-foundation-of-the-energy-model-is-shaking","Korea: the foundation of the energy model is shaking",[19,12792,12793,12794,12796,12797,12800],{},"Korea imports ",[23,12795,12018],{}," of its energy. ~70% of crude and ~20% of LNG transit the Middle East. Around 15% of LNG imports come from Qatar; after the Ras Laffan strike, Qatar declared force majeure on parts of its long-term gas contracts. The Democratic Party's Middle East Crisis Task Force responded by ",[23,12798,12799],{},"fully removing the 80% utilization cap on coal-fired generation"," and pulling forward reactor maintenance to push nuclear utilization to 80%.",[19,12802,12803],{},"Korea's power mix is roughly 33% coal, 31% nuclear, 25% LNG. The moment LNG needs to be replaced, coal and nuclear are the only structural levers.",[122,12805,12807],{"id":12806},"india-heat-plus-exploding-demand","India: heat plus exploding demand",[19,12809,12810,12811,12814],{},"India is the world's #2 coal consumer and producer. Summer peak power demand is projected to hit ",[23,12812,12813],{},"270 GW"," — nearly double Spain's total generation capacity. The government is accelerating coal output and holds ~3 months of inventory.",[122,12816,12818],{"id":12817},"indonesia-and-southeast-asia-export-controls-and-pass-through","Indonesia and Southeast Asia: export controls and pass-through",[19,12820,12821,12822,12825,12826,12829],{},"Indonesia, the world's largest coal exporter, has set its 2026 production target at ~",[23,12823,12824],{},"600 Mt"," (down sharply from ~800 Mt the prior year) and tightened export limits. The Philippines declared a ",[23,12827,12828],{},"state of national emergency"," citing the energy supply crisis. Vietnam, hit by unstable Indonesian supply, has begun looking at imports from the US and Laos.",[54,12831,12832,12847],{},[57,12833,12834],{},[60,12835,12836,12839,12842,12845],{},[63,12837,12838],{},"Country",[63,12840,12841],{},"Response",[63,12843,12844],{},"Coal share",[63,12846,9415],{},[70,12848,12849,12863,12877,12891,12905],{},[60,12850,12851,12854,12857,12860],{},[75,12852,12853],{},"Korea",[75,12855,12856],{},"80% cap removed; nuclear → 80%",[75,12858,12859],{},"33% of power",[75,12861,12862],{},"Supplementary budget on the way",[60,12864,12865,12868,12871,12874],{},[75,12866,12867],{},"India",[75,12869,12870],{},"Coal production accelerated",[75,12872,12873],{},">70% of power",[75,12875,12876],{},"Summer 270 GW peak",[60,12878,12879,12882,12885,12888],{},[75,12880,12881],{},"Indonesia",[75,12883,12884],{},"Domestic priority; export limits",[75,12886,12887],{},"World's largest exporter",[75,12889,12890],{},"Output cut from 800 → 600 Mt",[60,12892,12893,12896,12899,12902],{},[75,12894,12895],{},"Philippines",[75,12897,12898],{},"National emergency",[75,12900,12901],{},"~50%",[75,12903,12904],{},"LNG dependency had been rising fast",[60,12906,12907,12910,12913,12916],{},[75,12908,12909],{},"Vietnam",[75,12911,12912],{},"Eyeing US/Laos imports",[75,12914,12915],{},"~30%",[75,12917,12918],{},"Indonesian supply destabilizing",[122,12920,12922],{"id":12921},"newcastles-price-trajectory","Newcastle's price trajectory",[54,12924,12925,12937],{},[57,12926,12927],{},[60,12928,12929,12931,12934],{},[63,12930,3655],{},[63,12932,12933],{},"Newcastle ($/t)",[63,12935,12936],{},"Catalyst",[70,12938,12939,12949,12959,12969,12979,12989,12999,13009],{},[60,12940,12941,12944,12946],{},[75,12942,12943],{},"Late Jan 2026",[75,12945,12171],{},[75,12947,12948],{},"China's plan for 100 additional coal plants",[60,12950,12951,12954,12956],{},[75,12952,12953],{},"Early Feb",[75,12955,12182],{},[75,12957,12958],{},"Indonesia output cut headlines",[60,12960,12961,12964,12966],{},[75,12962,12963],{},"Feb 24 (pre-war)",[75,12965,12193],{},[75,12967,12968],{},"US-Iran diplomacy collapsing",[60,12970,12971,12974,12976],{},[75,12972,12973],{},"Mar 3 (war start)",[75,12975,12204],{},[75,12977,12978],{},"Hormuz blockade",[60,12980,12981,12984,12986],{},[75,12982,12983],{},"Mar 6",[75,12985,12215],{},[75,12987,12988],{},"Qatar LNG facility hit",[60,12990,12991,12994,12996],{},[75,12992,12993],{},"Mar 17",[75,12995,12226],{},[75,12997,12998],{},"Korea removes coal cap",[60,13000,13001,13004,13006],{},[75,13002,13003],{},"Mar 24",[75,13005,12237],{},[75,13007,13008],{},"Asia coal-return headlines stack",[60,13010,13011,13014,13018],{},[75,13012,13013],{},"Mar 27 (recent)",[75,13015,13016],{},[23,13017,12250],{},[75,13019,13020],{},"New 52-week high",[19,13022,13023,13024,13027],{},"GMO's Lucas White has flagged that a multi-month Hormuz closure could push coal another ",[23,13025,13026],{},"+46%"," higher.",[205,13029],{},[32,13031,13033],{"id":13032},"coal-names-worth-watching","Coal Names Worth Watching",[19,13035,13036,13037,8946,13040,13043,13044,13047],{},"This is where my CVC instincts cut hardest. Pick names by ",[23,13038,13039],{},"Seaborne exposure",[23,13041,13042],{},"metallurgical-coal share",", and ",[23,13045,13046],{},"contract vs. spot mix"," — those are the three that matter, in my view.",[122,13049,13051],{"id":13050},"peabody-energy-nyse-btu","Peabody Energy (NYSE: BTU)",[19,13053,13054,13055,13058],{},"America's largest private coal producer. PRB and Illinois Basin domestically, plus Seaborne thermal and metallurgical operations in NSW and Queensland — directly positioned for Asian demand. The stock jumped ",[23,13056,13057],{},"+7.9%"," on March 24 to close at $38.19.",[19,13060,13061,13062,13065],{},"The asset I'd most highlight is the ",[23,13063,13064],{},"Centurion mine"," — a $750M investment that started longwall production in February 2026 (two months ahead of plan), targeting 3.5 mt of premium HCC in 2026 and 4.7 mt by 2028. Met coal is a structural premium business with very different supply-demand dynamics from thermal coal. 2026 shipment guidance: 12.5 mt Seaborne thermal, 10.8 mt metallurgical. Cash on hand $575M; total liquidity $900M+.",[122,13067,13069],{"id":13068},"whitehaven-coal-asx-whcax","Whitehaven Coal (ASX: WHC.AX)",[19,13071,13072,13073,13076,13077,13080],{},"NSW-based, high-quality low-impurity thermal and met coal, sold across Japan, Korea, China, India, Taiwan, Southeast Asia, and Europe. Q2 2026 production rose ",[23,13074,13075],{},"+21% q/q to 11 mt",", with Narrabri up ",[23,13078,13079],{},"+48%",". Unit cost down to AUD 135/t, net debt down by AUD 100M. Stock at AUD 8.915 — top of the AUD 4.26–9.60 52-week range.",[122,13082,13084],{"id":13083},"other-names-worth-watching","Other names worth watching",[19,13086,13087,13090],{},[23,13088,13089],{},"Warrior Met Coal (NYSE: HCC)"," — Alabama-based pure-play premium HCC producer. Met coal has its own demand cycle, distinct from thermal, and pricing power holds as long as Asian steel demand stays bid.",[19,13092,13093,13096],{},[23,13094,13095],{},"Alliance Resource Partners (NASDAQ: ARLP)"," — Illinois Basin and Appalachia coal MLP. High-yield ($4.21 per common unit distributed in 2025) — built for income investors.",[19,13098,13099,13102],{},[23,13100,13101],{},"Natural Resource Partners (NYSE: NRP)"," — A royalty model. Doesn't mine; collects mineral rights lease income, so operating risk is low. 2025 free cash flow $169M; debt paid down $109M.",[54,13104,13105,13126],{},[57,13106,13107],{},[60,13108,13109,13111,13114,13117,13120,13123],{},[63,13110,12323],{},[63,13112,13113],{},"Name",[63,13115,13116],{},"Listing",[63,13118,13119],{},"Mkt cap",[63,13121,13122],{},"Profile",[63,13124,13125],{},"Post-war move",[70,13127,13128,13145,13163,13180,13197],{},[60,13129,13130,13134,13136,13138,13140,13143],{},[75,13131,13132],{},[23,13133,12346],{},[75,13135,12349],{},[75,13137,12352],{},[75,13139,12355],{},[75,13141,13142],{},"US #1; Australia Seaborne + PRB",[75,13144,12361],{},[60,13146,13147,13151,13153,13155,13158,13161],{},[75,13148,13149],{},[23,13150,12368],{},[75,13152,12371],{},[75,13154,12374],{},[75,13156,13157],{},"~AUD 9B",[75,13159,13160],{},"High-quality NSW; direct Asia sales",[75,13162,12383],{},[60,13164,13165,13169,13171,13173,13175,13178],{},[75,13166,13167],{},[23,13168,12390],{},[75,13170,12393],{},[75,13172,12352],{},[75,13174,1250],{},[75,13176,13177],{},"Alabama premium HCC pure-play",[75,13179,12403],{},[60,13181,13182,13186,13188,13190,13192,13195],{},[75,13183,13184],{},[23,13185,12410],{},[75,13187,12413],{},[75,13189,12416],{},[75,13191,12419],{},[75,13193,13194],{},"High-yield US MLP",[75,13196,12425],{},[60,13198,13199,13203,13205,13207,13209,13212],{},[75,13200,13201],{},[23,13202,12432],{},[75,13204,12435],{},[75,13206,12352],{},[75,13208,12440],{},[75,13210,13211],{},"Royalty model; low op risk",[75,13213,12446],{},[19,13215,13216,13217,13224],{},"US thermal demand is structurally declining per EIA, but ",[23,13218,13219,13220,13223],{},"met coal exports are forecast to ",[802,13221,13222],{},"rise"," in 2026"," — and BTU (via Centurion), HCC, and WHC have the most direct exposure to that.",[205,13226],{},[32,13228,13230],{"id":13229},"scenario-strategy-outlook","Scenario Strategy & Outlook",[19,13232,13233,13234,849],{},"The biggest call in this piece: coal stocks should be held as a ",[23,13235,13236],{},"tactical trade, not a structural long",[122,13238,13239],{"id":5452},"Scenario analysis",[54,13241,13242,13257],{},[57,13243,13244],{},[60,13245,13246,13248,13251,13254],{},[63,13247,1389],{},[63,13249,13250],{},"Newcastle range",[63,13252,13253],{},"Brent outlook",[63,13255,13256],{},"Implication for coal stocks",[70,13258,13259,13273,13287],{},[60,13260,13261,13264,13267,13270],{},[75,13262,13263],{},"Base (ceasefire by April)",[75,13265,13266],{},"$120–$140/t",[75,13268,13269],{},"$80–$90/bbl",[75,13271,13272],{},"Profit-taking; downside risk",[60,13274,13275,13278,13281,13284],{},[75,13276,13277],{},"Main (conflict continues months)",[75,13279,13280],{},"$145–$200/t",[75,13282,13283],{},"$100–$120/bbl",[75,13285,13286],{},"Further upside for BTU/WHC: +30–40%",[60,13288,13289,13292,13295,13298],{},[75,13290,13291],{},"Tail (unresolved by year-end)",[75,13293,13294],{},"$180–$260/t",[75,13296,13297],{},"$120–$150/bbl",[75,13299,13300],{},"1970s-style energy crisis; stagflation",[122,13302,13304],{"id":13303},"short-term-relative-long-inside-the-energy-sector","Short term: relative-long inside the energy sector",[19,13306,13307,13308,13311,13312,13315],{},"Note the macro: Brent above $112; S&P 500 down ",[23,13309,13310],{},"−7%"," from highs; Nasdaq off ",[23,13313,13314],{},"−10%+",". Sticky energy inflation has pushed central-bank cuts back, and \"higher for longer\" is alive again.",[19,13317,13318,13319,13322],{},"In this regime, holding coal stocks as an outright market long is more dangerous than it looks. The sharper expression is ",[23,13320,13321],{},"a relative-long inside the energy sector"," — paired against renewables or demand-sensitive shorts. From investment-committee experience, \"naked\" longs in this kind of environment are the ones that bite later.",[122,13324,13326],{"id":13325},"medium-to-long-term-the-paradoxical-accelerator-for-renewables","Medium-to-long term: the paradoxical accelerator for renewables",[19,13328,13329,13330,13333],{},"Counterintuitively, this crisis is the ",[23,13331,13332],{},"clearest accelerator for renewable and electrification investment"," in years. Birol said it bluntly: \"Solar isn't a romantic story anymore — it's a business.\" Ember's research lead added: \"Electrotech gives countries the tools to remove imported fuels entirely.\"",[19,13335,13336,13337,13340],{},"After Russia-Ukraine in 2022, European solar deployments doubled — then slowed when rates rose. This crisis re-cements the equation ",[23,13338,13339],{},"energy security = domestic energy = renewables",". The methane-slip → maritime-decarbonization corridor I follow daily moves up the urgency scale in lockstep, in my view.",[122,13342,1565],{"id":1564},[312,13344,13345,13351,13357,13363,13369],{},[315,13346,13347,13350],{},[23,13348,13349],{},"April 6 deadline"," — Trump's demand for Hormuz reopening. The post-deadline US response (resumed strikes on Iranian power infrastructure or not) is the largest single catalyst.",[315,13352,13353,13356],{},[23,13354,13355],{},"Ras Laffan repair timeline"," — Does the 3–5 year estimate get hardened? That sets long-term LNG structure.",[315,13358,13359,13362],{},[23,13360,13361],{},"IEA emergency stock release scale"," — Members agreed on a 400 mb release; the drawdown pace is the principal price-floor input.",[315,13364,13365,13368],{},[23,13366,13367],{},"China dynamics"," — RMB-denominated \"transit fee\" reports from Iran. Expansion suggests an entrenched fragmentation of energy flows.",[315,13370,13371,13374],{},[23,13372,13373],{},"Indonesian export policy"," — As the world's largest exporter, even small policy nudges feed straight into Newcastle.",[19,13376,13377,13378,13381],{},"Bottom line: ",[23,13379,13380],{},"short-term long, medium-term neutral",". Coal stocks are a tactical trade on the Iran crisis — not a structural long. But the \"import-dependence risk\" this crisis has exposed will accelerate investment urgency in distributed energy, battery storage, and maritime decarbonization. The seeds of the next structural shift live inside this crisis.",[205,13383],{},[19,13385,13386],{},[802,13387,13388],{},"This article is for informational purposes only and does not constitute a recommendation to buy or sell any financial instrument. Investment decisions are made at your own risk.",{"title":143,"searchDepth":1605,"depth":1605,"links":13390},[13391,13395,13401,13406,13412,13416,13422,13427],{"id":11855,"depth":1605,"text":11855,"children":13392},[13393,13394],{"id":11867,"depth":1610,"text":11868},{"id":11974,"depth":1610,"text":11975},{"id":12001,"depth":1605,"text":12001,"children":13396},[13397,13398,13399,13400],{"id":12011,"depth":1610,"text":12012},{"id":12029,"depth":1610,"text":12030},{"id":12040,"depth":1610,"text":12041},{"id":12145,"depth":1610,"text":12146},{"id":12265,"depth":1605,"text":12265,"children":13402},[13403,13404,13405],{"id":12271,"depth":1610,"text":12272},{"id":12285,"depth":1610,"text":12286},{"id":12296,"depth":1610,"text":12296},{"id":12458,"depth":1605,"text":12458,"children":13407},[13408,13409,13410,13411],{"id":4797,"depth":1610,"text":4797},{"id":12531,"depth":1610,"text":12532},{"id":12549,"depth":1610,"text":12550},{"id":764,"depth":1610,"text":764},{"id":12628,"depth":1605,"text":12629,"children":13413},[13414,13415],{"id":12645,"depth":1610,"text":12646},{"id":12744,"depth":1610,"text":12745},{"id":12778,"depth":1605,"text":12779,"children":13417},[13418,13419,13420,13421],{"id":12789,"depth":1610,"text":12790},{"id":12806,"depth":1610,"text":12807},{"id":12817,"depth":1610,"text":12818},{"id":12921,"depth":1610,"text":12922},{"id":13032,"depth":1605,"text":13033,"children":13423},[13424,13425,13426],{"id":13050,"depth":1610,"text":13051},{"id":13068,"depth":1610,"text":13069},{"id":13083,"depth":1610,"text":13084},{"id":13229,"depth":1605,"text":13230,"children":13428},[13429,13430,13431,13432],{"id":5452,"depth":1610,"text":13239},{"id":13303,"depth":1610,"text":13304},{"id":13325,"depth":1610,"text":13326},{"id":1564,"depth":1610,"text":1565},"ホルムズ海峡封鎖でLNG供給が寸断されるなか、アジア全域で石炭回帰が加速。Newcastle石炭は52週高値圏へ。石炭関連株の短期・中期の投資シナリオをCVCの海洋分析の視点で整理する。",{"date":13435,"image":13436,"alt":13437,"tags":13438,"tagsEn":13439,"published":1666},"28th Mar 2026","/blogs-img/blog-iran-coal-stocks.png","ホルムズ海峡封鎖とNewcastle石炭先物の急騰チャート",[12007,3370,4239,4240,4241],[13440,4050,4244,4245,4241],"Coal","/blogs/3-iran-coal-stocks",{"title":11837,"description":13433},"blogs/3-iran-coal-stocks","HzzHg-Tv122MlJAzVCHGt2cSly2G2-eHzaLZl3uNAkY",{"id":13446,"title":13447,"body":13448,"description":14687,"extension":1651,"meta":14688,"navigation":1666,"ogImage":13492,"path":14693,"seo":14694,"stem":14695,"__hash__":14696},"content/blogs/4-drone-logistics.md","ドローン物流が「実験」から「日常」へ——市場規模と有望企業の全貌 | Drone Logistics From Experiment to Everyday: Market Size and Top Companies",{"type":7,"value":13449,"toc":14648},[13450,14052,14054],[10,13451,13452,13456,13467,13471,13474,13477,13484,13487,13493,13495,13499,13502,13505,13608,13615,13618,13628,13634,13640,13643,13654,13661,13663,13666,13669,13673,13679,13686,13697,13702,13706,13717,13723,13727,13733,13737,13740,13744,13747,13899,13901,13904,13907,13909,13978,13981,13991,13997,14003,14006,14009,14011,14043,14046,14048],{"lang":12},[14,13453,13455],{"id":13454},"ドローン物流が実験から日常へ","ドローン物流が「実験」から「日常」へ",[19,13457,13458,13459,13462,13463,13466],{},"私はCVCで北米EVリユース蓄電・半固体電池に出資した経験に加えて、自社で河川洪水検知IoTシステムをAWS（IoT Core・Lambda・DynamoDB・API Gateway）でゼロから設計・構築した経験があります。",[23,13460,13461],{},"「ハードウェアと自律オペレーションは、ソフトウェア単独より3倍ややこしい」","——これが現場で身に染みた感覚です。だから今回ドローン物流を見るときも、機体スペックではなく",[23,13464,13465],{},"ユニットエコノミクス・規制の進度・運航ソフトウェアの完成度","の3点で解きほぐしてみたいと思います。",[32,13468,13470],{"id":13469},"空から届く時代がついに日常になる","空から届く時代がついに「日常」になる",[19,13472,13473],{},"最初に、市場が今どこに居るかをフラットに置きます。",[19,13475,13476],{},"ドローン配送がようやく「実験段階」を脱し、「日常のインフラ」へと移行しつつあります。2026年は、この転換点が明確に可視化された年です。1月、世界最大のドローン配送企業Ziplineが$800M（初回$600M + 追加$200M）を調達し、バリュエーションは**$7.6B（約1.1兆円）**に到達。同月、Alphabet傘下のWingはWalmartとのパートナーシップを150店舗追加に拡大し、2027年末までに全米270拠点のドローン配送ネットワークを構築する計画を発表。3月にはサンフランシスコ・ベイエリアへの進出も明らかにしました。",[19,13478,13479,13480,13483],{},"これらは単なるプレスの積み重ねではない、というのが私の見立てです。",[23,13481,13482],{},"ユニットエコノミクスと規制環境が同時に臨界点を超えた","ことを示すシグナルなんですよね。10年以上かけて成熟してきた技術が、eコマースの即日配送需要、ラストマイル物流のコスト圧力、そして医療アクセスの格差という3つの構造的課題と合流し、爆発的な成長フェーズに入った。",[19,13485,13486],{},"Ziplineの共同創業者兼CEO Keller Cliffton氏は「2026年、自律型物流は米国の複数の州で日常の一部になる」と宣言しています。誇大広告ではなく、同社の米国内配送量は過去7ヶ月間、**週次+15%**のペースで成長し続けています。",[19,13488,13489],{},[4681,13490],{"alt":13491,"src":13492},"ドローン配送センターと都市上空を飛ぶ配送ドローン群","/blogs-img/blog-drone-logistics.png",[205,13494],{},[32,13496,13498],{"id":13497},"市場規模cagr4548の爆発的成長","市場規模：CAGR45〜48%の爆発的成長",[19,13500,13501],{},"市場規模の数字は調査会社によって振れるのですが、共通しているのは**「年率40%超」**という異次元の成長率です。",[122,13503,13504],{"id":13504},"主要調査会社の予測比較",[54,13506,13507,13522],{},[57,13508,13509],{},[60,13510,13511,13514,13517,13520],{},[63,13512,13513],{},"調査会社",[63,13515,13516],{},"2025年実績",[63,13518,13519],{},"2030-35年予測",[63,13521,237],{},[70,13523,13524,13538,13552,13566,13580,13594],{},[60,13525,13526,13529,13532,13535],{},[75,13527,13528],{},"IMARC Group",[75,13530,13531],{},"$1.3B",[75,13533,13534],{},"$34.7B (2034)",[75,13536,13537],{},"43.8%",[60,13539,13540,13543,13546,13549],{},[75,13541,13542],{},"Mordor Intelligence",[75,13544,13545],{},"$0.66B",[75,13547,13548],{},"$6.78B (2031)",[75,13550,13551],{},"47.6%",[60,13553,13554,13557,13560,13563],{},[75,13555,13556],{},"Precedence Research",[75,13558,13559],{},"$2.05B",[75,13561,13562],{},"$103.7B (2035)",[75,13564,13565],{},"48.1%",[60,13567,13568,13571,13574,13577],{},[75,13569,13570],{},"Future Market Insights",[75,13572,13573],{},"$2.1B",[75,13575,13576],{},"$87.6B (2035)",[75,13578,13579],{},"45.5%",[60,13581,13582,13585,13588,13591],{},[75,13583,13584],{},"Grand View Research（ドローン全体）",[75,13586,13587],{},"$83.8B",[75,13589,13590],{},"$182.5B (2033)",[75,13592,13593],{},"9.5%",[60,13595,13596,13599,13602,13605],{},[75,13597,13598],{},"Fortune Business Insights（配送特化）",[75,13600,13601],{},"$3.47B",[75,13603,13604],{},"$21.0B (2034)",[75,13606,13607],{},"19.5%",[19,13609,13610,13611,13614],{},"数字のばらつきは「ドローン物流」の定義範囲——ハードウェア・ソフトウェア・インフラを含むか、配送サービスのみか——に起因します。が、いずれの調査でも",[23,13612,13613],{},"2030年代前半に数十兆円規模の市場","が見込まれている点では一致しています。",[122,13616,13617],{"id":13617},"成長を牽引する3つのドライバー",[19,13619,13620,13623,13624,13627],{},[23,13621,13622],{},"1. eコマースの「超即時配送」需要"," — Wingの上位25%のユーザーは",[23,13625,13626],{},"週3回","ドローン配送を利用しており、注文されるのは卵、挽肉、トマト、アボカド、携帯充電器といった「日常品」です。即日どころか「30分以内」が期待値になりつつある。",[19,13629,13630,13633],{},[23,13631,13632],{},"2. 医療物流の構造的ニーズ"," — ドローン配送は医療分野で最も先行的に実用化されました。Ziplineはルワンダで血液製剤の配送からスタートし、現在は5カ国のアフリカ諸国で5,000以上の病院・医療施設にサービスを提供。日本でも長崎県でのエリア型Level 4医薬品配送が2026年1月に実現しました。",[19,13635,13636,13639],{},[23,13637,13638],{},"3. 規制環境の整備"," — 米国ではFAAがBVLOS（目視外飛行）の規則整備を加速し、トランプ政権はドローン産業の成長を促進する大統領令を発出。日本では2022年12月のLevel 4解禁以降、2026年度のUTM（無人機交通管理システム）構築に向けたロードマップが進行中です。",[122,13641,13642],{"id":13642},"セグメント別の注目ポイント",[19,13644,13645,13646,13653],{},"アプリケーション別では、",[23,13647,13648,13649,13652],{},"小売・物流が45.6%",[23,13650,13651],{},"のシェアで最大セグメントですが、成長率が最も高いのは","医療物資配送（CAGR 51.4%）","。地域別ではアジア太平洋地域が**CAGR 51.9%**で全地域中最速です。中国・インド・日本・東南アジアの人口密集都市と島嶼・山間部の物流課題が、ドローン配送の「ユースケースの宝庫」になっている。",[19,13655,13656,13657,13660],{},"私はAWS IoTでセンサーシステムを実装した経験から思うのですが、ドローン物流の本当の参入障壁は",[23,13658,13659],{},"機体ではなく運航ソフトウェアとUTM","側にあります。フリート規模が10機を超えると、ソフトの完成度が一気にスケールの律速になる——というのは、河川洪水検知システムを6ヶ月で立ち上げた時の感覚と同じです。",[205,13662],{},[32,13664,13665],{"id":13665},"注目企業の徹底分析",[19,13667,13668],{},"ドローン物流市場は「実験期」を経て「スケーリング期」に入り、勝者と敗者の輪郭が見え始めています。",[122,13670,13672],{"id":13671},"zipline-international-業界の絶対王者","Zipline International — 業界の絶対王者",[19,13674,13675,13678],{},[23,13676,13677],{},"基本情報","：2014年創業、本社サンフランシスコ。バリュエーション$7.6B。累計配送200万件超、1億2,500万マイルの自律飛行実績。",[19,13680,13681,13682,13685],{},"Ziplineは名実ともにドローン物流業界のリーダーです。単なるドローンメーカーではなく、",[23,13683,13684],{},"機体・発射/着陸システム・物流ソフトウェアを垂直統合","したフルスタック企業である点が競合との最大の差別化要因。**Platform 1（P1）**は長距離配送用の固定翼ドローンで往復120マイルをカバーし、医療・企業・政府向けの大型貨物を運ぶ。**Platform 2（P2）**は家庭向けの短距離配送用で、最大8ポンドの荷物を半径10マイル以内に届ける。",[19,13687,13688,13689,13692,13693,13696],{},"成長の加速度は驚異的です。ダラスの最初の拠点では1日100件の配送に到達するのに",[23,13690,13691],{},"10週間","かかったが、新規拠点では",[23,13694,13695],{},"わずか2日","で同じ水準に到達。Q3の日次配送目標を30%上回り、Q4の目標は6週間前倒しで達成しました。",[19,13698,13699,13701],{},[23,13700,117],{},"：Fidelity Management & Research、Baillie Gifford、Valor Equity Partners、Tiger Global、Paradigm",[122,13703,13705],{"id":13704},"wing-aviationalphabet子会社-巨人の組織力","Wing Aviation（Alphabet子会社） — 巨人の組織力",[19,13707,13708,13709,13712,13713,13716],{},"WingはAlphabetのリソースを活かした「プラットフォーム戦略」で市場を攻略しています。Walmartとの提携は2023年にダラスの2店舗からスタートし、2025年にアトランタ、そして2026年1月に150店舗追加が発表された。2027年末までに",[23,13710,13711],{},"270店舗","のネットワークを構築し、対象人口は約",[23,13714,13715],{},"4,000万人","（米国人口の約12%）に達する見込み。直近の配送量は過去6ヶ月で3倍に成長しています。",[19,13718,13719,13722],{},[23,13720,13721],{},"投資的含意","：Alphabet（GOOG/GOOGL）を通じたエクスポージャー。Wing単体での投資は難しいが、Waymoに次ぐ「隠れた価値」として再評価される可能性があります。",[122,13724,13726],{"id":13725},"matternet-医療物流の先駆者","Matternet — 医療物流の先駆者",[19,13728,13729,13730,758],{},"シリコンバレー拠点で米国の複数の病院システムと提携。M2ドローンは最大2kgのペイロードを12.5マイル運搬でき、温度管理コンテナによる血液サンプルや医薬品の配送に強み。2025年1月にはサウジアラビアの航空当局（GACA）から運航承認を取得し、同国初の貨物ドローン運航者となった。累計調達額$74Mと小規模で、",[23,13731,13732],{},"M&Aターゲットになりやすいポジション",[122,13734,13736],{"id":13735},"wingcopter-ドイツ発のグローバルプレイヤー","Wingcopter — ドイツ発のグローバルプレイヤー",[19,13738,13739],{},"VTOL固定翼ドローン「Wingcopter 198」を開発。最高速度150km/h、航続距離100km（ペイロード2kg時）で、僻地・離島への配送に適する。日本では伊藤忠商事・ANAホールディングスと共同で、沖縄で53kmの血液輸送実証試験を実施。Zipline・Wing・Matternetと合わせて上位4社で市場シェアの**70%**を占めます。",[122,13741,13743],{"id":13742},"日本勢acsl-aeronext-2024年問題が追い風に","日本勢：ACSL × Aeronext — 「2024年問題」が追い風に",[19,13745,13746],{},"ACSL（東証グロース: 6232）は日本初のLevel 4型式認証を取得した国産ドローンメーカー。AeronextはACSLと共同開発した物流専用ドローン**「AirTruck」**（ペイロード5kg、航続20km）を通じて、地方自治体と連携したラストマイル配送の社会実装を進めています。トラックドライバーの時間外労働上限規制（2024年問題）と物流キャパシティ不足が、追い風として効きやすい構造です。",[54,13748,13749,13770],{},[57,13750,13751],{},[60,13752,13753,13755,13758,13761,13764,13767],{},[63,13754,434],{},[63,13756,13757],{},"国",[63,13759,13760],{},"調達額/時価総額",[63,13762,13763],{},"累計配送数",[63,13765,13766],{},"強み",[63,13768,13769],{},"注目イベント",[70,13771,13772,13794,13815,13835,13856,13878],{},[60,13773,13774,13779,13782,13785,13788,13791],{},[75,13775,13776],{},[23,13777,13778],{},"Zipline",[75,13780,13781],{},"米国",[75,13783,13784],{},"$7.6B（val）",[75,13786,13787],{},"200万件超",[75,13789,13790],{},"フルスタック垂直統合",[75,13792,13793],{},"IPO候補（2026-27年）",[60,13795,13796,13801,13804,13807,13809,13812],{},[75,13797,13798],{},[23,13799,13800],{},"Wing",[75,13802,13803],{},"米国（Alphabet）",[75,13805,13806],{},"Alphabet子会社",[75,13808,489],{},[75,13810,13811],{},"Walmart提携270店舗",[75,13813,13814],{},"ベイエリア進出",[60,13816,13817,13822,13824,13827,13829,13832],{},[75,13818,13819],{},[23,13820,13821],{},"Matternet",[75,13823,13781],{},[75,13825,13826],{},"$74M調達",[75,13828,489],{},[75,13830,13831],{},"医療特化、FAA型式認証",[75,13833,13834],{},"サウジ進出",[60,13836,13837,13842,13845,13848,13850,13853],{},[75,13838,13839],{},[23,13840,13841],{},"Wingcopter",[75,13843,13844],{},"ドイツ",[75,13846,13847],{},"$125M+調達",[75,13849,489],{},[75,13851,13852],{},"VTOL長距離、新興国展開",[75,13854,13855],{},"メキシコ・沖縄",[60,13857,13858,13863,13866,13869,13872,13875],{},[75,13859,13860],{},[23,13861,13862],{},"ACSL",[75,13864,13865],{},"日本",[75,13867,13868],{},"東証6232",[75,13870,13871],{},"実証段階",[75,13873,13874],{},"国産Level 4認証",[75,13876,13877],{},"UTM構築（2026年度）",[60,13879,13880,13885,13887,13890,13893,13896],{},[75,13881,13882],{},[23,13883,13884],{},"Amazon Prime Air",[75,13886,13781],{},[75,13888,13889],{},"Amazon子会社",[75,13891,13892],{},"~16,000件",[75,13894,13895],{},"50,000SKU対応",[75,13897,13898],{},"5州で運航中",[205,13900],{},[32,13902,13903],{"id":13903},"シナリオ別投資戦略",[19,13905,13906],{},"ドローン物流は2026年時点で市場規模が$1〜3B程度と、まだ「初期段階」にあります。上場企業としての純粋なエクスポージャーは限られているが、いくつかの投資経路があります。",[122,13908,4797],{"id":4797},[54,13910,13911,13924],{},[57,13912,13913],{},[60,13914,13915,13917,13920,13922],{},[63,13916,601],{},[63,13918,13919],{},"市場規模（2030年）",[63,13921,4808],{},[63,13923,4811],{},[70,13925,13926,13939,13952,13965],{},[60,13927,13928,13930,13933,13936],{},[75,13929,4818],{},[75,13931,13932],{},"$8〜12B",[75,13934,13935],{},"BVLOS規制の段階的整備。都市部限定のスケーリング",[75,13937,13938],{},"Zipline IPO成功。既存物流企業の参入加速",[60,13940,13941,13943,13946,13949],{},[75,13942,4829],{},[75,13944,13945],{},"$15〜25B",[75,13947,13948],{},"UTM構築完了。全米・欧州・アジア主要都市で日常化",[75,13950,13951],{},"カテゴリー全体のバリュエーション再評価。M&A活発化",[60,13953,13954,13956,13959,13962],{},[75,13955,4840],{},[75,13957,13958],{},"$30B超",[75,13960,13961],{},"AI自律飛行の飛躍的進化。規制の完全自由化",[75,13963,13964],{},"自動運転に次ぐ「自律型モビリティ」テーマとして株式市場を席巻",[60,13966,13967,13969,13972,13975],{},[75,13968,4851],{},[75,13970,13971],{},"$5B以下",[75,13973,13974],{},"重大事故による規制強化。公衆の受容性低下",[75,13976,13977],{},"成長鈍化、バリュエーション調整",[122,13979,13980],{"id":13980},"投資経路",[19,13982,13983,13986,13987,13990],{},[23,13984,13985],{},"プレIPO/ベンチャー投資"," — Ziplineは$7.6Bのバリュエーションで未上場。Fidelity、Baillie Gifford、Tiger Globalといったクロスオーバー投資家が参入しており、IPO前の最後の成長ラウンドとしてのポジショニングが可能。CVCで上場直前ラウンドの感触を見てきた経験から言うと、Ziplineは",[23,13988,13989],{},"自律型物流のカテゴリー・リーダー","としてのプレミアムが付きやすい銘柄だと思います。",[19,13992,13993,13996],{},[23,13994,13995],{},"Alphabet（GOOG/GOOGL）"," — Wingの270店舗スケールが実現すれば、再評価カタリスト。",[19,13998,13999,14002],{},[23,14000,14001],{},"日本の関連銘柄"," — ACSL（6232）はLevel 4認証を持つ唯一の国産メーカーとして希少性。伊藤忠商事（8001）やANAホールディングス（9202）はWingcopter協業を通じたエクスポージャー。",[122,14004,14005],{"id":14005},"中長期の構造変化",[19,14007,14008],{},"注目すべきは、この市場が**「勝者総取り」になりにくい構造**を持っていることです。地域ごとの規制差異、ユースケースの多様性（医療 vs リテール vs 産業用）、そしてハードウェアとソフトウェアの両方が必要なフルスタックの複雑性が、複数の勝者を生み出す土壌になっている。Ziplineの長距離医療配送、Wingの大規模リテール提携、Matternetの都市型医療、ACSLの日本国内Level 4——それぞれが異なるニッチで勝ち残る可能性があります。",[122,14010,764],{"id":764},[312,14012,14013,14019,14025,14031,14037],{},[315,14014,14015,14018],{},[23,14016,14017],{},"Zipline IPOのタイミングと条件"," — 2026年後半〜2027年のIPOウィンドウが有力",[315,14020,14021,14024],{},[23,14022,14023],{},"FAAのBVLOS最終規則"," — 最終化の時期と内容がスケーラビリティを決定",[315,14026,14027,14030],{},[23,14028,14029],{},"日本のUTM構築"," — 2026年度の完成目標",[315,14032,14033,14036],{},[23,14034,14035],{},"Amazon Prime Airの動向"," — 累計約16,000件と後塵だが、Amazonの物流ネットワーク統合次第で一気に逆転可能性",[315,14038,14039,14042],{},[23,14040,14041],{},"イラン戦争とサプライチェーンの寸断"," — グローバル物流の混乱は、ドローン物流の「代替手段としての価値」を押し上げる",[19,14044,14045],{},"結論として、ドローン物流は**「今、入るべきテーマ」**です。市場はまだ$1〜3B規模と小さいが、CAGR 45%超の成長軌道に乗っています。Ziplineを筆頭とするリーダー企業は「実験」から「スケーリング」へ明確にフェーズが移行しており、今後2〜3年で市場の勝者と構造が固まる。この窓が開いている今こそ、ポジションを取るべきタイミングだと見ています。",[205,14047],{},[19,14049,14050],{},[802,14051,4907],{},[205,14053],{},[10,14055,14056,14060,14070,14074,14077,14084,14091,14098,14103,14105,14109,14115,14196,14199,14203,14213,14219,14225,14229,14244,14251,14253,14257,14260,14264,14279,14290,14296,14300,14311,14315,14322,14326,14329,14333,14336,14481,14483,14487,14490,14492,14565,14569,14579,14585,14591,14595,14602,14604,14635,14642,14644],{"lang":809},[14,14057,14059],{"id":14058},"drone-logistics-from-experiment-to-everyday","Drone Logistics: From Experiment to Everyday",[19,14061,14062,14063,14066,14067,849],{},"I worked on EV second-life storage and semi-solid battery deals at a corporate VC, and I also designed and built our company's river-flood-detection IoT system from scratch on AWS (IoT Core, Lambda, DynamoDB, API Gateway). Both experiences hammered home the same lesson: ",[23,14064,14065],{},"hardware plus autonomous operations are about three times messier than pure software",". That's the lens I want to apply here — not \"drone specs,\" but ",[23,14068,14069],{},"unit economics, regulatory cadence, and ops-software maturity",[32,14071,14073],{"id":14072},"the-age-of-aerial-delivery-has-arrived","The Age of Aerial Delivery Has Arrived",[19,14075,14076],{},"Let me first put the picture down flat.",[19,14078,14079,14080,14083],{},"Drone delivery has finally moved beyond the experimental stage and is settling into everyday infrastructure. 2026 is the year that inflection became unmistakable. In January, Zipline — the world's largest drone delivery company — raised $800M (initial $600M + add-on $200M), reaching a ",[23,14081,14082],{},"$7.6B valuation",". The same month, Alphabet's Wing expanded its Walmart partnership by 150 stores, with a plan to build a 270-location nationwide network by end-2027. In March, Wing announced expansion into the San Francisco Bay Area.",[19,14085,14086,14087,14090],{},"These aren't just press releases. To me, they're the signal that ",[23,14088,14089],{},"unit economics and regulatory readiness crossed their critical thresholds simultaneously",". A decade of maturation collided with three structural forces — same-hour e-commerce demand, last-mile cost pressure, and gaps in healthcare access — and the result is a step change.",[19,14092,14093,14094,14097],{},"Zipline CEO Keller Cliffton put it bluntly: \"In 2026, autonomous logistics becomes part of daily life in multiple US states.\" Their US delivery volumes have grown ",[23,14095,14096],{},"+15% week-over-week for seven straight months",", so the rhetoric is grounded.",[19,14099,14100],{},[4681,14101],{"alt":14102,"src":13492},"Drone delivery hub with fleets of drones launching over a city skyline",[205,14104],{},[32,14106,14108],{"id":14107},"market-size-4548-cagr","Market Size: 45–48% CAGR",[19,14110,14111,14112,849],{},"Forecasts vary on definition, but they agree on one thing: ",[23,14113,14114],{},"annual growth above 40%",[54,14116,14117,14132],{},[57,14118,14119],{},[60,14120,14121,14124,14127,14130],{},[63,14122,14123],{},"Research firm",[63,14125,14126],{},"2025 estimate",[63,14128,14129],{},"2030–35 forecast",[63,14131,237],{},[70,14133,14134,14144,14154,14164,14174,14185],{},[60,14135,14136,14138,14140,14142],{},[75,14137,13528],{},[75,14139,13531],{},[75,14141,13534],{},[75,14143,13537],{},[60,14145,14146,14148,14150,14152],{},[75,14147,13542],{},[75,14149,13545],{},[75,14151,13548],{},[75,14153,13551],{},[60,14155,14156,14158,14160,14162],{},[75,14157,13556],{},[75,14159,13559],{},[75,14161,13562],{},[75,14163,13565],{},[60,14165,14166,14168,14170,14172],{},[75,14167,13570],{},[75,14169,13573],{},[75,14171,13576],{},[75,14173,13579],{},[60,14175,14176,14179,14181,14183],{},[75,14177,14178],{},"Grand View Research (drones)",[75,14180,13587],{},[75,14182,13590],{},[75,14184,13593],{},[60,14186,14187,14190,14192,14194],{},[75,14188,14189],{},"Fortune Business Insights",[75,14191,13601],{},[75,14193,13604],{},[75,14195,13607],{},[19,14197,14198],{},"The spread reflects whether \"drone logistics\" includes hardware/software/infrastructure or only the delivery service. But all forecasts agree on a multi-tens-of-billions market by the early 2030s.",[122,14200,14202],{"id":14201},"the-three-demand-drivers","The three demand drivers",[19,14204,14205,14208,14209,14212],{},[23,14206,14207],{},"1. \"Hyper-immediate\" e-commerce delivery."," Wing's top quartile of users orders ",[23,14210,14211],{},"three times a week",". The actual SKUs: eggs, ground beef, tomatoes, avocados, phone chargers — everyday goods. Same-hour delivery is becoming the expectation; \"within 30 minutes\" is increasingly normal.",[19,14214,14215,14218],{},[23,14216,14217],{},"2. Structural medical-logistics demand."," Drones broke into healthcare logistics first. Zipline started in Rwanda with blood products and now serves 5,000+ hospitals and clinics across five African countries. Japan ran its first Level 4 area-based pharmaceutical drone deliveries in Nagasaki in January 2026.",[19,14220,14221,14224],{},[23,14222,14223],{},"3. Regulatory tailwinds."," The US FAA is accelerating BVLOS (beyond visual line of sight) rulemaking, and the Trump administration issued an executive order to support drone industry growth. Japan's Level 4 framework opened in December 2022, and a national UTM (drone traffic management) build-out is on track for fiscal 2026.",[122,14226,14228],{"id":14227},"segment-notes","Segment notes",[19,14230,14231,14232,14235,14236,14239,14240,14243],{},"Retail and logistics is the largest application at ",[23,14233,14234],{},"45.6%"," share, but the fastest-growing segment is ",[23,14237,14238],{},"medical supplies (51.4% CAGR)",". Asia Pacific is the fastest-growing region at ",[23,14241,14242],{},"51.9% CAGR"," — dense urban centers plus archipelago/mountain logistics gaps make the use cases unusually rich.",[19,14245,14246,14247,14250],{},"Coming back to my IoT-build instinct: the real moat in drone logistics is not the airframe — it's the ",[23,14248,14249],{},"ops software and UTM stack",". Once a fleet exceeds ~10 aircraft, software completeness becomes the rate-limiter. That matches what I learned standing up the river-flood IoT demo in six months.",[205,14252],{},[32,14254,14256],{"id":14255},"players-worth-watching","Players Worth Watching",[19,14258,14259],{},"The market has moved from experimentation into scaling, and winners are starting to take shape.",[122,14261,14263],{"id":14262},"zipline-international-the-runaway-leader","Zipline International — the runaway leader",[19,14265,14266,14267,14270,14271,14274,14275,14278],{},"Founded 2014, HQ San Francisco. Valuation $7.6B. 2M+ deliveries; 125M+ autonomous miles flown. The crucial differentiator is ",[23,14268,14269],{},"vertical integration"," — airframe + launch/recovery + logistics software in a single stack. ",[23,14272,14273],{},"Platform 1 (P1)"," is a fixed-wing for 120-mile round trips (medical / corporate / government). ",[23,14276,14277],{},"Platform 2 (P2)"," is a residential short-haul drone, up to 8 lbs within ~10-mile radius.",[19,14280,14281,14282,14285,14286,14289],{},"Adoption acceleration is striking: the Dallas hub took ",[23,14283,14284],{},"10 weeks"," to hit 100 deliveries/day at first; new hubs reach the same level in ",[23,14287,14288],{},"two days",". Q3 daily targets exceeded by 30%; Q4 target hit six weeks early.",[19,14291,14292,14295],{},[23,14293,14294],{},"Investor base:"," Fidelity, Baillie Gifford, Valor Equity Partners, Tiger Global, Paradigm.",[122,14297,14299],{"id":14298},"wing-aviation-alphabet-platform-play","Wing Aviation (Alphabet) — platform play",[19,14301,14302,14303,14306,14307,14310],{},"Walmart partnership scaled from 2 stores in Dallas (2023) → Atlanta (2025) → +150 stores (Jan 2026). Network targets ",[23,14304,14305],{},"270 stores by end-2027",", addressing roughly ",[23,14308,14309],{},"40M Americans (~12% of the population)",". Delivery volume tripled in the past 6 months. Public-market exposure flows through GOOG/GOOGL — Wing alone is hard to size, but it could become the next Waymo-style \"hidden value\" line item.",[122,14312,14314],{"id":14313},"matternet-medical-logistics-specialist","Matternet — medical-logistics specialist",[19,14316,14317,14318,14321],{},"Silicon Valley-based, FAA Part 135 certified. M2 carries up to 2 kg over 12.5 miles, with temperature-controlled containers for blood and pharmaceuticals. Total raised $74M — small enough that ",[23,14319,14320],{},"M&A is the more likely outcome than IPO",", in my view.",[122,14323,14325],{"id":14324},"wingcopter-global-vtol-maker","Wingcopter — global VTOL maker",[19,14327,14328],{},"German VTOL fixed-wing maker. 150 km/h max speed, 100 km range with 2 kg payload. Active in Africa, Latin America, and Asia — including Okinawa blood-transport trials with Itochu and ANA Holdings. Together with Zipline, Wing, and Matternet, the top four hold ~70% market share.",[122,14330,14332],{"id":14331},"japanese-players-acsl-aeronext-riding-the-2024-problem","Japanese players: ACSL × Aeronext — riding the \"2024 problem\"",[19,14334,14335],{},"ACSL (TSE: 6232) is Japan's only domestic drone maker with Level 4 type certification. Through \"AirTruck\" (5 kg payload, 20 km range), co-developed with Aeronext, the pair is rolling out municipal last-mile delivery. Japan's \"2024 problem\" — a structural truck-driver-hours shortage — adds a regulatory tailwind to the demand side.",[54,14337,14338,14358],{},[57,14339,14340],{},[60,14341,14342,14344,14346,14349,14352,14355],{},[63,14343,1228],{},[63,14345,12838],{},[63,14347,14348],{},"Funding/Mkt cap",[63,14350,14351],{},"Total deliveries",[63,14353,14354],{},"Strength",[63,14356,14357],{},"Watch event",[70,14359,14360,14381,14401,14420,14440,14461],{},[60,14361,14362,14366,14369,14372,14375,14378],{},[75,14363,14364],{},[23,14365,13778],{},[75,14367,14368],{},"US",[75,14370,14371],{},"$7.6B (val)",[75,14373,14374],{},"2M+",[75,14376,14377],{},"Full-stack vertical integration",[75,14379,14380],{},"IPO candidate (2026–27)",[60,14382,14383,14387,14390,14393,14395,14398],{},[75,14384,14385],{},[23,14386,13800],{},[75,14388,14389],{},"US (Alphabet)",[75,14391,14392],{},"Alphabet sub",[75,14394,1281],{},[75,14396,14397],{},"270-store Walmart partnership",[75,14399,14400],{},"Bay Area expansion",[60,14402,14403,14407,14409,14412,14414,14417],{},[75,14404,14405],{},[23,14406,13821],{},[75,14408,14368],{},[75,14410,14411],{},"$74M raised",[75,14413,1281],{},[75,14415,14416],{},"Medical + FAA Part 135",[75,14418,14419],{},"Saudi expansion",[60,14421,14422,14426,14429,14432,14434,14437],{},[75,14423,14424],{},[23,14425,13841],{},[75,14427,14428],{},"Germany",[75,14430,14431],{},"$125M+ raised",[75,14433,1281],{},[75,14435,14436],{},"VTOL long range; emerging mkts",[75,14438,14439],{},"Mexico, Okinawa",[60,14441,14442,14446,14449,14452,14455,14458],{},[75,14443,14444],{},[23,14445,13862],{},[75,14447,14448],{},"Japan",[75,14450,14451],{},"TSE 6232",[75,14453,14454],{},"pilot stage",[75,14456,14457],{},"Domestic Level 4 cert",[75,14459,14460],{},"UTM build-out (FY2026)",[60,14462,14463,14467,14469,14472,14475,14478],{},[75,14464,14465],{},[23,14466,13884],{},[75,14468,14368],{},[75,14470,14471],{},"Amazon sub",[75,14473,14474],{},"~16,000",[75,14476,14477],{},"50,000 SKU integration",[75,14479,14480],{},"Operating in 5 states",[205,14482],{},[32,14484,14486],{"id":14485},"investment-scenarios-routes","Investment Scenarios & Routes",[19,14488,14489],{},"In 2026, the addressable market is still only $1–3B — early stage. Pure-play public exposure is limited, but several routes exist.",[122,14491,13239],{"id":5452},[54,14493,14494,14508],{},[57,14495,14496],{},[60,14497,14498,14500,14503,14506],{},[63,14499,1389],{},[63,14501,14502],{},"2030 market size",[63,14504,14505],{},"Development",[63,14507,5467],{},[70,14509,14510,14523,14537,14551],{},[60,14511,14512,14514,14517,14520],{},[75,14513,6951],{},[75,14515,14516],{},"$8–12B",[75,14518,14519],{},"Phased BVLOS rules, urban scaling",[75,14521,14522],{},"Zipline IPO success; incumbents enter",[60,14524,14525,14528,14531,14534],{},[75,14526,14527],{},"Main",[75,14529,14530],{},"$15–25B",[75,14532,14533],{},"UTM completed; routine in major cities globally",[75,14535,14536],{},"Category re-rating; M&A wave",[60,14538,14539,14542,14545,14548],{},[75,14540,14541],{},"Upside tail",[75,14543,14544],{},"$30B+",[75,14546,14547],{},"AI autonomy leap; full regulatory liberalization",[75,14549,14550],{},"Autonomous-mobility theme sweeps markets",[60,14552,14553,14556,14559,14562],{},[75,14554,14555],{},"Downside tail",[75,14557,14558],{},"\u003C$5B",[75,14560,14561],{},"Major accident triggers regulatory freeze",[75,14563,14564],{},"Growth stalls; valuation correction",[122,14566,14568],{"id":14567},"investment-routes","Investment routes",[19,14570,14571,14574,14575,14578],{},[23,14572,14573],{},"Pre-IPO / venture."," Zipline at $7.6B is still private, with Fidelity, Baillie Gifford, and Tiger Global anchoring crossover rounds. From watching late-stage rounds inside CVC, this is the kind of company that prices the ",[23,14576,14577],{},"\"category leader\" premium"," at IPO.",[19,14580,14581,14584],{},[23,14582,14583],{},"Alphabet (GOOG/GOOGL)."," Wing-related re-rating is possible if the 270-store rollout lands.",[19,14586,14587,14590],{},[23,14588,14589],{},"Japanese names."," ACSL (6232) is the only domestic Level 4-certified maker — scarcity premium. Itochu (8001) and ANA Holdings (9202) carry indirect exposure via Wingcopter partnerships.",[122,14592,14594],{"id":14593},"why-winner-takes-all-doesnt-apply-here","Why \"winner-takes-all\" doesn't apply here",[19,14596,14597,14598,14601],{},"The structural feature I'd most underline: this market is ",[23,14599,14600],{},"not a winner-takes-all market",". Regional regulatory differences, the diversity of use cases (healthcare vs. retail vs. industrial), and the hardware-plus-software full-stack complexity make space for multiple winners. Zipline's long-haul medical, Wing's retail platform plays, Matternet's urban healthcare, ACSL's Japanese Level 4 — each can occupy a distinct, durable niche.",[122,14603,1565],{"id":1564},[312,14605,14606,14612,14618,14624,14629],{},[315,14607,14608,14611],{},[23,14609,14610],{},"Zipline IPO timing & terms"," — 2026 H2 to 2027 IPO window most likely",[315,14613,14614,14617],{},[23,14615,14616],{},"FAA BVLOS final rule"," — timing and substance gate scaling",[315,14619,14620,14623],{},[23,14621,14622],{},"Japan's UTM build-out"," — fiscal 2026 completion target",[315,14625,14626,14628],{},[23,14627,13884],{}," — only ~16,000 deliveries to date, but Amazon-network integration could flip the leaderboard quickly",[315,14630,14631,14634],{},[23,14632,14633],{},"Iran war supply-chain disruption"," — global logistics fragility highlights drone delivery's \"alternative-route\" value, especially in medical and emergency supplies",[19,14636,14637,14638,14641],{},"Bottom line: drone logistics is a ",[23,14639,14640],{},"\"position now\" theme",". Market size is still small ($1–3B), but the 45%+ CAGR runway is real. Leaders like Zipline have decisively transitioned from experimentation into scaling, and the next 2–3 years will lock in the structural winners. The window to position is open — and it won't stay open forever.",[205,14643],{},[19,14645,14646],{},[802,14647,13388],{},{"title":143,"searchDepth":1605,"depth":1605,"links":14649},[14650,14651,14656,14663,14669,14670,14674,14681],{"id":13469,"depth":1605,"text":13470},{"id":13497,"depth":1605,"text":13498,"children":14652},[14653,14654,14655],{"id":13504,"depth":1610,"text":13504},{"id":13617,"depth":1610,"text":13617},{"id":13642,"depth":1610,"text":13642},{"id":13665,"depth":1605,"text":13665,"children":14657},[14658,14659,14660,14661,14662],{"id":13671,"depth":1610,"text":13672},{"id":13704,"depth":1610,"text":13705},{"id":13725,"depth":1610,"text":13726},{"id":13735,"depth":1610,"text":13736},{"id":13742,"depth":1610,"text":13743},{"id":13903,"depth":1605,"text":13903,"children":14664},[14665,14666,14667,14668],{"id":4797,"depth":1610,"text":4797},{"id":13980,"depth":1610,"text":13980},{"id":14005,"depth":1610,"text":14005},{"id":764,"depth":1610,"text":764},{"id":14072,"depth":1605,"text":14073},{"id":14107,"depth":1605,"text":14108,"children":14671},[14672,14673],{"id":14201,"depth":1610,"text":14202},{"id":14227,"depth":1610,"text":14228},{"id":14255,"depth":1605,"text":14256,"children":14675},[14676,14677,14678,14679,14680],{"id":14262,"depth":1610,"text":14263},{"id":14298,"depth":1610,"text":14299},{"id":14313,"depth":1610,"text":14314},{"id":14324,"depth":1610,"text":14325},{"id":14331,"depth":1610,"text":14332},{"id":14485,"depth":1605,"text":14486,"children":14682},[14683,14684,14685,14686],{"id":5452,"depth":1610,"text":13239},{"id":14567,"depth":1610,"text":14568},{"id":14593,"depth":1610,"text":14594},{"id":1564,"depth":1610,"text":1565},"CAGR45%超で急成長するドローン物流市場。Zipline、Wing、Matternet、日本勢など主要プレイヤーの事業構造をCVC視点で解剖し、短期・中期の投資シナリオを提示する。",{"date":14689,"image":13492,"alt":14690,"tags":14691,"tagsEn":14692,"published":1666},"29th Mar 2026","Fly Logisticsのドローン配送センターと都市上空を飛ぶ配送ドローン群",[4240,1660,1658,5599,4241],[4245,1665,1663,5599,4241],"/blogs/4-drone-logistics",{"title":13447,"description":14687},"blogs/4-drone-logistics","iTcM9v4UZvRgdlNKLf1zvSL_tiMq0CMl3H-pQm6zUGQ",{"id":14698,"title":14699,"body":14700,"description":16108,"extension":1651,"meta":16109,"navigation":1666,"ogImage":14748,"path":16113,"seo":16114,"stem":16115,"__hash__":16116},"content/blogs/5-space-startup-trends.md","宇宙スタートアップ投資が示す8つのメガトレンド——「地球の延長」から「宇宙の産業化」へ | 8 Megatrends in Space Startup Investing: From Earth Extension to Space Industrialization",{"type":7,"value":14701,"toc":16070},[14702,15387,15389],[10,14703,14704,14707,14718,14722,14725,14732,14743,14749,14751,14755,14762,14773,14775,14779,14786,14869,14880,14887,14889,14893,14896,14900,14911,14915,14918,15005,15007,15011,15024,15031,15033,15037,15040,15047,15049,15053,15060,15067,15069,15073,15080,15087,15089,15093,15104,15111,15113,15116,15119,15123,15252,15254,15323,15326,15333,15344,15346,15378,15381,15383],{"lang":12},[14,14705,14706],{"id":14706},"宇宙スタートアップ投資が示す8つのメガトレンド",[19,14708,14709,14710,14713,14714,14717],{},"私はCVCで自社技術の棚卸しとスタートアップ技術の系統分類・シナジーマップ作成を担当した経験があります。",[23,14711,14712],{},"「テーマ単位ではなく、互いにレバレッジが効く塊として見る」","——これが投資テーマを設計するときの基本姿勢でした。今回の宇宙スタートアップも、個別の8トレンドを並べるのではなく、",[23,14715,14716],{},"どの組み合わせがバリューチェーンとして連動するか","で読み解いてみたいと思っています。",[32,14719,14721],{"id":14720},"宇宙vcの投資先が語る次の10年","宇宙VCの投資先が語る「次の10年」",[19,14723,14724],{},"最初に、業界が今どこに立っているかをフラットに置きます。",[19,14726,14727,14728,14731],{},"宇宙産業は転換点にあります。SpaceXが打ち上げコストを劇的に引き下げ、NASAのArtemisプログラムが月面への有人帰還を準備し、民間資本が加速的に流入するなか、宇宙は「探検のフロンティア」から**「産業化のプラットフォーム」",[23,14729,14730],{},"へと変貌しつつあります。PwCの分析では、宇宙経済全体は2034年に","$1.7兆**、2064年には**$6.1兆**に達する見通し。",[19,14733,14734,14735,14742],{},"ここで重要なのは、伸びるのが「打ち上げや衛星通信といった従来型」ではなく、",[23,14736,14737,14738,14741],{},"「宇宙空間そのものを生産・製造・資源調達の場として活用する産業化レイヤー」",[23,14739,14740],{},"だという点です。宇宙特化VCの投資パターンを俯瞰すると、この産業化を構成する","8つのメガトレンド","が鮮明に浮かび上がります。",[19,14744,14745],{},[4681,14746],{"alt":14747,"src":14748},"地球の延長から宇宙の産業化プラットフォームへの変革を示すインフォグラフィック","/blogs-img/blog-space-trends.png",[205,14750],{},[32,14752,14754],{"id":14753},"トレンド1商業宇宙ステーションiss後の軌道上不動産","トレンド1：商業宇宙ステーション——ISS後の「軌道上不動産」",[19,14756,14757,14758,14761],{},"ISSは2030年代前半の運用終了が見込まれており、その後継として",[23,14759,14760],{},"民間主導の商業宇宙ステーション","の開発競争が激化しています。NASAはCommercial LEO Destinations（CLD）プログラムを通じて複数の企業に資金を投じ、「宇宙の不動産」を民間に移管する方針を明確にしています。",[19,14763,14764,14765,14768,14769,14772],{},"商業ステーションの用途はISS時代と根本的に違います。科学研究だけでなく、",[23,14766,14767],{},"製造、バイオテク、メディア制作、宇宙観光","など多目的な「軌道上プラットフォーム」としての活用が想定されている。建設には従来の航空宇宙の製造手法ではなく、",[23,14770,14771],{},"造船所型の大規模モジュール製造技術","が必要です。私が普段見ている海運脱炭素のサプライチェーンと、ここの製造プロセスは構造的にかなり似通っている、というのが個人的な発見でした。",[205,14774],{},[32,14776,14778],{"id":14777},"トレンド2宇宙空間製造微小重力が生む地上では作れないもの","トレンド2：宇宙空間製造——微小重力が生む「地上では作れないもの」",[19,14780,14781,14782,14785],{},"宇宙空間での製造（In-Space Manufacturing, ISM）は、2025年時点で**$2.2B",[23,14783,14784],{},"の市場規模を持ち、2030年に","$5.2B**、2040年には**$62.8〜$135B**に成長すると予測されています（調査会社により幅あり）。CAGR 19〜30%の高成長セグメント。",[54,14787,14788,14805],{},[57,14789,14790],{},[60,14791,14792,14795,14797,14800,14803],{},[63,14793,14794],{},"ISM市場予測",[63,14796,228],{},[63,14798,14799],{},"2030年",[63,14801,14802],{},"2040年",[63,14804,237],{},[70,14806,14807,14823,14839,14855],{},[60,14808,14809,14812,14814,14817,14820],{},[75,14810,14811],{},"Allied Market Research",[75,14813,250],{},[75,14815,14816],{},"$21.3B（ISM+サービシング+輸送）",[75,14818,14819],{},"$135.3B",[75,14821,14822],{},"20.3%",[60,14824,14825,14828,14830,14833,14836],{},[75,14826,14827],{},"MarketsandMarkets",[75,14829,250],{},[75,14831,14832],{},"$4.6B",[75,14834,14835],{},"$62.8B",[75,14837,14838],{},"29.7%",[60,14840,14841,14844,14847,14850,14852],{},[75,14842,14843],{},"Research & Markets",[75,14845,14846],{},"$2.18B",[75,14848,14849],{},"$5.23B",[75,14851,250],{},[75,14853,14854],{},"19.1%",[60,14856,14857,14860,14863,14865,14867],{},[75,14858,14859],{},"商業セクター比率",[75,14861,14862],{},"58.7%",[75,14864,250],{},[75,14866,250],{},[75,14868,250],{},[19,14870,14871,14872,14875,14876,14879],{},"地球上では重力によって結晶構造が歪むタンパク質の結晶成長、対流の影響を受けない超高純度の光ファイバー製造、均質な合金の鋳造——これらは微小重力環境でのみ最適化できます。特に",[23,14873,14874],{},"医薬品","と",[23,14877,14878],{},"先端半導体材料","の分野で、地上では再現不可能な品質の製品を宇宙で生産し地球に帰還させるビジネスモデルが立ち上がりつつあります。",[19,14881,14882,14883,14886],{},"このビジネスの核心は、宇宙で製造した高付加価値材料を地球に安全に持ち帰る",[23,14884,14885],{},"再突入能力","にあります。打ち上げコストの低下と相まって「宇宙工場→地球への出荷」というサプライチェーンの経済性が現実味を帯びてきた、というのが現在地です。",[205,14888],{},[32,14890,14892],{"id":14891},"トレンド3月面経済水インフラモビリティの三位一体","トレンド3：月面経済——水・インフラ・モビリティの三位一体",[19,14894,14895],{},"月面経済はもはやSFではない、というのが私の基本的な見方です。PwCの推計では、月面活動からの累積収益は2026〜2050年で**$93.9B〜$127.3B**に達する可能性。",[122,14897,14899],{"id":14898},"月の水宇宙の石油","月の水——宇宙の「石油」",[19,14901,14902,14903,14906,14907,14910],{},"月面経済の最大のドライバーは",[23,14904,14905],{},"月極域の水氷","です。電気分解して水素と酸素に分離すればロケット推進剤になるので、月面での燃料生産（ISRU: In-Situ Resource Utilization）が実現すれば、深宇宙探査の経済性が根本から変わる。月面ISRU水抽出市場は2024年に**$1.1B**、2033年に**$6.5B**（CAGR 18.7%）に成長すると予測されています。Artemis Accordsの署名国は2026年1月時点で",[23,14908,14909],{},"61カ国","に達し、月面資源利用の国際法的フレームワークも整備されつつあります。",[122,14912,14914],{"id":14913},"月面モビリティローバーは宇宙のピックアップトラック","月面モビリティ——ローバーは「宇宙のピックアップトラック」",[19,14916,14917],{},"月面ローバー市場は2025年に**$3.8B**、2034年に**$12.6B**（CAGR 14.2%）に成長見通し。NASAはLunar Terrain Vehicle Services（LTVS）契約に最大$4.6Bを投じ、2名の宇宙飛行士が14日間・1回の遠征で20km走行できる無加圧ローバーの開発を委託しています。",[54,14919,14920,14935],{},[57,14921,14922],{},[60,14923,14924,14927,14929,14932],{},[63,14925,14926],{},"月面経済の主要市場",[63,14928,228],{},[63,14930,14931],{},"2030年代",[63,14933,14934],{},"備考",[70,14936,14937,14951,14965,14978,14992],{},[60,14938,14939,14942,14945,14948],{},[75,14940,14941],{},"月面探査技術全体",[75,14943,14944],{},"$11.4B",[75,14946,14947],{},"$18.9B (2029)",[75,14949,14950],{},"CAGR 13.4%",[60,14952,14953,14956,14959,14962],{},[75,14954,14955],{},"月面ローバー",[75,14957,14958],{},"$3.8B",[75,14960,14961],{},"$12.6B (2034)",[75,14963,14964],{},"CAGR 14.2%",[60,14966,14967,14970,14972,14975],{},[75,14968,14969],{},"月面ISRU水抽出",[75,14971,1263],{},[75,14973,14974],{},"$6.5B (2033)",[75,14976,14977],{},"CAGR 18.7%",[60,14979,14980,14983,14986,14989],{},[75,14981,14982],{},"宇宙マイニング全体",[75,14984,14985],{},"$3.07B (2026)",[75,14987,14988],{},"$7.4B (2031)",[75,14990,14991],{},"CAGR 19.2%",[60,14993,14994,14997,14999,15002],{},[75,14995,14996],{},"PwC月面累積収益",[75,14998,250],{},[75,15000,15001],{},"$93.9〜$127.3B (2026-2050)",[75,15003,15004],{},"5分野合計",[205,15006],{},[32,15008,15010],{"id":15009},"トレンド4軌道上サービシング宇宙のガソリンスタンド","トレンド4：軌道上サービシング——「宇宙のガソリンスタンド」",[19,15012,15013,15014,7758,15017,7758,15020,15023],{},"衛星の寿命は燃料の枯渇によって制約されるケースが多い。軌道上で衛星に燃料を補給できれば、寿命を数年〜十年単位で延長できます。OSAM（On-orbit Servicing, Assembly, and Manufacturing）市場は急成長していて、衛星の寿命延長サービスだけでなく、",[23,15015,15016],{},"デブリ除去",[23,15018,15019],{},"軌道上点検・修理",[23,15021,15022],{},"軌道間輸送","といった多機能なサービスが商業化されつつあります。宇宙デブリ除去市場は単独で**$750M**規模と推定。",[19,15025,15026,15027,15030],{},"ビジネスモデルは石油産業のインフラに近い、というのが個人的な解釈です。燃料を宇宙空間にデポ（貯蔵庫）として配備し、必要に応じて顧客衛星にサービスを提供する。将来的には月面で生産した推進剤を軌道上デポに供給するという、",[23,15028,15029],{},"月面ISRU→軌道上サービシングのバリューチェーン統合","も視野に入る。",[205,15032],{},[32,15034,15036],{"id":15035},"トレンド5再利用型宇宙アクセス飛行機のように宇宙へ","トレンド5：再利用型宇宙アクセス——「飛行機のように宇宙へ」",[19,15038,15039],{},"SpaceXのFalcon 9やStarshipが打ち上げコストを激減させましたが、次のパラダイムシフトは**「飛行機のように離着陸するスペースプレーン」**です。水平離着陸（HTHL）の宇宙機は、既存の空港インフラを活用でき、ターンアラウンドタイムの劇的な短縮が期待できます。",[19,15041,15042,15043,15046],{},"この技術が実現すれば、宇宙へのアクセスは「ロケット発射」から「フライト予約」へと変貌します。",[23,15044,15045],{},"地球への再突入能力","を持つ専用ビークルの開発も重要なサブテーマで、宇宙製造のバリューチェーンにおける「ラストワンマイル」を担います。",[205,15048],{},[32,15050,15052],{"id":15051},"トレンド6衛星iotすべてのbluetoothデバイスを宇宙から接続する","トレンド6：衛星IoT——すべてのBluetoothデバイスを宇宙から接続する",[19,15054,15055,15056,15059],{},"地上の通信インフラが届かない場所——海洋、砂漠、極地、農村部——にある何十億ものIoTデバイスを衛星経由で接続するというコンセプトが現実化しつつあります。最新の技術では",[23,15057,15058],{},"既存のBluetoothチップ","を搭載した一般的なデバイスを、特別なハードウェア変更なしに衛星から直接接続できるようになってきた。",[19,15061,15062,15063,15066],{},"私はAWS IoTで河川洪水検知システムをゼロから組んだ経験があるのですが、",[23,15064,15065],{},"カバレッジの空白地帯がIoTのボトルネック","だったのは、この実装で痛感したことでした。衛星IoTがそこを潰す——というのは、ITインフラ側の人間としても投資する側としても腹落ちする話です。SATCOM on the moveが2026年に**$42.8B**（CAGR 19.3%）と予測されており、その中でもIoT特化の低軌道衛星コンステレーションは最速成長セグメントの一つです。",[205,15068],{},[32,15070,15072],{"id":15071},"トレンド7宇宙放射線シールド長期滞在の見えない壁を突破する","トレンド7：宇宙放射線シールド——長期滞在の「見えない壁」を突破する",[19,15074,15075,15076,15079],{},"月面や深宇宙での長期滞在において、最大の技術的障壁の一つが",[23,15077,15078],{},"宇宙放射線","です。銀河宇宙線（GCR）や太陽粒子線（SPE）は、地磁気に守られた地球低軌道とは比較にならないレベルで人体や電子機器にダメージを与える。",[19,15081,15082,15083,15086],{},"従来の遮蔽手法——金属板や水タンクによる物理的シールド——では重量が課題でした。軽量かつ高性能な",[23,15084,15085],{},"新素材による放射線シールド","が切望されており、この分野は月面経済、商業ステーション、深宇宙探査のすべてに横断的に関わるイネーブリング・テクノロジー。市場浸透度が低い今こそ投資のタイミングだと見ています。",[205,15088],{},[32,15090,15092],{"id":15091},"トレンド8宇宙農業バイオテック宇宙の生態系を作る","トレンド8：宇宙農業・バイオテック——「宇宙の生態系」を作る",[19,15094,15095,15096,15099,15100,15103],{},"長期的な宇宙滞在を支えるためには、食料・医薬品・バイオマテリアルの現地生産が不可欠です。",[23,15097,15098],{},"制御環境型バイオファーム","の技術は、宇宙ステーションや月面基地での食料生産に直結。応用範囲は宇宙に限らず、地球上でも気候変動に対応した垂直農業、希少植物由来の医薬品生産など、",[23,15101,15102],{},"地球と宇宙の両方にまたがるデュアルユース","のビジネスモデルが構築可能です。",[19,15105,15106,15107,15110],{},"微小重力環境を活用した",[23,15108,15109],{},"バイオメディカル研究","も有望で、「宇宙の病院から地球の家庭へ」というコンセプトで、宇宙環境での創薬研究や組織工学を地上医療にフィードバックするモデルが模索されています。",[205,15112],{},[32,15114,15115],{"id":15115},"投資フレームワークと今後の展望",[19,15117,15118],{},"私の中ではここがいちばん大事なところです。8トレンドを並べて見せるだけでは投資判断にならないので、**「相互依存性」と「政府アンカー需要」**の2軸で整理します。",[122,15120,15122],{"id":15121},"_8つのトレンドの位置づけ","8つのトレンドの位置づけ",[54,15124,15125,15141],{},[57,15126,15127],{},[60,15128,15129,15132,15135,15138],{},[63,15130,15131],{},"トレンド",[63,15133,15134],{},"市場成熟度",[63,15136,15137],{},"主な市場規模",[63,15139,15140],{},"投資ステージ",[70,15142,15143,15157,15171,15185,15199,15211,15225,15239],{},[60,15144,15145,15148,15151,15154],{},[75,15146,15147],{},"商業宇宙ステーション",[75,15149,15150],{},"開発期",[75,15152,15153],{},"ISS後継需要$B規模",[75,15155,15156],{},"Series A-C",[60,15158,15159,15162,15165,15168],{},[75,15160,15161],{},"宇宙空間製造",[75,15163,15164],{},"実証→初期商業",[75,15166,15167],{},"$2.2B→$5.2B (2030)",[75,15169,15170],{},"Seed-Series A",[60,15172,15173,15176,15179,15182],{},[75,15174,15175],{},"月面経済",[75,15177,15178],{},"探査→商業準備",[75,15180,15181],{},"$11.4B (探査技術, 2025)",[75,15183,15184],{},"Pre-seed-Series A",[60,15186,15187,15190,15193,15196],{},[75,15188,15189],{},"軌道上サービシング",[75,15191,15192],{},"初期商業",[75,15194,15195],{},"$2.6B (2026)",[75,15197,15198],{},"Series A-B",[60,15200,15201,15204,15206,15209],{},[75,15202,15203],{},"再利用型宇宙アクセス",[75,15205,15150],{},[75,15207,15208],{},"打ち上げ市場全体$10B+",[75,15210,15170],{},[60,15212,15213,15216,15219,15222],{},[75,15214,15215],{},"衛星IoT",[75,15217,15218],{},"商業化初期",[75,15220,15221],{},"SATCOM全体$42.8B (2026)",[75,15223,15224],{},"Post-seed-Series A",[60,15226,15227,15230,15233,15236],{},[75,15228,15229],{},"宇宙放射線シールド",[75,15231,15232],{},"基礎研究→実証",[75,15234,15235],{},"横断的イネーブラー",[75,15237,15238],{},"Seed",[60,15240,15241,15244,15247,15250],{},[75,15242,15243],{},"宇宙農業・バイオテック",[75,15245,15246],{},"実証期",[75,15248,15249],{},"デュアルユース",[75,15251,15170],{},[122,15253,4797],{"id":4797},[54,15255,15256,15269],{},[57,15257,15258],{},[60,15259,15260,15262,15265,15267],{},[63,15261,601],{},[63,15263,15264],{},"宇宙経済規模（2035年）",[63,15266,4808],{},[63,15268,4811],{},[70,15270,15271,15284,15297,15310],{},[60,15272,15273,15275,15278,15281],{},[75,15274,4818],{},[75,15276,15277],{},"$600〜$800B",[75,15279,15280],{},"Artemis有人月面着陸が2027年に実現。商業ステーション1基が運用開始",[75,15282,15283],{},"月面関連・ISMが最初のスケーリングフェーズへ",[60,15285,15286,15288,15291,15294],{},[75,15287,4829],{},[75,15289,15290],{},"$1.0〜$1.5T",[75,15292,15293],{},"複数の商業ステーションが並行運用。月面ISRUが初の商業生産を達成",[75,15295,15296],{},"軌道上サービシング・月面インフラが本格的な収益化段階に",[60,15298,15299,15301,15304,15307],{},[75,15300,4840],{},[75,15302,15303],{},"$2T超",[75,15305,15306],{},"Starship完全運用+打ち上げコスト$100/kg以下。月面水の商業採掘開始",[75,15308,15309],{},"宇宙製造・月面経済が「新しいインターネット」級の市場創造",[60,15311,15312,15314,15317,15320],{},[75,15313,4851],{},[75,15315,15316],{},"$400B以下",[75,15318,15319],{},"Artemis大幅遅延。商業ステーション開発中止。規制強化",[75,15321,15322],{},"VC投資の冬。テーマ全体のバリュエーション調整",[122,15324,15325],{"id":15325},"投資家としてのポジショニング",[19,15327,15328,15329,15332],{},"8つのトレンドに共通するのは、いずれも",[23,15330,15331],{},"政府のアンカー需要（NASAのCLPS・LTVS・CLD契約等）が存在する","という点です。純粋な民間需要だけに依存するのではなく、政府調達がリスクを引き下げ、商業市場を育てるという構図は、過去のGPS産業やインターネットの発展経路と重なります。",[19,15334,15335,15336,15339,15340,15343],{},"注目すべきは、これらのトレンドが",[23,15337,15338],{},"互いに相互依存","していることです。月面の水がなければ軌道上燃料デポは成立しない。商業ステーションがなければ宇宙製造はスケールしない。再突入技術がなければ宇宙で製造したものを地球に届けられない。宇宙放射線シールドがなければ長期滞在は困難。",[23,15341,15342],{},"1つのテーマへの投資は、他のテーマの成功にもレバレッジが効く","——というのが、私のシナジーマップ作成の経験から見たいちばん本質的な構造です。",[122,15345,764],{"id":764},[312,15347,15348,15354,15360,15366,15372],{},[315,15349,15350,15353],{},[23,15351,15352],{},"Artemis III（有人月面着陸）のスケジュール"," — 2027年目標。月面経済の投資テーマが一気に加速するトリガー",[315,15355,15356,15359],{},[23,15357,15358],{},"Starshipの完全運用タイミング"," — 打ち上げコストの桁違いの低下は、宇宙産業全体のゲームチェンジャー",[315,15361,15362,15365],{},[23,15363,15364],{},"FAAのBVLOS・再突入規制の進展"," — 商業的な貨物再突入の規制枠組みが整備されれば、宇宙製造→地球帰還のビジネスモデルが本格化",[315,15367,15368,15371],{},[23,15369,15370],{},"ISS退役の正式日程"," — 2030年代前半。商業ステーション開発のデッドラインを設定",[315,15373,15374,15377],{},[23,15375,15376],{},"中国・インドの月面計画"," — 国際競争の激化は市場全体のパイを拡大するが、地政学リスクも内包",[19,15379,15380],{},"結論として、宇宙は**「次のインフラ投資」**です。20世紀にインターネットが「通信のインフラ」から「経済のインフラ」へと進化したように、宇宙は「探査の場」から「産業のプラットフォーム」へと不可逆的に進化しています。8つのメガトレンドが相互に補強し合いながら形成される宇宙経済は、今後10〜20年で最も大きなα（超過リターン）を生み出すテーマの一つだと見ています。",[205,15382],{},[19,15384,15385],{},[802,15386,4907],{},[205,15388],{},[10,15390,15391,15395,15405,15409,15412,15425,15436,15441,15443,15447,15454,15465,15467,15471,15485,15543,15553,15560,15562,15566,15572,15576,15591,15595,15605,15684,15686,15690,15703,15710,15712,15716,15723,15730,15732,15736,15743,15754,15756,15760,15767,15774,15776,15780,15787,15793,15795,15799,15805,15809,15934,15936,16005,16009,16016,16023,16025,16057,16064,16066],{"lang":809},[14,15392,15394],{"id":15393},"_8-megatrends-in-space-startup-investing","8 Megatrends in Space Startup Investing",[19,15396,15397,15398,15401,15402,849],{},"I spent part of my CVC role mapping our parent company's tech stack and our portfolio's startup technologies into a synergy map. The biggest takeaway: ",[23,15399,15400],{},"don't think theme-by-theme; think in clusters where one bet leverages another",". So with space startups, I want to read the eight trends not as a list but as ",[23,15403,15404],{},"a graph of mutual dependencies",[32,15406,15408],{"id":15407},"what-space-vcs-are-betting-on","What Space VCs Are Betting On",[19,15410,15411],{},"Let me put the picture down flat first.",[19,15413,15414,15415,15418,15419,10338,15422,849],{},"The space industry is at an inflection point. SpaceX has dramatically lowered launch costs, NASA's Artemis is preparing for crewed lunar return, and private capital is flowing in at speed. Space is shifting from an \"exploration frontier\" into an ",[23,15416,15417],{},"\"industrial platform\"",". PwC projects the total space economy at ",[23,15420,15421],{},"$1.7T by 2034",[23,15423,15424],{},"$6.1T by 2064",[19,15426,15427,15428,15431,15432,15435],{},"The growth is not in the legacy \"launch + comms\" stack — it's in the ",[23,15429,15430],{},"industrialization layer that uses space itself as a place to produce, manufacture, and source materials",". Looking across space VC portfolios, ",[23,15433,15434],{},"eight megatrends"," stand out clearly.",[19,15437,15438],{},[4681,15439],{"alt":15440,"src":14748},"Infographic showing the transformation from Earth's extension to space's industrial platform",[205,15442],{},[32,15444,15446],{"id":15445},"trend-1-commercial-space-stations-orbital-real-estate-after-iss","Trend 1: Commercial Space Stations — \"Orbital Real Estate\" After ISS",[19,15448,15449,15450,15453],{},"ISS retires in the early 2030s, and the race for ",[23,15451,15452],{},"privately-led commercial stations"," is intensifying. NASA's Commercial LEO Destinations (CLD) program is funding multiple companies and openly transferring \"space real estate\" to the private sector.",[19,15455,15456,15457,15460,15461,15464],{},"Use cases differ structurally from ISS-era research stations. These are multipurpose orbital platforms: ",[23,15458,15459],{},"manufacturing, biotech, media production, space tourism",". Construction needs ",[23,15462,15463],{},"shipyard-scale modular manufacturing",", not aerospace-style serial assembly. Coming from maritime work, the supply-chain shape here looks structurally familiar — that's actually one of my better personal anchors for understanding it.",[205,15466],{},[32,15468,15470],{"id":15469},"trend-2-in-space-manufacturing-what-cant-be-made-on-earth","Trend 2: In-Space Manufacturing — \"What Can't Be Made on Earth\"",[19,15472,15473,15474,15477,15478,10338,15481,15484],{},"ISM is ",[23,15475,15476],{},"$2.2B in 2025",", projected at ",[23,15479,15480],{},"$5.2B by 2030",[23,15482,15483],{},"$62.8–$135B by 2040"," (forecasts vary), at 19–30% CAGR.",[54,15486,15487,15504],{},[57,15488,15489],{},[60,15490,15491,15494,15496,15499,15502],{},[63,15492,15493],{},"ISM forecast",[63,15495,1018],{},[63,15497,15498],{},"2030",[63,15500,15501],{},"2040",[63,15503,237],{},[70,15505,15506,15519,15531],{},[60,15507,15508,15510,15512,15515,15517],{},[75,15509,14811],{},[75,15511,250],{},[75,15513,15514],{},"$21.3B (ISM + servicing + xport)",[75,15516,14819],{},[75,15518,14822],{},[60,15520,15521,15523,15525,15527,15529],{},[75,15522,14827],{},[75,15524,250],{},[75,15526,14832],{},[75,15528,14835],{},[75,15530,14838],{},[60,15532,15533,15535,15537,15539,15541],{},[75,15534,14843],{},[75,15536,14846],{},[75,15538,14849],{},[75,15540,250],{},[75,15542,14854],{},[19,15544,15545,15546,10338,15549,15552],{},"Microgravity enables products you cannot make on Earth: protein crystal growth without gravitational distortion, ultra-pure fiber optic without convection, homogeneous alloy castings. ",[23,15547,15548],{},"Pharmaceuticals",[23,15550,15551],{},"advanced semiconductor materials"," are where the \"make-in-space, return-to-Earth\" model is starting to pencil out economically.",[19,15554,15555,15556,15559],{},"The pivotal capability here is ",[23,15557,15558],{},"safe re-entry of high-value cargo",". Combined with falling launch costs, the \"space factory → Earth shipping\" supply chain is finally crossing into commercial-economics territory.",[205,15561],{},[32,15563,15565],{"id":15564},"trend-3-lunar-economy-water-infrastructure-mobility","Trend 3: Lunar Economy — Water, Infrastructure, Mobility",[19,15567,15568,15569,849],{},"The lunar economy is no longer science fiction — that's the framing I'd most defend. PwC estimates cumulative lunar revenue of ",[23,15570,15571],{},"$93.9B–$127.3B from 2026–2050",[122,15573,15575],{"id":15574},"lunar-water-the-oil-of-space","Lunar water — the \"oil\" of space",[19,15577,15578,15579,15582,15583,15586,15587,15590],{},"The single biggest driver is ",[23,15580,15581],{},"water ice at the lunar poles",". Electrolyze it and you get hydrogen + oxygen propellant. With Earth-launch fuel costing tens of thousands per kg, on-Moon fuel production (ISRU) fundamentally changes deep-space economics. The lunar ISRU water market is forecast at ",[23,15584,15585],{},"$1.1B in 2024 → $6.5B in 2033"," (18.7% CAGR). Artemis Accord signatories reached ",[23,15588,15589],{},"61 nations"," by January 2026.",[122,15592,15594],{"id":15593},"lunar-mobility-rovers-as-space-pickup-trucks","Lunar mobility — rovers as \"space pickup trucks\"",[19,15596,15597,15598,15601,15602,15604],{},"The lunar rover market grows from ",[23,15599,15600],{},"$3.8B in 2025 → $12.6B by 2034"," (14.2% CAGR). NASA's LTVS contract commits up to ",[23,15603,14832],{}," for unpressurized rovers carrying 2 astronauts 20 km per sortie over 14 days.",[54,15606,15607,15621],{},[57,15608,15609],{},[60,15610,15611,15614,15616,15619],{},[63,15612,15613],{},"Lunar economy markets",[63,15615,1018],{},[63,15617,15618],{},"2030s",[63,15620,9415],{},[70,15622,15623,15635,15647,15659,15671],{},[60,15624,15625,15628,15630,15632],{},[75,15626,15627],{},"Lunar exploration tech",[75,15629,14944],{},[75,15631,14947],{},[75,15633,15634],{},"13.4% CAGR",[60,15636,15637,15640,15642,15644],{},[75,15638,15639],{},"Lunar rovers",[75,15641,14958],{},[75,15643,14961],{},[75,15645,15646],{},"14.2% CAGR",[60,15648,15649,15652,15654,15656],{},[75,15650,15651],{},"Lunar ISRU water",[75,15653,1263],{},[75,15655,14974],{},[75,15657,15658],{},"18.7% CAGR",[60,15660,15661,15664,15666,15668],{},[75,15662,15663],{},"Space mining",[75,15665,14985],{},[75,15667,14988],{},[75,15669,15670],{},"19.2% CAGR",[60,15672,15673,15676,15678,15681],{},[75,15674,15675],{},"PwC cumulative lunar",[75,15677,250],{},[75,15679,15680],{},"$93.9–$127.3B (2026–2050)",[75,15682,15683],{},"5 segments",[205,15685],{},[32,15687,15689],{"id":15688},"trend-4-on-orbit-servicing-the-gas-station-of-space","Trend 4: On-Orbit Servicing — \"The Gas Station of Space\"",[19,15691,15692,15693,8946,15696,8946,15699,15702],{},"Satellite life is often capped by fuel depletion. Refueling on-orbit extends life by years, sometimes decades. OSAM (on-orbit servicing, assembly, and manufacturing) is expanding fast — life extension, ",[23,15694,15695],{},"debris removal",[23,15697,15698],{},"inspection and repair",[23,15700,15701],{},"orbit transfer",". The debris-removal sub-market alone sits at ~$750M.",[19,15704,15705,15706,15709],{},"The business model rhymes with the oil industry. Pre-positioned propellant depots in orbit, on-demand service to client satellites — and longer-term, ",[23,15707,15708],{},"lunar-ISRU-to-orbit-depot"," value-chain integration is on the horizon.",[205,15711],{},[32,15713,15715],{"id":15714},"trend-5-reusable-space-access-to-space-like-an-airplane","Trend 5: Reusable Space Access — \"To Space Like an Airplane\"",[19,15717,15718,15719,15722],{},"Falcon 9 and Starship cut launch cost. The next paradigm shift is ",[23,15720,15721],{},"horizontal-takeoff-and-landing spaceplanes",": existing airport infrastructure, dramatically shorter turnaround.",[19,15724,15725,15726,15729],{},"If realized, space access flips from \"rocket launch\" to \"flight booking.\" A critical sub-theme is ",[23,15727,15728],{},"purpose-built re-entry vehicles"," — the \"last mile\" of the space-manufacturing supply chain that brings high-value goods home safely.",[205,15731],{},[32,15733,15735],{"id":15734},"trend-6-satellite-iot-connecting-every-bluetooth-device-from-space","Trend 6: Satellite IoT — Connecting Every Bluetooth Device From Space",[19,15737,15738,15739,15742],{},"Billions of IoT devices in places that terrestrial networks don't reach — oceans, deserts, polar regions, rural — are about to become directly addressable from satellites. Recent tech allows ",[23,15740,15741],{},"off-the-shelf Bluetooth chips"," to connect to satellites without any hardware modification.",[19,15744,15745,15746,15749,15750,15753],{},"I built a flood-detection IoT system on AWS (IoT Core, Lambda, DynamoDB, API Gateway) from scratch, and the experience that stuck with me was that ",[23,15747,15748],{},"coverage gaps are the actual bottleneck, not silicon",". Satellite IoT closes that gap. SATCOM on-the-move is forecast at ",[23,15751,15752],{},"$42.8B by 2026"," (19.3% CAGR), with IoT-focused LEO constellations as the fastest-growing sub-segment.",[205,15755],{},[32,15757,15759],{"id":15758},"trend-7-space-radiation-shielding-breaking-the-invisible-wall","Trend 7: Space Radiation Shielding — Breaking the \"Invisible Wall\"",[19,15761,15762,15763,15766],{},"For long-duration stays on the Moon and in deep space, the biggest single technical wall is ",[23,15764,15765],{},"radiation",". Galactic cosmic rays (GCR) and solar particle events (SPE) damage humans and electronics at levels the magnetosphere-protected LEO simply doesn't see.",[19,15768,15769,15770,15773],{},"Traditional shielding (metal plates, water tanks) carries an unaffordable mass penalty. ",[23,15771,15772],{},"Lightweight high-performance materials"," are the missing piece. This is a horizontal enabler — it gates the lunar economy, commercial stations, and deep space exploration simultaneously. With low penetration today, the entry timing is, in my view, particularly interesting.",[205,15775],{},[32,15777,15779],{"id":15778},"trend-8-space-agriculture-biotech-building-a-space-ecosystem","Trend 8: Space Agriculture & Biotech — Building a \"Space Ecosystem\"",[19,15781,15782,15783,15786],{},"Long-duration missions need on-site production of food, medicine, and biomaterials. ",[23,15784,15785],{},"Controlled-environment biofarming"," maps directly onto stations and lunar bases — but the use cases are dual: climate-resilient vertical farming on Earth, rare-plant pharmaceutical production, specialized bioprocesses in simulated space conditions.",[19,15788,15789,15792],{},[23,15790,15791],{},"Microgravity biomedical research"," is parallel: protein crystal growth and organoid culture in space could accelerate drug development on Earth. The \"from space hospital to Earth home\" framing is being actively explored.",[205,15794],{},[32,15796,15798],{"id":15797},"investment-framework-outlook","Investment Framework & Outlook",[19,15800,15801,15802,849],{},"The most important section. Reading the eight trends as a list under-prices the structure — they need to be read on ",[23,15803,15804],{},"two axes: mutual dependency, and government anchor demand",[122,15806,15808],{"id":15807},"positioning-the-8-trends","Positioning the 8 trends",[54,15810,15811,15827],{},[57,15812,15813],{},[60,15814,15815,15818,15821,15824],{},[63,15816,15817],{},"Trend",[63,15819,15820],{},"Market maturity",[63,15822,15823],{},"Key market size",[63,15825,15826],{},"Investment stage",[70,15828,15829,15842,15856,15870,15883,15895,15908,15921],{},[60,15830,15831,15834,15836,15839],{},[75,15832,15833],{},"Commercial space stations",[75,15835,14505],{},[75,15837,15838],{},"Post-ISS demand, $B-scale",[75,15840,15841],{},"Series A–C",[60,15843,15844,15847,15850,15853],{},[75,15845,15846],{},"In-space manufacturing",[75,15848,15849],{},"Demo → early commercial",[75,15851,15852],{},"$2.2B → $5.2B (2030)",[75,15854,15855],{},"Seed–Series A",[60,15857,15858,15861,15864,15867],{},[75,15859,15860],{},"Lunar economy",[75,15862,15863],{},"Exploration → commercial prep",[75,15865,15866],{},"$11.4B (exp tech, 2025)",[75,15868,15869],{},"Pre-seed–Series A",[60,15871,15872,15875,15878,15880],{},[75,15873,15874],{},"On-orbit servicing",[75,15876,15877],{},"Early commercial",[75,15879,15195],{},[75,15881,15882],{},"Series A–B",[60,15884,15885,15888,15890,15893],{},[75,15886,15887],{},"Reusable space access",[75,15889,14505],{},[75,15891,15892],{},"Launch market $10B+ total",[75,15894,15855],{},[60,15896,15897,15900,15902,15905],{},[75,15898,15899],{},"Satellite IoT",[75,15901,15877],{},[75,15903,15904],{},"SATCOM total $42.8B (2026)",[75,15906,15907],{},"Post-seed–Series A",[60,15909,15910,15913,15916,15919],{},[75,15911,15912],{},"Space radiation shielding",[75,15914,15915],{},"Basic research → demo",[75,15917,15918],{},"Cross-cutting enabler",[75,15920,15238],{},[60,15922,15923,15926,15929,15932],{},[75,15924,15925],{},"Space agriculture & biotech",[75,15927,15928],{},"Demo",[75,15930,15931],{},"Dual-use",[75,15933,15855],{},[122,15935,13239],{"id":5452},[54,15937,15938,15951],{},[57,15939,15940],{},[60,15941,15942,15944,15947,15949],{},[63,15943,1389],{},[63,15945,15946],{},"Space economy (2035)",[63,15948,14505],{},[63,15950,5467],{},[70,15952,15953,15966,15979,15992],{},[60,15954,15955,15957,15960,15963],{},[75,15956,6951],{},[75,15958,15959],{},"$600–$800B",[75,15961,15962],{},"Artemis crewed lunar landing in 2027; one commercial station operational",[75,15964,15965],{},"Lunar/ISM enter their first scaling phase",[60,15967,15968,15970,15973,15976],{},[75,15969,14527],{},[75,15971,15972],{},"$1.0–$1.5T",[75,15974,15975],{},"Multiple commercial stations operating; lunar ISRU achieves first commercial production",[75,15977,15978],{},"OSAM and lunar infrastructure reach full monetization",[60,15980,15981,15983,15986,15989],{},[75,15982,14541],{},[75,15984,15985],{},"$2T+",[75,15987,15988],{},"Full Starship ops + sub-$100/kg launch; commercial lunar water mining",[75,15990,15991],{},"Space manufacturing/lunar economy reach \"new internet\"-scale markets",[60,15993,15994,15996,15999,16002],{},[75,15995,14555],{},[75,15997,15998],{},"\u003C$400B",[75,16000,16001],{},"Major Artemis delay; commercial stations cancelled; regulatory tightening",[75,16003,16004],{},"Space VC winter; thematic valuation correction",[122,16006,16008],{"id":16007},"positioning","Positioning",[19,16010,16011,16012,16015],{},"What unites all eight trends is ",[23,16013,16014],{},"government anchor demand"," — NASA CLPS, LTVS, CLD contracts, etc. The pattern of public procurement de-risking commercial scale-up rhymes with the early development of GPS and the internet. Pure private demand alone wouldn't carry the timeline.",[19,16017,16018,16019,16022],{},"The deeper structure is ",[23,16020,16021],{},"mutual dependency",". No lunar water → no orbital fuel depot. No commercial stations → no space manufacturing scale. No re-entry tech → can't deliver what you make in space. No radiation shielding → no long-duration human presence. Investing in any one trend leverages the success of the others. From building synergy maps inside CVC, this is the kind of structure that compounds rather than substitutes.",[122,16024,1565],{"id":1564},[312,16026,16027,16033,16039,16045,16051],{},[315,16028,16029,16032],{},[23,16030,16031],{},"Artemis III (crewed lunar landing) schedule"," — currently 2027; the trigger that re-rates lunar startups",[315,16034,16035,16038],{},[23,16036,16037],{},"Starship full operationalization"," — order-of-magnitude launch-cost drop is the industry-wide game changer",[315,16040,16041,16044],{},[23,16042,16043],{},"FAA BVLOS and re-entry rules"," — commercial cargo re-entry frameworks unlock the full space-manufacturing return loop",[315,16046,16047,16050],{},[23,16048,16049],{},"ISS retirement timeline"," — early 2030s; sets the deadline that pulls forward private investment",[315,16052,16053,16056],{},[23,16054,16055],{},"China and India lunar programs"," — international competition expands the pie but layers in geopolitical risk",[19,16058,16059,16060,16063],{},"Bottom line: space is ",[23,16061,16062],{},"the next infrastructure investment",". As the internet went from \"communications infrastructure\" to \"economic infrastructure\" in the 20th century, space is shifting from \"a place to explore\" to \"a platform for industry\" — and that move is now irreversible. The space economy, formed by these eight reinforcing megatrends, is, in my view, one of the largest sources of alpha for the next 10–20 years.",[205,16065],{},[19,16067,16068],{},[802,16069,13388],{},{"title":143,"searchDepth":1605,"depth":1605,"links":16071},[16072,16073,16074,16075,16079,16080,16081,16082,16083,16084,16090,16091,16092,16093,16097,16098,16099,16100,16101,16102],{"id":14720,"depth":1605,"text":14721},{"id":14753,"depth":1605,"text":14754},{"id":14777,"depth":1605,"text":14778},{"id":14891,"depth":1605,"text":14892,"children":16076},[16077,16078],{"id":14898,"depth":1610,"text":14899},{"id":14913,"depth":1610,"text":14914},{"id":15009,"depth":1605,"text":15010},{"id":15035,"depth":1605,"text":15036},{"id":15051,"depth":1605,"text":15052},{"id":15071,"depth":1605,"text":15072},{"id":15091,"depth":1605,"text":15092},{"id":15115,"depth":1605,"text":15115,"children":16085},[16086,16087,16088,16089],{"id":15121,"depth":1610,"text":15122},{"id":4797,"depth":1610,"text":4797},{"id":15325,"depth":1610,"text":15325},{"id":764,"depth":1610,"text":764},{"id":15407,"depth":1605,"text":15408},{"id":15445,"depth":1605,"text":15446},{"id":15469,"depth":1605,"text":15470},{"id":15564,"depth":1605,"text":15565,"children":16094},[16095,16096],{"id":15574,"depth":1610,"text":15575},{"id":15593,"depth":1610,"text":15594},{"id":15688,"depth":1605,"text":15689},{"id":15714,"depth":1605,"text":15715},{"id":15734,"depth":1605,"text":15735},{"id":15758,"depth":1605,"text":15759},{"id":15778,"depth":1605,"text":15779},{"id":15797,"depth":1605,"text":15798,"children":16103},[16104,16105,16106,16107],{"id":15807,"depth":1610,"text":15808},{"id":5452,"depth":1610,"text":13239},{"id":16007,"depth":1610,"text":16008},{"id":1564,"depth":1610,"text":1565},"宇宙関連VCの投資パターンから浮かび上がる8つの構造的トレンドを分析。月面経済、宇宙製造、軌道上サービシング、再突入技術など、次の10年を定義するテーマと投資機会をCVC視点で整理する。",{"date":14689,"image":14748,"alt":14747,"tags":16110,"tagsEn":16111,"published":1666},[4240,1660,5599,4241,10631],[4245,1665,5599,4241,16112],"Space","/blogs/5-space-startup-trends",{"title":14699,"description":16108},"blogs/5-space-startup-trends","GuERL4IkTrtvKXCQVvAJSzTOCNnp2XsAFMPvZXPwEUI",{"id":16118,"title":16119,"body":16120,"description":17977,"extension":1651,"meta":17978,"navigation":1666,"ogImage":17980,"path":17988,"seo":17989,"stem":17990,"__hash__":17991},"content/blogs/6-semiconductor-vc-series-1.md","Tier1 VCの半導体投資はAI需要のボトルネックへ収斂している｜連載①概観と主要プレイヤー | Semiconductor VC Series①: Converging on AI Bottlenecks",{"type":7,"value":16121,"toc":17939},[16122,17048,17050],[10,16123,16124,16128,16134,16137,16141,16144,16155,16162,16164,16167,16187,16189,16192,16252,16255,16302,16305,16312,16314,16318,16321,16327,16330,16332,16336,16442,16444,16448,16452,16455,16458,16484,16490,16494,16497,16499,16513,16518,16522,16525,16527,16541,16546,16550,16553,16555,16581,16586,16590,16593,16595,16615,16621,16623,16626,16932,16934,16937,16940,16943,17013,17016,17027,17029,17033,17036,17039,17042,17044],{"lang":12},[14,16125,16127],{"id":16126},"tier1-vcの半導体投資はai需要のボトルネックへ収斂している","Tier1 VCの半導体投資はAI需要のボトルネックへ収斂している",[19,16129,16130,16131,7215],{},"私はもともと半導体プロセス開発の現場の人間で、EEPROMの車載プロセス開発をしていました。半導体は「投資対象」である前に、私にとっては",[23,16132,16133],{},"自分の手で歩留まりを動かしてきた現場",[19,16135,16136],{},"その後CVCに移って、Tier1 VCがどこに賭けているかを観察するようになって気づいたのは、**「彼らはチップを買っているのではなく、AIの制約を買っている」**ということでした。今回の連載①では、その視点で5つのTier1 VC（Sequoia、a16z、Kleiner Perkins、Bessemer、SoftBank）の動きを整理します。",[32,16138,16140],{"id":16139},"半導体ブームの裏側で何が起きているか","半導体ブームの「裏側」で何が起きているか",[19,16142,16143],{},"最初に、市場の現在地をフラットに置いておきます。",[19,16145,16146,16147,16150,16151,16154],{},"2025年、世界半導体市場は過去最高となる**$791.7B（WSTS）",[23,16148,16149],{},"に達しました。2026年には","$975B",[23,16152,16153],{},"への拡大が見込まれ、AIインフラへの投資が市場全体を牽引しています。同時に、300mmファブ装置の支出は2026年に","$133B**、2027年には**$151B**に達するとSEMIは予測。",[19,16156,16157,16158,16161],{},"数字だけ見れば「半導体ブーム」に見えるが、Tier1 VCが実際に賭けているのはチップそのものではない。",[23,16159,16160],{},"計算能力の向上が生み出す「ボトルネック」","——設計スループット、データ移動（インターコネクト）、パッケージング、そして次世代の計算原理——こそが、トップVCの投資が集まる場所です。",[205,16163],{},[32,16165,16166],{"id":16166},"エグゼクティブサマリ",[312,16168,16169,16175,16181],{},[315,16170,16171,16174],{},[23,16172,16173],{},"成長ドライバーはAIインフラ。"," 市場規模の拡大と装置投資の増加は「供給制約が緩む」より「投資負荷が続く」シグナルに近い。スタートアップ側からすれば、量産・テスト能力の確保が評価の分岐点になります。",[315,16176,16177,16180],{},[23,16178,16179],{},"VCの差分は「ステージの取り方」と「提携構造」。"," 巨大シードでアーキテクチャ賭けに出る（a16z）、設計自動化で開発リードタイム短縮（Sequoia/Bessemer）、M&Aで垂直統合（SoftBank）——資本政策が根本的に異なります。",[315,16182,16183,16186],{},[23,16184,16185],{},"地理は依然米国中心。"," だが、インドのRISC-V/IP、欧州のAIアクセラレータ/フォトニクス、イスラエルのAIチップなど、サプライチェーン分散に沿った投資が可視化されています。",[205,16188],{},[32,16190,16191],{"id":16191},"世界半導体市場の現在地",[54,16193,16194,16210],{},[57,16195,16196],{},[60,16197,16198,16201,16204,16207],{},[63,16199,16200],{},"年",[63,16202,16203],{},"市場規模",[63,16205,16206],{},"前年比",[63,16208,16209],{},"主な成長領域",[70,16211,16212,16225,16238],{},[60,16213,16214,16217,16220,16222],{},[75,16215,16216],{},"2024",[75,16218,16219],{},"$611.2B",[75,16221,250],{},[75,16223,16224],{},"メモリ回復、ロジック",[60,16226,16227,16229,16232,16235],{},[75,16228,1018],{},[75,16230,16231],{},"$791.7B",[75,16233,16234],{},"+29.4%",[75,16236,16237],{},"AIデータセンター、HBM",[60,16239,16240,16243,16246,16249],{},[75,16241,16242],{},"2026（予測）",[75,16244,16245],{},"$975.0B",[75,16247,16248],{},"+23.2%",[75,16250,16251],{},"AIインフラ継続拡大",[19,16253,16254],{},"装置投資（300mmファブ・SEMI予測）：",[54,16256,16257,16268],{},[57,16258,16259],{},[60,16260,16261,16263,16266],{},[63,16262,16200],{},[63,16264,16265],{},"装置支出",[63,16267,12161],{},[70,16269,16270,16280,16291],{},[60,16271,16272,16274,16277],{},[75,16273,1018],{},[75,16275,16276],{},"〜$120B",[75,16278,16279],{},"AI需要立ち上がり",[60,16281,16282,16285,16288],{},[75,16283,16284],{},"2026",[75,16286,16287],{},"$133B",[75,16289,16290],{},"AI需要×各国供給網再編",[60,16292,16293,16296,16299],{},[75,16294,16295],{},"2027",[75,16297,16298],{},"$151B",[75,16300,16301],{},"供給投資の継続",[19,16303,16304],{},"この拡大は「半導体全体が儲かる」ではなく、Memory/Logicなど特定カテゴリに偏りやすい。EDA・テスト・パッケージングといった周辺領域は、ボトルネックとして間接的に恩恵を受ける構造です。",[19,16306,16307,16308,16311],{},"私が現場でDFM導入をやった経験から言うと、",[23,16309,16310],{},"装置投資のピーク期ほど、製造プロセス側の「目に見えない歩留まり改善余地」が利益の質を決める","——というのは普遍的な構造で、今回もEDA/検査/DFM周辺はインベスタブルだと感じています。",[205,16313],{},[32,16315,16317],{"id":16316},"aiインフラのボトルネック構造","AIインフラのボトルネック構造",[19,16319,16320],{},"GPUやアクセラレータの演算能力が伸び続けるほど、「その周辺」が限界要因になります。",[134,16322,16325],{"className":16323,"code":16324,"language":139},[137],"AIインフラ需要の増加\n        │\n        ▼ ボトルネックの再定義\n┌───────────┬─────────────┬──────────────┬────────────┐\n│ 設計生産性 │ データ移動   │ パッケージング │ 新計算原理─ │\n│ EDA/検証  │SerDes/光I/O │ CPO/Chiplet  │アナログ/量子─│\n└────┬─────┴─────┬───────┴───────┬──────┴─────┬───────┘\n     ↓           ↓               ↓            ↓\n シード〜A     Series B〜E     レイト          長期\n ソフト比率高  顧客/製造連携    資本集約     技術/標準不確実\n",[141,16326,16324],{"__ignoreMap":143},[19,16328,16329],{},"VCはこの4つの「制約領域」にリスク分散しながらポジションを取っています。",[205,16331],{},[32,16333,16335],{"id":16334},"vc別の型比較","VC別の\"型\"比較",[54,16337,16338,16357],{},[57,16339,16340],{},[60,16341,16342,16345,16348,16351,16354],{},[63,16343,16344],{},"VC",[63,16346,16347],{},"重点技術領域",[63,16349,16350],{},"ステージ傾向",[63,16352,16353],{},"地理",[63,16355,16356],{},"提携・LP構造",[70,16358,16359,16376,16392,16408,16425],{},[60,16360,16361,16364,16367,16370,16373],{},[75,16362,16363],{},"Sequoia",[75,16365,16366],{},"設計AI（EDA/検証）、RISC-V/IP、CPO",[75,16368,16369],{},"Seed〜Growth",[75,16371,16372],{},"米国・インド",[75,16374,16375],{},"設計AI領域に公式コミット。関連ビークルがCPOに参加",[60,16377,16378,16381,16384,16387,16389],{},[75,16379,16380],{},"a16z",[75,16382,16383],{},"新計算（アナログ/混載）、AIネットワーク",[75,16385,16386],{},"巨大Seed＋大型B",[75,16388,13781],{},[75,16390,16391],{},"Strategic Partnershipを明示、GTM設計が特徴",[60,16393,16394,16397,16400,16403,16405],{},[75,16395,16396],{},"Kleiner Perkins",[75,16398,16399],{},"RISC-V/IP、コヒレントDSP",[75,16401,16402],{},"Early〜Series D",[75,16404,13781],{},[75,16406,16407],{},"TSMC支援ファンド報道など「供給側との接続」示唆",[60,16409,16410,16413,16416,16419,16422],{},[75,16411,16412],{},"Bessemer",[75,16414,16415],{},"SerDes/IP、量子、EDA自動化",[75,16417,16418],{},"Series C中心＋アーリー",[75,16420,16421],{},"米国・欧州・イスラエル",[75,16423,16424],{},"AIチップ/量子/EDAで「深技術の商用化」",[60,16426,16427,16430,16433,16436,16439],{},[75,16428,16429],{},"SoftBank",[75,16431,16432],{},"AIチップ/CPU/DPU、垂直統合",[75,16434,16435],{},"Growth〜M&A",[75,16437,16438],{},"米国・英国",[75,16440,16441],{},"Arm核のAI Computing Segment。LPに戦略資本（Apple等）",[205,16443],{},[32,16445,16447],{"id":16446},"各vc解説","各VC解説",[122,16449,16451],{"id":16450},"sequoia設計運用ボトルネックを連結する","Sequoia：設計→運用ボトルネックを連結する",[19,16453,16454],{},"Sequoiaは「AIが高性能化するほど、チップ設計のスループットがボトルネックになる」という仮説を前面に出します。",[19,16456,16457],{},"代表案件：",[312,16459,16460,16466,16472,16478],{},[315,16461,16462,16465],{},[23,16463,16464],{},"Ricursive Intelligence（$35M、2025年シード、リード）","：AIによるチップ設計加速。Sequoiaが初回ラウンドをリードした象徴的案件。",[315,16467,16468,16471],{},[23,16469,16470],{},"InCore Semiconductors（$3M、2023年シード）","：インドのRISC-V IP。",[315,16473,16474,16477],{},[23,16475,16476],{},"Mindgrove Technologies（約$2.3M、2023年シード）","：インドSoCスタートアップ。",[315,16479,16480,16483],{},[23,16481,16482],{},"Ayar Labs（$500M、2026年Series E、関連ビークル参加）","：Co-Packaged Optics（CPO）。NVIDIA・AMD等の戦略投資家と共に参加。評価額$3.75B。",[19,16485,16486,16489],{},[23,16487,16488],{},"読み解き","：シード段階で「設計の生産性」を押さえつつ、大型レイトでは「次のデータ移動（光I/O）」にも資本を延ばす。設計から運用ボトルネックまでを連結する投資設計になっている、という構造です。",[122,16491,16493],{"id":16492},"a16z計算原理そのものへの巨大シード","a16z：計算原理そのものへの巨大シード",[19,16495,16496],{},"a16zは「計算原理を変えるような勝ち筋に、研究開発段階から大口でアンダーライトする」のが特徴です。",[19,16498,16457],{},[312,16500,16501,16507],{},[315,16502,16503,16506],{},[23,16504,16505],{},"Unconventional AI（$475M、2025年シード、共同リード）","：アナログ/ミックスドシグナルで確率的計算に最適化。",[315,16508,16509,16512],{},[23,16510,16511],{},"Nexthop AI（$500M、2026年Series B）","：AIクラスタ向けイーサネットスイッチ。",[19,16514,16515,16517],{},[23,16516,16488],{},"：同じAIインフラ投資でも、a16zは「カテゴリを再定義できる」と判断したらシード段階から数百億円を投下する。Strategic Partnerships（顧客・公共・大企業への導入経路の制度化）を明示している点も特徴で、資本だけでなくGTMを設計する色が濃い。",[122,16519,16521],{"id":16520},"kleiner-perkins水平方向のipデータセンター光","Kleiner Perkins：\"水平方向\"のIP×データセンター光",[19,16523,16524],{},"Kleiner PerkinsはAIインフラの水平方向——IP（RISC-V）とデータセンターの光/インターコネクト——に選球眼が集まっています。",[19,16526,16457],{},[312,16528,16529,16535],{},[315,16530,16531,16534],{},[23,16532,16533],{},"Akeana（累計$100M超、2024年ステルス解除）","：RISC-V CPU/IPポートフォリオ。",[315,16536,16537,16540],{},[23,16538,16539],{},"Retym（$75M、2025年Series D）","：クラウド/AI向けコヒレントDSP。",[19,16542,16543,16545],{},[23,16544,16488],{},"：ハイパースケーラーを顧客とする領域で「設計の差別化（IP/DSP）」を取りにいく投資スタイル。TSMC支援ファンドの立ち上げ報道は、「資金×ファブ供給（エコシステム）」の結びつきが強まる流れを示唆します。",[122,16547,16549],{"id":16548},"bessemer深技術の商用化を組み合わせて張る","Bessemer：深技術の商用化を組み合わせて張る",[19,16551,16552],{},"BessemerはSerDes/IP、量子、EDA自動化という一見バラバラな領域を、「深い技術負債と長いR&Dサイクルを伴うが、当たれば大きい」という共通軸で捉えています。",[19,16554,16457],{},[312,16556,16557,16563,16569,16575],{},[315,16558,16559,16562],{},[23,16560,16561],{},"ChipAgents（$21M、2025年Series A、リード）","：設計・検証のAIエージェント化。",[315,16564,16565,16568],{},[23,16566,16567],{},"Kandou（$92.3M、2020年Series C、リード）","：SerDes/IPとリタイマー。",[315,16570,16571,16574],{},[23,16572,16573],{},"Rigetti Computing（$79M、2020年Series C、リード）","：量子ハードウェア。",[315,16576,16577,16580],{},[23,16578,16579],{},"Habana Labs（$75M Series Bに参加、2018年、Intelが2019年に約$2B で買収）","：AIチップ。",[19,16582,16583,16585],{},[23,16584,16488],{},"：イスラエルネットワーク（Habana Labs）、スイス（Kandou）など、地域性も活用しながら「深技術×商用化の道筋」が立つ案件を選ぶ。",[122,16587,16589],{"id":16588},"softbank-vision-fundgroup垂直統合という別次元のリスクテイク","SoftBank Vision Fund/Group：垂直統合という別次元のリスクテイク",[19,16591,16592],{},"SoftBankは他のTier1 VCと決定的に異なります。ファンド投資だけでなく、買収を含む「垂直統合（Arm×周辺シリコン）」へ踏み込む。",[19,16594,16457],{},[312,16596,16597,16603,16609],{},[315,16598,16599,16602],{},[23,16600,16601],{},"Fungible（$200M、2019年Series C、SVFリード）","：DPU。後にMicrosoft傘下へ。",[315,16604,16605,16608],{},[23,16606,16607],{},"Graphcore（2024年買収、金額非開示）","：英国AIチップ。",[315,16610,16611,16614],{},[23,16612,16613],{},"Ampere Computing（$6.5B買収、2025年発表）","：ArmベースのクラウドサーバーCPU。",[19,16616,16617,16618,16620],{},"AI Computing SegmentとしてArm・Ampere・Graphcoreを同一セグメントで運営し、半導体事業を中核に据える姿勢を公式に明確化しています。",[23,16619,16488],{},"：これはVCというより「戦略金融×事業会社」の設計に近い。リスクテイクの桁も期間も通常のVCファンドとは異なります。",[205,16622],{},[32,16624,16625],{"id":16625},"代表投資案件タイムライン",[54,16627,16628,16648],{},[57,16629,16630],{},[60,16631,16632,16634,16636,16639,16642,16645],{},[63,16633,16200],{},[63,16635,16344],{},[63,16637,16638],{},"投資先",[63,16640,16641],{},"ラウンド",[63,16643,16644],{},"金額（$M）",[63,16646,16647],{},"技術領域",[70,16649,16650,16669,16688,16706,16723,16741,16758,16775,16792,16811,16828,16845,16863,16880,16897,16916],{},[60,16651,16652,16655,16657,16660,16663,16666],{},[75,16653,16654],{},"2018",[75,16656,16412],{},[75,16658,16659],{},"Habana Labs",[75,16661,16662],{},"Series B（参加）",[75,16664,16665],{},"75",[75,16667,16668],{},"AIアクセラレータ",[60,16670,16671,16674,16676,16679,16682,16685],{},[75,16672,16673],{},"2019",[75,16675,16429],{},[75,16677,16678],{},"Fungible",[75,16680,16681],{},"Series C",[75,16683,16684],{},"200",[75,16686,16687],{},"DPU",[60,16689,16690,16693,16695,16698,16700,16703],{},[75,16691,16692],{},"2020",[75,16694,16412],{},[75,16696,16697],{},"Kandou",[75,16699,16681],{},[75,16701,16702],{},"92.3",[75,16704,16705],{},"SerDes/IP",[60,16707,16708,16710,16712,16715,16717,16720],{},[75,16709,16692],{},[75,16711,16412],{},[75,16713,16714],{},"Rigetti",[75,16716,16681],{},[75,16718,16719],{},"79",[75,16721,16722],{},"量子",[60,16724,16725,16728,16730,16733,16735,16738],{},[75,16726,16727],{},"2023",[75,16729,16363],{},[75,16731,16732],{},"InCore Semiconductors",[75,16734,15238],{},[75,16736,16737],{},"3",[75,16739,16740],{},"RISC-V IP",[60,16742,16743,16745,16747,16750,16752,16755],{},[75,16744,16727],{},[75,16746,16363],{},[75,16748,16749],{},"Mindgrove Technologies",[75,16751,15238],{},[75,16753,16754],{},"2.3",[75,16756,16757],{},"SoC",[60,16759,16760,16762,16764,16767,16770,16773],{},[75,16761,16216],{},[75,16763,16363],{},[75,16765,16766],{},"Quantum Circuits",[75,16768,16769],{},"Series B",[75,16771,16772],{},">60",[75,16774,16722],{},[60,16776,16777,16779,16781,16784,16787,16790],{},[75,16778,16216],{},[75,16780,16396],{},[75,16782,16783],{},"Akeana",[75,16785,16786],{},"累計",[75,16788,16789],{},">100",[75,16791,16740],{},[60,16793,16794,16796,16799,16802,16805,16808],{},[75,16795,16216],{},[75,16797,16798],{},"SoftBank Group",[75,16800,16801],{},"Graphcore",[75,16803,16804],{},"M&A",[75,16806,16807],{},"非開示",[75,16809,16810],{},"AIチップ",[60,16812,16813,16815,16817,16820,16822,16825],{},[75,16814,1018],{},[75,16816,16363],{},[75,16818,16819],{},"Ricursive Intelligence",[75,16821,15238],{},[75,16823,16824],{},"35",[75,16826,16827],{},"設計AI/EDA",[60,16829,16830,16832,16834,16837,16839,16842],{},[75,16831,1018],{},[75,16833,16380],{},[75,16835,16836],{},"Unconventional AI",[75,16838,15238],{},[75,16840,16841],{},"475",[75,16843,16844],{},"アナログ新計算",[60,16846,16847,16849,16851,16854,16857,16860],{},[75,16848,1018],{},[75,16850,16412],{},[75,16852,16853],{},"ChipAgents",[75,16855,16856],{},"Series A",[75,16858,16859],{},"21",[75,16861,16862],{},"EDA自動化",[60,16864,16865,16867,16869,16872,16875,16877],{},[75,16866,1018],{},[75,16868,16396],{},[75,16870,16871],{},"Retym",[75,16873,16874],{},"Series D",[75,16876,16665],{},[75,16878,16879],{},"コヒレントDSP",[60,16881,16882,16884,16886,16889,16891,16894],{},[75,16883,1018],{},[75,16885,16798],{},[75,16887,16888],{},"Ampere Computing",[75,16890,16804],{},[75,16892,16893],{},"6,500",[75,16895,16896],{},"Arm CPU",[60,16898,16899,16901,16904,16907,16910,16913],{},[75,16900,16284],{},[75,16902,16903],{},"Sequoia（関連）",[75,16905,16906],{},"Ayar Labs",[75,16908,16909],{},"Series E",[75,16911,16912],{},"500",[75,16914,16915],{},"CPO/光I/O",[60,16917,16918,16920,16922,16925,16927,16929],{},[75,16919,16284],{},[75,16921,16380],{},[75,16923,16924],{},"Nexthop AI",[75,16926,16769],{},[75,16928,16912],{},[75,16930,16931],{},"AIスイッチ",[205,16933],{},[32,16935,16936],{"id":16936},"シナリオ分析と投資家へのインプリケーション",[19,16938,16939],{},"ここからが投資家の出口側の話です。",[122,16941,16942],{"id":16942},"シナリオ別ポジショニング",[54,16944,16945,16963],{},[57,16946,16947],{},[60,16948,16949,16951,16954,16957,16960],{},[63,16950,601],{},[63,16952,16953],{},"確率",[63,16955,16956],{},"AI需要の方向",[63,16958,16959],{},"ボトルネック",[63,16961,16962],{},"推奨アロケーション",[70,16964,16965,16981,16997],{},[60,16966,16967,16970,16972,16975,16978],{},[75,16968,16969],{},"ベース：継続的拡張",[75,16971,624],{},[75,16973,16974],{},"+25%/年",[75,16976,16977],{},"設計・データ移動が交互に律速",[75,16979,16980],{},"EDA/設計AI＋光インターコネクト＝コア配分",[60,16982,16983,16986,16988,16991,16994],{},[75,16984,16985],{},"メイン：加速",[75,16987,646],{},[75,16989,16990],{},"+35%/年",[75,16992,16993],{},"パッケージング・電力が新ボトル",[75,16995,16996],{},"CPO/Chiplet/電力半導体に追加配分",[60,16998,16999,17002,17004,17007,17010],{},[75,17000,17001],{},"テール：減速",[75,17003,665],{},[75,17005,17006],{},"+10%/年",[75,17008,17009],{},"キャパ過剰感→価格圧力",[75,17011,17012],{},"装置・材料サイクル銘柄を削減、IP/EDAに退避",[122,17014,17015],{"id":17015},"日本投資家の現実的アングル",[19,17017,17018,17019,17022,17023,17026],{},"CVCで日本の半導体エコシステムも見てきた経験から言うと、",[23,17020,17021],{},"日本からこの波に乗るとしたら、製造装置・材料・テスト（国内に強いプレイヤー）と、設計IP・EDA自動化（ソフト寄りで参入障壁が低い）の2つの切り口が現実的","です。私が現場で最も強く感じてきたのは、",[23,17024,17025],{},"プロセス・装置データのエコシステムは外から作るのは難しい","ということで、ここは日本勢が固有の優位を持っている領域でもあります。",[205,17028],{},[32,17030,17032],{"id":17031},"zyl0の視点","ZYL0の視点",[19,17034,17035],{},"私の足元のスタンスを一言で言うと、**「半導体投資は『チップを買う』のではなく『AIの制約を買う』」**です。",[19,17037,17038],{},"GPU性能が指数関数的に伸び続けるなら、設計・接続・実装・供給の各レイヤーが次々とボトルネックに変わっていく——これは私が現場で「装置投資が回るほど、歩留まりとプロセスデータが利益を決める」と学んだ感覚と同じ構造です。",[19,17040,17041],{},"次回（連載②）では、各技術領域の具体的な投資ロジックとケーススタディを深掘りします。",[205,17043],{},[19,17045,17046],{},[802,17047,804],{},[205,17049],{},[10,17051,17052,17056,17059,17066,17070,17072,17089,17096,17098,17102,17122,17124,17128,17182,17185,17228,17235,17242,17244,17248,17254,17257,17259,17263,17363,17365,17369,17373,17379,17382,17408,17414,17418,17424,17438,17443,17447,17454,17468,17473,17477,17483,17509,17514,17518,17524,17544,17547,17552,17554,17558,17818,17820,17824,17828,17897,17901,17916,17918,17922,17927,17930,17933,17935],{"lang":809},[14,17053,17055],{"id":17054},"semiconductor-vc-series-converging-on-ai-bottlenecks","Semiconductor VC Series ①: Converging on AI Bottlenecks",[19,17057,17058],{},"I came up through semiconductor process development. Semiconductors aren't an abstract investment category for me — they're an operational reality I've worked with.",[19,17060,17061,17062,17065],{},"When I moved to corporate VC and watched how Tier-1 VCs allocated capital, the pattern that jumped out was that ",[23,17063,17064],{},"they aren't buying chips — they're buying AI's constraints",". This is Part 1 of a three-part series on what that thesis looks like across Sequoia, a16z, Kleiner Perkins, Bessemer, and SoftBank.",[32,17067,17069],{"id":17068},"whats-actually-happening-behind-the-boom","What's Actually Happening Behind the Boom",[19,17071,15411],{},[19,17073,17074,17075,17078,17079,17082,17083,10338,17086,849],{},"In 2025 the global semiconductor market hit an all-time high of ",[23,17076,17077],{},"$791.7B (WSTS)",", with consensus pointing to ",[23,17080,17081],{},"$975B in 2026"," as AI infrastructure pulls the whole stack. SEMI projects 300mm fab equipment spending of ",[23,17084,17085],{},"$133B in 2026",[23,17087,17088],{},"$151B in 2027",[19,17090,17091,17092,17095],{},"Read the surface and it looks like a chip boom. Read the VC books and it's something different: ",[23,17093,17094],{},"the bottlenecks"," — design throughput, data movement, packaging, and new computing paradigms — are where Tier-1 capital is concentrated.",[205,17097],{},[32,17099,17101],{"id":17100},"executive-summary","Executive Summary",[312,17103,17104,17110,17116],{},[315,17105,17106,17109],{},[23,17107,17108],{},"AI infrastructure is the growth driver."," The rise in market size and equipment investment signals sustained capital intensity, not easing supply constraints. For startups, securing production and testing capability is increasingly the key evaluation criterion.",[315,17111,17112,17115],{},[23,17113,17114],{},"VC differentiation is in stage selection and partnership structure."," Mega-seed architectural bets (a16z), design productivity (Sequoia/Bessemer), or M&A-driven vertical integration (SoftBank) — fundamentally different capital strategies.",[315,17117,17118,17121],{},[23,17119,17120],{},"Geography stays U.S.-centric but diversifies."," India (RISC-V/IP), Europe (AI accelerators / photonics), and Israel (AI chips) are visible as supply-chain diversification plays.",[205,17123],{},[32,17125,17127],{"id":17126},"global-semiconductor-market-snapshot","Global Semiconductor Market Snapshot",[54,17129,17130,17146],{},[57,17131,17132],{},[60,17133,17134,17137,17140,17143],{},[63,17135,17136],{},"Year",[63,17138,17139],{},"Market size",[63,17141,17142],{},"YoY",[63,17144,17145],{},"Key drivers",[70,17147,17148,17159,17170],{},[60,17149,17150,17152,17154,17156],{},[75,17151,16216],{},[75,17153,16219],{},[75,17155,250],{},[75,17157,17158],{},"Memory recovery, logic",[60,17160,17161,17163,17165,17167],{},[75,17162,1018],{},[75,17164,16231],{},[75,17166,16234],{},[75,17168,17169],{},"AI data centers, HBM",[60,17171,17172,17175,17177,17179],{},[75,17173,17174],{},"2026 (est.)",[75,17176,16245],{},[75,17178,16248],{},[75,17180,17181],{},"Continued AI infrastructure",[19,17183,17184],{},"300mm equipment spending (SEMI):",[54,17186,17187,17198],{},[57,17188,17189],{},[60,17190,17191,17193,17196],{},[63,17192,17136],{},[63,17194,17195],{},"Equipment spend",[63,17197,5185],{},[70,17199,17200,17210,17219],{},[60,17201,17202,17204,17207],{},[75,17203,1018],{},[75,17205,17206],{},"~$120B",[75,17208,17209],{},"AI ramp",[60,17211,17212,17214,17216],{},[75,17213,16284],{},[75,17215,16287],{},[75,17217,17218],{},"AI demand + supply-chain reshaping",[60,17220,17221,17223,17225],{},[75,17222,16295],{},[75,17224,16298],{},[75,17226,17227],{},"Continued investment cycle",[19,17229,17230,17231,17234],{},"Concentration is in Memory/Logic. Peripheral layers — EDA, test, packaging — benefit indirectly ",[802,17232,17233],{},"as"," bottlenecks rather than as direct growth plays.",[19,17236,17237,17238,17241],{},"From doing DFM rollout on the floor: ",[23,17239,17240],{},"the more capital flows into equipment, the more profit quality is decided by the invisible yield-improvement margin around process data",". That's a universal pattern, and EDA / inspection / DFM-adjacent plays remain investable, in my view, through this cycle.",[205,17243],{},[32,17245,17247],{"id":17246},"the-ai-infrastructure-bottleneck-map","The AI Infrastructure Bottleneck Map",[134,17249,17252],{"className":17250,"code":17251,"language":139},[137],"Growing AI Infrastructure Demand\n        │\n        ▼ Bottlenecks Redefined\n┌───────────────┬───────────────────────────┬────────────────────────┬──────────────────────┐\n│ Design Prod.  │        Data Movement      │       Packaging        │    New Computing     │\n│ EDA/Verif.    │        SerDes/Optics      │       CPO/Chiplet      │    Analog/Quantum    │\n└───────┬───────┴────────────┬──────────────┴─────────────┬──────────┴───────────┬──────────┘\n       ↓                     ↓                            ↓                     ↓\n Seed→A: high           Series B→E                       Late                  Long\n software ratio  Customer/Manufacturing Connection  capital intensive     tech/standard risk\n",[141,17253,17251],{"__ignoreMap":143},[19,17255,17256],{},"Tier-1 VCs spread risk across these four constraint layers and capture upside where bottlenecks repeatedly resurface.",[205,17258],{},[32,17260,17262],{"id":17261},"vc-strategy-comparison","VC Strategy Comparison",[54,17264,17265,17283],{},[57,17266,17267],{},[60,17268,17269,17271,17274,17277,17280],{},[63,17270,16344],{},[63,17272,17273],{},"Focus areas",[63,17275,17276],{},"Stage preference",[63,17278,17279],{},"Geography",[63,17281,17282],{},"Partnership / LP structure",[70,17284,17285,17301,17316,17331,17347],{},[60,17286,17287,17289,17292,17295,17298],{},[75,17288,16363],{},[75,17290,17291],{},"Design AI (EDA/verification), RISC-V/IP, CPO",[75,17293,17294],{},"Seed → Growth",[75,17296,17297],{},"US, India",[75,17299,17300],{},"Public commitment to design AI; related vehicle in CPO",[60,17302,17303,17305,17308,17311,17313],{},[75,17304,16380],{},[75,17306,17307],{},"New compute (analog/mixed-signal), AI networking",[75,17309,17310],{},"Mega-seed + large Series B",[75,17312,14368],{},[75,17314,17315],{},"Strategic Partnerships program for institutional GTM",[60,17317,17318,17320,17323,17326,17328],{},[75,17319,16396],{},[75,17321,17322],{},"RISC-V/IP, coherent DSP",[75,17324,17325],{},"Early → Series D",[75,17327,14368],{},[75,17329,17330],{},"TSMC-backed fund reports → tightening supply-side ties",[60,17332,17333,17335,17338,17341,17344],{},[75,17334,16412],{},[75,17336,17337],{},"SerDes/IP, quantum, EDA automation",[75,17339,17340],{},"Series C core + early",[75,17342,17343],{},"US, Europe, Israel",[75,17345,17346],{},"Deep-tech commercialization across AI chips/quantum/EDA",[60,17348,17349,17351,17354,17357,17360],{},[75,17350,16429],{},[75,17352,17353],{},"AI chips/CPU/DPU, vertical integration",[75,17355,17356],{},"Growth → M&A",[75,17358,17359],{},"US, UK",[75,17361,17362],{},"Strategic LPs (Apple, etc.); Arm-centered AI Computing Segment",[205,17364],{},[32,17366,17368],{"id":17367},"individual-vc-profiles","Individual VC Profiles",[122,17370,17372],{"id":17371},"sequoia-linking-design-to-operations","Sequoia — Linking Design to Operations",[19,17374,17375,17376,849],{},"Sequoia's thesis: ",[23,17377,17378],{},"as AI scales, chip design throughput becomes the bottleneck",[19,17380,17381],{},"Selected deals:",[312,17383,17384,17390,17396,17402],{},[315,17385,17386,17389],{},[23,17387,17388],{},"Ricursive Intelligence ($35M, 2025 Seed, lead)"," — AI-accelerated chip design.",[315,17391,17392,17395],{},[23,17393,17394],{},"InCore Semiconductors ($3M, 2023 Seed)"," — India RISC-V IP.",[315,17397,17398,17401],{},[23,17399,17400],{},"Mindgrove Technologies (~$2.3M, 2023 Seed)"," — India SoC.",[315,17403,17404,17407],{},[23,17405,17406],{},"Ayar Labs ($500M, 2026 Series E, related vehicle)"," — Co-Packaged Optics. Co-investors include NVIDIA and AMD; valuation $3.75B.",[19,17409,17410,17413],{},[23,17411,17412],{},"Read",": seed-stage focus on design productivity, late-stage extension into data movement (optical I/O) — wiring the design-to-operations bottleneck chain together.",[122,17415,17417],{"id":17416},"a16z-mega-seeds-on-compute-paradigms","a16z — Mega-Seeds on Compute Paradigms",[19,17419,17420,17421,849],{},"a16z's signature move is ",[23,17422,17423],{},"massive seed checks on companies that could redefine entire compute categories",[312,17425,17426,17432],{},[315,17427,17428,17431],{},[23,17429,17430],{},"Unconventional AI ($475M, 2025 Seed, co-lead)"," — analog/mixed-signal optimized for probabilistic compute.",[315,17433,17434,17437],{},[23,17435,17436],{},"Nexthop AI ($500M, 2026 Series B)"," — AI-cluster Ethernet switches; explicit thesis that the network is the new cluster bottleneck.",[19,17439,17440,17442],{},[23,17441,17412],{},": where others spread early bets, a16z concentrates capital into paradigm-level bets at seed. The institutionalized Strategic Partnerships program (engineered GTM into government and enterprise) is a structural advantage that's hard to replicate.",[122,17444,17446],{"id":17445},"kleiner-perkins-horizontal-ip-datacenter-optics","Kleiner Perkins — Horizontal IP × Datacenter Optics",[19,17448,17449,17450,17453],{},"Kleiner gravitates to ",[23,17451,17452],{},"horizontal infrastructure",": IP (RISC-V) and datacenter optics/interconnect.",[312,17455,17456,17462],{},[315,17457,17458,17461],{},[23,17459,17460],{},"Akeana (>$100M cumulative, 2024 stealth launch)"," — RISC-V CPU/IP portfolio.",[315,17463,17464,17467],{},[23,17465,17466],{},"Retym ($75M, 2025 Series D)"," — cloud/AI-optimized coherent DSP.",[19,17469,17470,17472],{},[23,17471,17412],{},": target companies that win through design differentiation (IP/DSP) where the customer base is hyperscalers. The reported TSMC-backed fund signals deepening supply-side integration.",[122,17474,17476],{"id":17475},"bessemer-commercializing-deep-tech","Bessemer — Commercializing Deep Tech",[19,17478,17479,17480,849],{},"Bessemer treats SerDes/IP, quantum, and EDA automation as a single portfolio unified by ",[23,17481,17482],{},"deep technical moats with credible commercialization paths",[312,17484,17485,17491,17497,17503],{},[315,17486,17487,17490],{},[23,17488,17489],{},"ChipAgents ($21M, 2025 Series A, lead)"," — agentic AI for chip design and verification.",[315,17492,17493,17496],{},[23,17494,17495],{},"Kandou ($92.3M, 2020 Series C, lead)"," — SerDes/IP, retimers.",[315,17498,17499,17502],{},[23,17500,17501],{},"Rigetti ($79M, 2020 Series C, lead)"," — quantum hardware + cloud.",[315,17504,17505,17508],{},[23,17506,17507],{},"Habana Labs ($75M Series B participant, 2018; Intel ~$2B acquisition, 2019)"," — AI chips.",[19,17510,17511,17513],{},[23,17512,17412],{},": deep European/Israeli networks let Bessemer source technically differentiated companies early. Pattern is consistent — deep tech, plausible deployment path.",[122,17515,17517],{"id":17516},"softbank-vertical-integration-as-strategy","SoftBank — Vertical Integration as Strategy",[19,17519,17520,17521,849],{},"Categorically different from the others. Not just investing in chip startups but ",[23,17522,17523],{},"assembling an integrated AI computing segment via M&A",[312,17525,17526,17532,17538],{},[315,17527,17528,17531],{},[23,17529,17530],{},"Fungible ($200M, 2019 Series C, SVF lead)"," — DPU; later acquired by Microsoft.",[315,17533,17534,17537],{},[23,17535,17536],{},"Graphcore (M&A, 2024, undisclosed)"," — UK AI chip.",[315,17539,17540,17543],{},[23,17541,17542],{},"Ampere Computing ($6.5B announced 2025)"," — Arm-based cloud server CPU.",[19,17545,17546],{},"Arm + Ampere + Graphcore now sit inside one publicly disclosed AI Computing Segment. Strategic LPs in Fund I (PIF, Mubadala, Apple, Foxconn, Qualcomm, Sharp) embed semiconductor strategy at the fund-capital level.",[19,17548,17549,17551],{},[23,17550,17412],{},": this is closer to strategic finance + operating company than traditional VC. Risk and time horizon are categorically different.",[205,17553],{},[32,17555,17557],{"id":17556},"investment-timeline","Investment Timeline",[54,17559,17560,17579],{},[57,17561,17562],{},[60,17563,17564,17566,17568,17571,17574,17577],{},[63,17565,17136],{},[63,17567,16344],{},[63,17569,17570],{},"Portfolio",[63,17572,17573],{},"Round",[63,17575,17576],{},"Amount ($M)",[63,17578,948],{},[70,17580,17581,17597,17611,17625,17640,17654,17669,17683,17698,17713,17728,17743,17758,17773,17787,17803],{},[60,17582,17583,17585,17587,17589,17592,17594],{},[75,17584,16654],{},[75,17586,16412],{},[75,17588,16659],{},[75,17590,17591],{},"Series B (participant)",[75,17593,16665],{},[75,17595,17596],{},"AI accelerator",[60,17598,17599,17601,17603,17605,17607,17609],{},[75,17600,16673],{},[75,17602,16429],{},[75,17604,16678],{},[75,17606,16681],{},[75,17608,16684],{},[75,17610,16687],{},[60,17612,17613,17615,17617,17619,17621,17623],{},[75,17614,16692],{},[75,17616,16412],{},[75,17618,16697],{},[75,17620,16681],{},[75,17622,16702],{},[75,17624,16705],{},[60,17626,17627,17629,17631,17633,17635,17637],{},[75,17628,16692],{},[75,17630,16412],{},[75,17632,16714],{},[75,17634,16681],{},[75,17636,16719],{},[75,17638,17639],{},"Quantum",[60,17641,17642,17644,17646,17648,17650,17652],{},[75,17643,16727],{},[75,17645,16363],{},[75,17647,16732],{},[75,17649,15238],{},[75,17651,16737],{},[75,17653,16740],{},[60,17655,17656,17658,17660,17663,17665,17667],{},[75,17657,16727],{},[75,17659,16363],{},[75,17661,17662],{},"Mindgrove",[75,17664,15238],{},[75,17666,16754],{},[75,17668,16757],{},[60,17670,17671,17673,17675,17677,17679,17681],{},[75,17672,16216],{},[75,17674,16363],{},[75,17676,16766],{},[75,17678,16769],{},[75,17680,16772],{},[75,17682,17639],{},[60,17684,17685,17687,17689,17691,17694,17696],{},[75,17686,16216],{},[75,17688,16396],{},[75,17690,16783],{},[75,17692,17693],{},"Cumulative",[75,17695,16789],{},[75,17697,16740],{},[60,17699,17700,17702,17704,17706,17708,17710],{},[75,17701,16216],{},[75,17703,16798],{},[75,17705,16801],{},[75,17707,16804],{},[75,17709,1281],{},[75,17711,17712],{},"AI chip",[60,17714,17715,17717,17719,17721,17723,17725],{},[75,17716,1018],{},[75,17718,16363],{},[75,17720,16819],{},[75,17722,15238],{},[75,17724,16824],{},[75,17726,17727],{},"Design AI/EDA",[60,17729,17730,17732,17734,17736,17738,17740],{},[75,17731,1018],{},[75,17733,16380],{},[75,17735,16836],{},[75,17737,15238],{},[75,17739,16841],{},[75,17741,17742],{},"Analog new compute",[60,17744,17745,17747,17749,17751,17753,17755],{},[75,17746,1018],{},[75,17748,16412],{},[75,17750,16853],{},[75,17752,16856],{},[75,17754,16859],{},[75,17756,17757],{},"EDA automation",[60,17759,17760,17762,17764,17766,17768,17770],{},[75,17761,1018],{},[75,17763,16396],{},[75,17765,16871],{},[75,17767,16874],{},[75,17769,16665],{},[75,17771,17772],{},"Coherent DSP",[60,17774,17775,17777,17779,17781,17783,17785],{},[75,17776,1018],{},[75,17778,16798],{},[75,17780,16888],{},[75,17782,16804],{},[75,17784,16893],{},[75,17786,16896],{},[60,17788,17789,17791,17794,17796,17798,17800],{},[75,17790,16284],{},[75,17792,17793],{},"Sequoia (related)",[75,17795,16906],{},[75,17797,16909],{},[75,17799,16912],{},[75,17801,17802],{},"CPO/Optical I/O",[60,17804,17805,17807,17809,17811,17813,17815],{},[75,17806,16284],{},[75,17808,16380],{},[75,17810,16924],{},[75,17812,16769],{},[75,17814,16912],{},[75,17816,17817],{},"AI switch",[205,17819],{},[32,17821,17823],{"id":17822},"scenario-analysis-investor-implications","Scenario Analysis & Investor Implications",[122,17825,17827],{"id":17826},"allocation-by-scenario","Allocation by scenario",[54,17829,17830,17847],{},[57,17831,17832],{},[60,17833,17834,17836,17838,17841,17844],{},[63,17835,1389],{},[63,17837,1392],{},[63,17839,17840],{},"AI demand path",[63,17842,17843],{},"Active bottleneck",[63,17845,17846],{},"Suggested allocation",[70,17848,17849,17865,17881],{},[60,17850,17851,17854,17856,17859,17862],{},[75,17852,17853],{},"Base — continued expansion",[75,17855,624],{},[75,17857,17858],{},"+25%/yr",[75,17860,17861],{},"Design + data movement alternate",[75,17863,17864],{},"EDA/design AI + optical interconnect = core",[60,17866,17867,17870,17872,17875,17878],{},[75,17868,17869],{},"Main — acceleration",[75,17871,646],{},[75,17873,17874],{},"+35%/yr",[75,17876,17877],{},"Packaging + power emerge as new limits",[75,17879,17880],{},"Add CPO/Chiplet/power-semis",[60,17882,17883,17886,17888,17891,17894],{},[75,17884,17885],{},"Tail — slowdown",[75,17887,665],{},[75,17889,17890],{},"+10%/yr",[75,17892,17893],{},"Capacity overhang, pricing pressure",[75,17895,17896],{},"Trim equipment/material cyclicals; rotate to IP/EDA",[122,17898,17900],{"id":17899},"practical-angles-for-japan-based-investors","Practical angles for Japan-based investors",[19,17902,17903,17904,17907,17908,17911,17912,17915],{},"From watching the Japanese semiconductor ecosystem from inside CVC, the realistic angles are: ",[23,17905,17906],{},"(1) manufacturing equipment, materials, and test"," — incumbents with global moats; ",[23,17909,17910],{},"(2) design IP and EDA automation"," — software-leaning, lower entry barriers. The thing I felt most strongly from the floor is that ",[23,17913,17914],{},"process/equipment data ecosystems are very hard to build from outside",", and this is where Japan retains structural advantage.",[205,17917],{},[32,17919,17921],{"id":17920},"zyl0s-take","ZYL0's Take",[19,17923,1558,17924,849],{},[23,17925,17926],{},"semiconductor investing means buying AI's constraints, not its compute",[19,17928,17929],{},"If GPU performance keeps scaling, design, interconnect, packaging, and supply chain rotate through the role of \"current bottleneck.\" That mirrors a pattern I learned from process work: the more capital pours into equipment, the more profit quality gets decided by yield and process data downstream. The VC bets reflect that.",[19,17931,17932],{},"Part 2 goes deeper on each technology domain — investment logic, case studies, and what separates winners from losers per vertical.",[205,17934],{},[19,17936,17937],{},[802,17938,1603],{},{"title":143,"searchDepth":1605,"depth":1605,"links":17940},[17941,17942,17943,17944,17945,17946,17953,17954,17958,17959,17960,17961,17962,17963,17964,17971,17972,17976],{"id":16139,"depth":1605,"text":16140},{"id":16166,"depth":1605,"text":16166},{"id":16191,"depth":1605,"text":16191},{"id":16316,"depth":1605,"text":16317},{"id":16334,"depth":1605,"text":16335},{"id":16446,"depth":1605,"text":16447,"children":17947},[17948,17949,17950,17951,17952],{"id":16450,"depth":1610,"text":16451},{"id":16492,"depth":1610,"text":16493},{"id":16520,"depth":1610,"text":16521},{"id":16548,"depth":1610,"text":16549},{"id":16588,"depth":1610,"text":16589},{"id":16625,"depth":1605,"text":16625},{"id":16936,"depth":1605,"text":16936,"children":17955},[17956,17957],{"id":16942,"depth":1610,"text":16942},{"id":17015,"depth":1610,"text":17015},{"id":17031,"depth":1605,"text":17032},{"id":17068,"depth":1605,"text":17069},{"id":17100,"depth":1605,"text":17101},{"id":17126,"depth":1605,"text":17127},{"id":17246,"depth":1605,"text":17247},{"id":17261,"depth":1605,"text":17262},{"id":17367,"depth":1605,"text":17368,"children":17965},[17966,17967,17968,17969,17970],{"id":17371,"depth":1610,"text":17372},{"id":17416,"depth":1610,"text":17417},{"id":17445,"depth":1610,"text":17446},{"id":17475,"depth":1610,"text":17476},{"id":17516,"depth":1610,"text":17517},{"id":17556,"depth":1605,"text":17557},{"id":17822,"depth":1605,"text":17823,"children":17973},[17974,17975],{"id":17826,"depth":1610,"text":17827},{"id":17899,"depth":1610,"text":17900},{"id":17920,"depth":1605,"text":17921},"市場規模$975Bへ拡大する半導体業界でTier1 VCはどこに賭けているのか。Sequoia、a16z、Kleiner Perkins、Bessemer、SoftBankの戦略を、半導体プロセス開発出身のCVCの視点で比較分析する。連載全3回の初回。",{"date":17979,"image":17980,"alt":17981,"tags":17982,"tagsEn":17985,"published":1666},"11th Apr 2026","/blogs-img/blog-semiconductor-vc-1.png","Tier1 VCと半導体投資の全体像",[17983,17984,1657,1660,4240],"半導体","VC投資",[17986,17987,1662,1665,4245],"Semiconductor","VC Investing","/blogs/6-semiconductor-vc-series-1",{"title":16119,"description":17977},"blogs/6-semiconductor-vc-series-1","5Y1AA_lyJEyFvT7WuVFacvXjlFJsxkHuCI1MM8WD7Q8",{"id":17993,"title":17994,"body":17995,"description":20271,"extension":1651,"meta":20272,"navigation":1666,"ogImage":20273,"path":20277,"seo":20278,"stem":20279,"__hash__":20280},"content/blogs/7-semiconductor-vc-series-2.md","「設計」「データ移動」「パッケージング」に資金が集まる理由｜連載②技術別深掘り | Why Design, Data Movement & Packaging Attract VC Capital — Series②",{"type":7,"value":17996,"toc":20213},[17997,19124],[10,17998,17999,18003,18006,18010,18013,18016,18019,18021,18023,18043,18045,18049,18251,18253,18257,18261,18264,18268,18271,18278,18282,18285,18288,18293,18346,18348,18352,18355,18358,18364,18368,18371,18374,18380,18384,18387,18390,18394,18397,18400,18404,18407,18410,18415,18466,18468,18472,18476,18479,18493,18500,18504,18507,18510,18530,18533,18539,18543,18546,18593,18596,18598,18602,18606,18609,18616,18620,18623,18626,18630,18633,18636,18641,18685,18687,18691,19015,19017,19020,19023,19091,19098,19100,19102,19105,19108,19115,19118,19120],{"lang":12},[14,18000,18002],{"id":18001},"設計データ移動パッケージングに資金が集まる理由","「設計」「データ移動」「パッケージング」に資金が集まる理由",[19,18004,18005],{},"私はEEPROMの車載プロセス開発でDFM（Design For Manufacturing）を社内導入し、収率10%改善を主導した経験があります。設計・プロセス・パッケージングが同時に動く現場の感覚から言うと、AIの計算ボトルネックは**「単一レイヤーの問題ではなく、3つのレイヤーが互いに律速し合う構造問題」**です。だからこそTier1 VCがこの3領域に同時に資金を投下しているのは、私には理にかなって見えます。連載第2回は、その技術的な必然性を、代表案件を通じて解いていきます。",[32,18007,18009],{"id":18008},"なぜ周辺レイヤーに大型資金が流れるのか","なぜ「周辺レイヤー」に大型資金が流れるのか",[19,18011,18012],{},"最初に、論点をフラットに置きます。",[19,18014,18015],{},"前回（連載①）で確認した通り、Tier1 VCが半導体に賭けるのは「計算そのもの」ではなく「AIの制約領域」です。GPUの演算能力が指数関数的に伸びれば、必ずその「周辺」がボトルネックに移ります。1万枚のGPUを繋ぐにはスイッチとインターコネクトが要る。チップ単体の性能限界を超えるにはパッケージング（CPO/Chiplet）が要る。設計の複雑性が増せばEDAと検証の自動化が要る。",[19,18017,18018],{},"それぞれの領域に、なぜ今、大型の資金が流れているのか。代表案件をケーススタディとして読み解いていきます。",[205,18020],{},[32,18022,16166],{"id":16166},[312,18024,18025,18031,18037],{},[315,18026,18027,18030],{},[23,18028,18029],{},"投資のROI設計は「単一チップ勝負」から「スタック勝負（設計→実装→運用）」へ。"," CPOやコヒレントDSP、スイッチングなど、運用上の制約を解く領域で$500M超の大型ラウンドが立ち上がっています。",[315,18032,18033,18036],{},[23,18034,18035],{},"生成AIの普及で設計の複雑性が増し、EDA自動化（Agentic AI）が新たな投資テーマに。"," 設計・検証のリードタイム短縮は、製造リスクが相対的に低いソフト寄りの案件として投資回転が早い。",[315,18038,18039,18042],{},[23,18040,18041],{},"「新計算（アナログ/量子）」は長期テーマとして存続するが、資金は「研究→プロダクト→市場」への接続ができる会社に集まる。"," 技術の実現可能性だけでなく、顧客・クラウド・政府との連携が評価基準になっています。",[205,18044],{},[32,18046,18048],{"id":18047},"技術スタック-投資マッピング","技術スタック × 投資マッピング",[54,18050,18051,18069],{},[57,18052,18053],{},[60,18054,18055,18057,18060,18063,18066],{},[63,18056,16647],{},[63,18058,18059],{},"代表企業（VC）",[63,18061,18062],{},"主要ラウンド",[63,18064,18065],{},"投資額",[63,18067,18068],{},"技術の役割",[70,18070,18071,18086,18101,18116,18131,18146,18161,18175,18190,18205,18221,18236],{},[60,18072,18073,18075,18078,18080,18083],{},[75,18074,16827],{},[75,18076,18077],{},"Ricursive（Sequoia）",[75,18079,15238],{},[75,18081,18082],{},"$35M",[75,18084,18085],{},"チップ設計サイクルの短縮",[60,18087,18088,18090,18093,18095,18098],{},[75,18089,16827],{},[75,18091,18092],{},"ChipAgents（Bessemer）",[75,18094,16856],{},[75,18096,18097],{},"$21M",[75,18099,18100],{},"設計・検証のエージェント化",[60,18102,18103,18105,18108,18110,18113],{},[75,18104,16740],{},[75,18106,18107],{},"Akeana（Kleiner Perkins）",[75,18109,16786],{},[75,18111,18112],{},">$100M",[75,18114,18115],{},"自由に使えるCPU命令セット",[60,18117,18118,18120,18123,18125,18128],{},[75,18119,16705],{},[75,18121,18122],{},"Kandou（Bessemer）",[75,18124,16681],{},[75,18126,18127],{},"$92.3M",[75,18129,18130],{},"高速チップ間通信の物理層",[60,18132,18133,18135,18138,18140,18143],{},[75,18134,16879],{},[75,18136,18137],{},"Retym（Kleiner Perkins）",[75,18139,16874],{},[75,18141,18142],{},"$75M",[75,18144,18145],{},"長距離光通信の信号処理",[60,18147,18148,18150,18153,18155,18158],{},[75,18149,16915],{},[75,18151,18152],{},"Ayar Labs（Sequoia系）",[75,18154,16909],{},[75,18156,18157],{},"$500M",[75,18159,18160],{},"銅配線の限界をシリコンフォトニクスで超える",[60,18162,18163,18165,18168,18170,18172],{},[75,18164,16931],{},[75,18166,18167],{},"Nexthop AI（a16z）",[75,18169,16769],{},[75,18171,18157],{},[75,18173,18174],{},"AI大規模クラスタのネットワーク",[60,18176,18177,18179,18182,18184,18187],{},[75,18178,16844],{},[75,18180,18181],{},"Unconventional AI（a16z）",[75,18183,15238],{},[75,18185,18186],{},"$475M",[75,18188,18189],{},"電力効率の桁違い改善を狙う確率的計算",[60,18191,18192,18195,18198,18200,18202],{},[75,18193,18194],{},"AIチップ（買収）",[75,18196,18197],{},"Graphcore（SoftBank）",[75,18199,16804],{},[75,18201,16807],{},[75,18203,18204],{},"英国発グラフプロセッサ",[60,18206,18207,18210,18213,18215,18218],{},[75,18208,18209],{},"CPU/Arm（買収）",[75,18211,18212],{},"Ampere（SoftBank）",[75,18214,16804],{},[75,18216,18217],{},"$6.5B",[75,18219,18220],{},"クラウドサーバー向けArmベースCPU",[60,18222,18223,18225,18228,18230,18233],{},[75,18224,16722],{},[75,18226,18227],{},"Rigetti（Bessemer）",[75,18229,16681],{},[75,18231,18232],{},"$79M",[75,18234,18235],{},"量子クラウド提供+ハードウェア",[60,18237,18238,18240,18243,18245,18248],{},[75,18239,16722],{},[75,18241,18242],{},"Quantum Circuits（Sequoia）",[75,18244,16769],{},[75,18246,18247],{},">$60M",[75,18249,18250],{},"フォールトトレラント量子コンピュータ",[205,18252],{},[32,18254,18256],{"id":18255},"領域設計ipeda-開発リードタイム短縮への賭け","領域①：設計・IP/EDA — 「開発リードタイム短縮」への賭け",[122,18258,18260],{"id":18259},"なぜ今設計自動化にvcが入るのか","なぜ今、設計自動化にVCが入るのか",[19,18262,18263],{},"先端チップの設計工数は年々増大していますよね。7nm→3nm→2nmと微細化するほど、設計規則の複雑性は指数関数的に増え、テープアウト1回のコストも億円単位になります。検証・シミュレーション・デバッグが開発期間の大半を占める現状に対し、AIエージェントを当てる仮説が立ち上がっています。",[122,18265,18267],{"id":18266},"ケーススタディricursive-intelligence35m2025年seed","ケーススタディ：Ricursive Intelligence（$35M、2025年Seed）",[19,18269,18270],{},"Sequoiaがリードした$35Mシードは、「AIがチップ設計そのものを加速する」テーマをTier1が最上流から取りにいった案件なんですよね。",[19,18272,18273,18274,18277],{},"注目点は",[23,18275,18276],{},"ソフトウェア比率が高い","ことなんですよね。製造工程（ファブ）を持たず、設計フローにAIを組み込むアプローチは、相対的に製造リスクが低い。一方で、Synopsys・Cadenceというグローバルなツールチェーン巨人のプロダクト戦略・M&A戦略に強く影響される。勝ち筋は「差別化されたワークフロー統合」か「特定の設計フロー（例：検証/STA）での圧倒的優位」になりやすい。",[122,18279,18281],{"id":18280},"ケーススタディchipagents21m2025年series-a","ケーススタディ：ChipAgents（$21M、2025年Series A）",[19,18283,18284],{},"Bessemerがリードしたこの案件は、既存EDAチェーンの「検証/デバッグ」をAIエージェントで置き換える仮説なんですよね。",[19,18286,18287],{},"「設計工数の相当部分は検証とバグ修正に費やされる」という現場の現実を起点に、Agentic AIのループ（タスク定義→実行→検証→修正）を設計フローに組み込む。ChipAgentsが述べる通り、目標は「エンジニアが書く必要があるコードを減らす」ことではなく「検証→修正のサイクルを自動化する」ことなんですよね。",[19,18289,18290],{},[23,18291,18292],{},"EDA自動化の投資ロジックまとめ：",[54,18294,18295,18304],{},[57,18296,18297],{},[60,18298,18299,18302],{},[63,18300,18301],{},"観点",[63,18303,68],{},[70,18305,18306,18314,18322,18330,18338],{},[60,18307,18308,18311],{},[75,18309,18310],{},"市場の大きさ",[75,18312,18313],{},"全世界のチップ設計者数×設計サイクル短縮価値",[60,18315,18316,18319],{},[75,18317,18318],{},"製造リスク",[75,18320,18321],{},"低（ソフトウェア主体）",[60,18323,18324,18327],{},[75,18325,18326],{},"競合リスク",[75,18328,18329],{},"EDA大手（Synopsys/Cadence）のM&A対象になりうる",[60,18331,18332,18335],{},[75,18333,18334],{},"出口",[75,18336,18337],{},"M&A（EDA大手/チップ設計会社）またはIPO",[60,18339,18340,18343],{},[75,18341,18342],{},"タイムライン",[75,18344,18345],{},"比較的短い（2〜5年）",[205,18347],{},[32,18349,18351],{"id":18350},"領域インターコネクトデータ移動-gpuを繋ぐが主戦場","領域②：インターコネクト/データ移動 — 「GPUを繋ぐ」が主戦場",[122,18353,18354],{"id":18354},"銅配線の限界とシリコンフォトニクスの台頭",[19,18356,18357],{},"AI大規模クラスタ（数千〜数万GPU）では、チップ間・ラック間・データセンター間のデータ移動が帯域・遅延・電力の全てでボトルネックになっています。",[134,18359,18362],{"className":18360,"code":18361,"language":139},[137],"従来の銅配線：帯域 ↑ → 電力消費 ↑↑↑、距離 → 減衰\nシリコンフォトニクス：光で通信 → 高帯域・低電力・長距離\nCPO（Co-Packaged Optics）：光I/OをチップパッケージにCo-Packageng\n        → 銅配線の物理限界を迂回する\n",[141,18363,18361],{"__ignoreMap":143},[122,18365,18367],{"id":18366},"ケーススタディayar-labs500m2026年series-e","ケーススタディ：Ayar Labs（$500M、2026年Series E）",[19,18369,18370],{},"評価額$3.75Bに達したこのラウンドは、CPO（Co-Packaged Optics）分野で最大規模の資金調達の一つなんですよね。",[19,18372,18373],{},"戦略投資家にNVIDIA・AMDが名を連ねる構造は重要なんですよね。彼らがAyar Labsに投資するのは「財務的リターン」だけでなく「次世代パッケージング技術を自分たちの製品ロードマップに組み込む」ためなんですよね。Sequoia系ビークルが参加していることは、ピュアファイナンシャルのリターンも十分あるという判断を示す。",[19,18375,18376,18379],{},[23,18377,18378],{},"投資ロジック："," 銅配線の物理限界（帯域密度・電力）は設計上の解決が難しい。光I/OをCo-Packageするというアーキテクチャ転換が起きれば、その製造技術・IP・テスト能力を持つ会社の価値は急増します。",[122,18381,18383],{"id":18382},"ケーススタディretym75m2025年series-d","ケーススタディ：Retym（$75M、2025年Series D）",[19,18385,18386],{},"クラウド/AI向けコヒレントDSPを専門とするRetymのSeries Dには、Kleiner Perkinsが参加していますよね。",[19,18388,18389],{},"特徴は「距離×帯域×消費電力」の現実制約に設計思想を合わせた点なんですよね。データセンター内（数十〜数百m）、メトロ（数十km）、長距離（数百km〜）でそれぞれ最適なDSPアーキテクチャが異なります。汎用ではなく特定の「距離レンジ」に最適化したシリコンを作ることで、差別化を図る。",[122,18391,18393],{"id":18392},"ケーススタディkandou923m2020年series-c","ケーススタディ：Kandou（$92.3M、2020年Series C）",[19,18395,18396],{},"Bessemerがリードしたこの案件は、SerDes（シリアライザ/デシリアライザ）とリタイマーの商用化を前提にした投資なんですよね。",[19,18398,18399],{},"KandouのMatterhorn IPはUSB4やThunderbolt実装で実績を持つ。サーバー・デバイスの世代交代（USB4 Gen3など）に合わせて実装が変わる局面を「実装の勝ち筋」として取りにいくスタイルなんですよね。スタンダードの世代交代とともに需要が立ち上がる、周期性のあるビジネスモデルと言えると思います。",[122,18401,18403],{"id":18402},"ケーススタディnexthop-ai500m2026年series-b","ケーススタディ：Nexthop AI（$500M、2026年Series B）",[19,18405,18406],{},"a16zが主要投資家として$500Mを投じたNexthop AIは、「AI時代のイーサネットスイッチ」を標榜します。",[19,18408,18409],{},"投資ロジックは明快なんですよね。「AIクラスタが大規模化するほど、スイッチのボトルネックが顕在化します。既存のスイッチはAIの通信パターン（全-全接続的な集合通信）に最適化されていない」という認識から出発し、AI専用スイッチングのアーキテクチャを取りにいく。",[19,18411,18412],{},[23,18413,18414],{},"インターコネクト投資の共通ロジック：",[54,18416,18417,18425],{},[57,18418,18419],{},[60,18420,18421,18423],{},[63,18422,18301],{},[63,18424,68],{},[70,18426,18427,18435,18443,18451,18459],{},[60,18428,18429,18432],{},[75,18430,18431],{},"技術ドライバー",[75,18433,18434],{},"GPU/アクセラレータの性能スケールがI/Oを際立たせる",[60,18436,18437,18440],{},[75,18438,18439],{},"顧客",[75,18441,18442],{},"ハイパースケーラー（Google/AWS/Microsoft/Meta等）",[60,18444,18445,18448],{},[75,18446,18447],{},"資本集約度",[75,18449,18450],{},"中〜高（量産・認定が必要）",[60,18452,18453,18456],{},[75,18454,18455],{},"リスク",[75,18457,18458],{},"量産歩留まり、サプライヤー認定、地政学",[60,18460,18461,18463],{},[75,18462,18334],{},[75,18464,18465],{},"M&A（チップ大手）またはIPO",[205,18467],{},[32,18469,18471],{"id":18470},"領域aiアクセラレータ新計算-計算原理への賭け","領域③：AIアクセラレータ/新計算 — 計算原理への賭け",[122,18473,18475],{"id":18474},"パラダイム賭け-vs-垂直統合","パラダイム賭け vs. 垂直統合",[19,18477,18478],{},"「計算原理を変える」というテーマには2つのアプローチが存在します。",[552,18480,18481,18487],{},[315,18482,18483,18486],{},[23,18484,18485],{},"スタートアップへの巨大シード","（a16z × Unconventional AI）：リスクが高いが、実現すれば市場を再定義できます。",[315,18488,18489,18492],{},[23,18490,18491],{},"買収による垂直統合","（SoftBank × Graphcore/Ampere）：技術の実現可能性リスクは低いが、統合コストと市場タイミングリスクがあります。",[19,18494,18495,18496,18499],{},"どちらも「単なるチップ投資」ではなく、",[23,18497,18498],{},"エコシステムとしてのポジション取り","だという点で共通していますよね。",[122,18501,18503],{"id":18502},"ケーススタディunconventional-ai475m2025年seed","ケーススタディ：Unconventional AI（$475M、2025年Seed）",[19,18505,18506],{},"a16zが共同リードした$475Mシードは、半導体投資史上最大規模のシードラウンドの一つなんですよね。",[19,18508,18509],{},"アナログ/ミックスドシグナルで確率的計算を扱うというアプローチは、デジタルCMOSによる現行アーキテクチャとは根本的に異なります。主な課題は：",[312,18511,18512,18518,18524],{},[315,18513,18514,18517],{},[23,18515,18516],{},"量産・再現性","：アナログ回路は製造ばらつきの影響を受けやすい",[315,18519,18520,18523],{},[23,18521,18522],{},"ツールチェーン","：設計・検証ツールが整っていない",[315,18525,18526,18529],{},[23,18527,18528],{},"エコシステム","：ソフトウェア/フレームワーク側の対応が必要",[19,18531,18532],{},"一方、実現すれば「電力効率の桁違い改善」というインパクトは現行GPUに対して圧倒的な競争優位になりうる。",[19,18534,18535,18538],{},[23,18536,18537],{},"a16zがこの賭けをする理由","：Strategic Partnershipsで企業・政府導入の経路を制度化しているa16zにとって、「カテゴリを再定義するアーキテクチャ」を囲い込む初期コストは、後で払う「選択肢なしコスト」より安い。",[122,18540,18542],{"id":18541},"softbankの垂直統合armgraphcoreampere","SoftBankの垂直統合：Arm×Graphcore×Ampere",[19,18544,18545],{},"SoftBankのアプローチは、個別案件の勝負ではなくセグメント全体の設計なんですよね。",[54,18547,18548,18560],{},[57,18549,18550],{},[60,18551,18552,18554,18557],{},[63,18553,5925],{},[63,18555,18556],{},"役割",[63,18558,18559],{},"SoftBankの取得形態",[70,18561,18562,18573,18583],{},[60,18563,18564,18567,18570],{},[75,18565,18566],{},"Arm",[75,18568,18569],{},"CPU IP、命令セット標準、エコシステムの核",[75,18571,18572],{},"2016年買収（$32B）",[60,18574,18575,18577,18580],{},[75,18576,16801],{},[75,18578,18579],{},"グラフ最適化AI処理（GC200/BOW等）",[75,18581,18582],{},"2024年買収",[60,18584,18585,18587,18590],{},[75,18586,16888],{},[75,18588,18589],{},"Armベースクラウドサーバーへの最適化",[75,18591,18592],{},"2025年$6.5Bで買収発表",[19,18594,18595],{},"このセグメントは「Armのエコシステムを基盤に、AIワークロードに特化したシリコンを垂直統合する」という事業戦略なんですよね。VCリターンではなく事業IRR（グループシナジー含む）で評価される性質のものなんですよね。",[205,18597],{},[32,18599,18601],{"id":18600},"領域量子-長期だが資本が戻り始めている","領域④：量子 — 長期だが資本が戻り始めている",[122,18603,18605],{"id":18604},"量子投資の商用化フェーズへの移行","量子投資の「商用化フェーズ」への移行",[19,18607,18608],{},"2020年前後の量子ブームは、多くのハードウェアスタートアップが誕生し、その後の市場との乖離で一部が撤退・合併しました。しかし2024〜2025年には「プラットフォーム競争」「クラウド提供」を前提にした大型ラウンドが戻り始めています。",[19,18610,18611,18612,18615],{},"変化の核心は：「量子コンピュータが量子優位を達成できるか」ではなく、",[23,18613,18614],{},"「エラー訂正・クラウド提供・ソフトウェアスタックをセットで商用化できるか」"," という評価軸への移行なんですよね。",[122,18617,18619],{"id":18618},"ケーススタディquantum-circuitsseries-b-60m2024年","ケーススタディ：Quantum Circuits（Series B > $60M、2024年）",[19,18621,18622],{},"Sequoiaがリード投資家として参加したQuantum Circuitsのシリーズ B は、フォールトトレラント量子コンピュータの商用化を狙う。",[19,18624,18625],{},"同社の特徴は超伝導量子回路の設計とエラー訂正への集中なんですよね。「ノイズのある中規模量子（NISQ）」ではなく、長期的にスケールするフォールトトレラントアーキテクチャへの賭けとして位置づけられる。",[122,18627,18629],{"id":18628},"ケーススタディrigetti-computing79m2020年series-c","ケーススタディ：Rigetti Computing（$79M、2020年Series C）",[19,18631,18632],{},"Bessemerがリードしたこの案件は、量子ハードウェア+クラウド提供（Quantum Cloud Services）という垂直統合モデルなんですよね。",[19,18634,18635],{},"後にSPACでNASDAQ上場（2022年）し、量子スタートアップとして初期の公開市場実績を残しました。クラウド提供という「ソフト的な入り口」を持つことで、企業顧客へのアクセスを維持する戦略は、量子の商用化において重要な設計判断だった。",[19,18637,18638],{},[23,18639,18640],{},"量子投資の評価基準（現在）：",[54,18642,18643,18653],{},[57,18644,18645],{},[60,18646,18647,18650],{},[63,18648,18649],{},"評価軸",[63,18651,18652],{},"詳細",[70,18654,18655,18663,18671,18678],{},[60,18656,18657,18660],{},[75,18658,18659],{},"技術",[75,18661,18662],{},"エラー率・量子ビット数・コヒーレンス時間",[60,18664,18665,18668],{},[75,18666,18667],{},"商用化",[75,18669,18670],{},"クラウドAPI・SDK・顧客導入実績",[60,18672,18673,18675],{},[75,18674,18528],{},[75,18676,18677],{},"アルゴリズム・ソフトウェアパートナー",[60,18679,18680,18682],{},[75,18681,18334],{},[75,18683,18684],{},"大手クラウド（AWS/Azure/GCP）への買収またはIPO",[205,18686],{},[32,18688,18690],{"id":18689},"代表投資案件-完全版テーブル","代表投資案件 完全版テーブル",[54,18692,18693,18715],{},[57,18694,18695],{},[60,18696,18697,18699,18701,18704,18706,18708,18710,18713],{},[63,18698,16344],{},[63,18700,16638],{},[63,18702,18703],{},"投資年",[63,18705,16641],{},[63,18707,16644],{},[63,18709,16647],{},[63,18711,18712],{},"ステージ",[63,18714,16353],{},[70,18716,18717,18736,18754,18772,18790,18809,18827,18845,18864,18883,18902,18922,18941,18960,18978,18997],{},[60,18718,18719,18721,18723,18725,18727,18729,18731,18734],{},[75,18720,16363],{},[75,18722,16766],{},[75,18724,16216],{},[75,18726,16769],{},[75,18728,16772],{},[75,18730,16722],{},[75,18732,18733],{},"B",[75,18735,13781],{},[60,18737,18738,18740,18742,18744,18746,18748,18750,18752],{},[75,18739,16363],{},[75,18741,16732],{},[75,18743,16727],{},[75,18745,15238],{},[75,18747,16737],{},[75,18749,16740],{},[75,18751,15238],{},[75,18753,12091],{},[60,18755,18756,18758,18760,18762,18764,18766,18768,18770],{},[75,18757,16363],{},[75,18759,16749],{},[75,18761,16727],{},[75,18763,15238],{},[75,18765,16754],{},[75,18767,16757],{},[75,18769,15238],{},[75,18771,12091],{},[60,18773,18774,18776,18778,18780,18782,18784,18786,18788],{},[75,18775,16363],{},[75,18777,16819],{},[75,18779,1018],{},[75,18781,15238],{},[75,18783,16824],{},[75,18785,16827],{},[75,18787,15238],{},[75,18789,13781],{},[60,18791,18792,18794,18796,18798,18800,18802,18804,18807],{},[75,18793,16903],{},[75,18795,16906],{},[75,18797,16284],{},[75,18799,16909],{},[75,18801,16912],{},[75,18803,16915],{},[75,18805,18806],{},"E",[75,18808,13781],{},[60,18810,18811,18813,18815,18817,18819,18821,18823,18825],{},[75,18812,16380],{},[75,18814,16836],{},[75,18816,1018],{},[75,18818,15238],{},[75,18820,16841],{},[75,18822,16844],{},[75,18824,15238],{},[75,18826,13781],{},[60,18828,18829,18831,18833,18835,18837,18839,18841,18843],{},[75,18830,16380],{},[75,18832,16924],{},[75,18834,16284],{},[75,18836,16769],{},[75,18838,16912],{},[75,18840,16931],{},[75,18842,18733],{},[75,18844,13781],{},[60,18846,18847,18849,18851,18853,18855,18857,18859,18862],{},[75,18848,16396],{},[75,18850,16783],{},[75,18852,16216],{},[75,18854,16786],{},[75,18856,16789],{},[75,18858,16740],{},[75,18860,18861],{},"Early",[75,18863,13781],{},[60,18865,18866,18868,18870,18872,18874,18876,18878,18881],{},[75,18867,16396],{},[75,18869,16871],{},[75,18871,1018],{},[75,18873,16874],{},[75,18875,16665],{},[75,18877,16879],{},[75,18879,18880],{},"D",[75,18882,13781],{},[60,18884,18885,18887,18889,18891,18893,18895,18897,18900],{},[75,18886,16412],{},[75,18888,16853],{},[75,18890,1018],{},[75,18892,16856],{},[75,18894,16859],{},[75,18896,16862],{},[75,18898,18899],{},"A",[75,18901,13781],{},[60,18903,18904,18906,18908,18910,18912,18914,18916,18919],{},[75,18905,16412],{},[75,18907,16697],{},[75,18909,16692],{},[75,18911,16681],{},[75,18913,16702],{},[75,18915,16705],{},[75,18917,18918],{},"C",[75,18920,18921],{},"スイス",[60,18923,18924,18926,18929,18931,18933,18935,18937,18939],{},[75,18925,16412],{},[75,18927,18928],{},"Rigetti Computing",[75,18930,16692],{},[75,18932,16681],{},[75,18934,16719],{},[75,18936,16722],{},[75,18938,18918],{},[75,18940,13781],{},[60,18942,18943,18945,18947,18949,18951,18953,18955,18957],{},[75,18944,16412],{},[75,18946,16659],{},[75,18948,16654],{},[75,18950,16662],{},[75,18952,16665],{},[75,18954,16668],{},[75,18956,18733],{},[75,18958,18959],{},"イスラエル",[60,18961,18962,18964,18966,18968,18970,18972,18974,18976],{},[75,18963,16429],{},[75,18965,16678],{},[75,18967,16673],{},[75,18969,16681],{},[75,18971,16684],{},[75,18973,16687],{},[75,18975,18918],{},[75,18977,13781],{},[60,18979,18980,18982,18984,18986,18988,18990,18992,18994],{},[75,18981,16798],{},[75,18983,16801],{},[75,18985,16216],{},[75,18987,16804],{},[75,18989,16807],{},[75,18991,16810],{},[75,18993,16804],{},[75,18995,18996],{},"英国",[60,18998,18999,19001,19003,19005,19007,19009,19011,19013],{},[75,19000,16798],{},[75,19002,16888],{},[75,19004,1018],{},[75,19006,16804],{},[75,19008,16893],{},[75,19010,16896],{},[75,19012,16804],{},[75,19014,13781],{},[205,19016],{},[32,19018,19019],{"id":19019},"シナリオ別の領域別アロケーション",[19,19021,19022],{},"ここまでの整理を踏まえて、3つのシナリオで領域別のアロケーションをどう振り分けるかを置いておきます。",[54,19024,19025,19043],{},[57,19026,19027],{},[60,19028,19029,19031,19033,19035,19038,19041],{},[63,19030,601],{},[63,19032,16953],{},[63,19034,16827],{},[63,19036,19037],{},"インターコネクト",[63,19039,19040],{},"新計算",[63,19042,16722],{},[70,19044,19045,19061,19076],{},[60,19046,19047,19050,19052,19055,19057,19059],{},[75,19048,19049],{},"ベース：継続拡張",[75,19051,624],{},[75,19053,19054],{},"コア",[75,19056,19054],{},[75,19058,638],{},[75,19060,633],{},[60,19062,19063,19066,19068,19070,19072,19074],{},[75,19064,19065],{},"メイン：パッケージング律速",[75,19067,646],{},[75,19069,675],{},[75,19071,630],{},[75,19073,638],{},[75,19075,633],{},[60,19077,19078,19081,19083,19085,19087,19089],{},[75,19079,19080],{},"テール：減速＋投資冬",[75,19082,665],{},[75,19084,675],{},[75,19086,638],{},[75,19088,668],{},[75,19090,668],{},[19,19092,19093,19094,19097],{},"メインシナリオで効くのは、",[23,19095,19096],{},"CPOとAIスイッチング","だと見ています。私の現場感覚で言うと、銅配線の物理限界はもはや「設計でなんとかする」レベルを超えていて、Co-Package側にしか解決の余地がない。ここはVCのモメンタムも実装の必然性も両方揃っている、珍しいスポットです。",[205,19099],{},[32,19101,17032],{"id":17031},[19,19103,19104],{},"私の足元のスタンスを一言で言うと、**「『大型ラウンド＝確実性が高い』は誤解で、『非対称性が大きいから大金を出している』」**です。",[19,19106,19107],{},"Unconventional AIの$475Mシードも、Nexthop/Ayar Labsの$500Mも、「確度が高いから大金を出している」わけではない。**「外れたとき以上に、当たったときのカテゴリ創造価値が大きいから大金を出している」**のです。",[19,19109,19110,19111,19114],{},"技術観点で個人的に最も興味深いのはCPOです。光I/OとチップをCo-Packageするという思想は、半導体・光学・パッケージングという全く異なる産業の交差点にある。",[23,19112,19113],{},"日本には半導体装置・光部品・精密実装という3つの強み","があり、このCPO領域は日本企業にとっても参入余地のある数少ない「最先端の交差点」だと思う。",[19,19116,19117],{},"次回（連載③）は戦略的示唆と投資提案——どうポートフォリオを組むか、短中長期でどこに賭けるかを整理します。",[205,19119],{},[19,19121,19122],{},[802,19123,804],{},[10,19125,19126,19130,19137,19141,19144,19151,19158,19160,19162,19182,19184,19188,19383,19385,19389,19393,19396,19400,19403,19410,19414,19417,19420,19425,19476,19478,19482,19486,19489,19495,19499,19502,19505,19511,19515,19518,19521,19525,19528,19535,19539,19542,19545,19550,19601,19603,19607,19611,19614,19628,19635,19639,19642,19645,19665,19672,19678,19682,19727,19734,19736,19740,19744,19750,19756,19760,19763,19767,19770,19775,19818,19820,19824,20105,20107,20111,20114,20182,20189,20191,20193,20198,20201,20204,20207,20209],{"lang":809},[14,19127,19129],{"id":19128},"why-design-data-movement-packaging-attract-vc-capital","Why Design, Data Movement & Packaging Attract VC Capital",[19,19131,19132,19133,19136],{},"I led automotive EEPROM process development and ran an internal DFM (Design For Manufacturing) rollout that delivered a 10% yield improvement. From the floor, what's blindingly obvious is that ",[23,19134,19135],{},"design, process, and packaging are not separate problems"," — they rate-limit each other in a coupled system. So when Tier-1 VCs simultaneously pour capital into all three layers, that lines up with the engineering reality, not against it. Part 2 walks through the technical necessity behind each domain through the representative deals.",[32,19138,19140],{"id":19139},"why-peripheral-layers-are-catching-the-big-checks","Why \"Peripheral Layers\" Are Catching the Big Checks",[19,19142,19143],{},"Let me reset the framing flat first.",[19,19145,19146,19147,19150],{},"In Part 1, the key claim was that Tier-1 VCs bet on AI's ",[802,19148,19149],{},"constraints",", not its compute. The logic that powers Part 2 is straightforward: as GPU performance scales exponentially, the surrounding layers become the limiting factor. Connecting 10,000 GPUs needs switches and interconnects. Exceeding single-chip performance limits needs packaging (CPO/Chiplet). Growing design complexity needs EDA and verification automation.",[19,19152,19153,19154,19157],{},"Why are large rounds flowing into each of these areas ",[802,19155,19156],{},"right now","? I'll work through the representative deals as case studies.",[205,19159],{},[32,19161,17101],{"id":17100},[312,19163,19164,19170,19176],{},[315,19165,19166,19169],{},[23,19167,19168],{},"Investment ROI is shifting from single-chip to full-stack (design→packaging→operations)."," CPO, coherent DSP, and AI switching — areas that solve operational constraints — are now commanding $500M+ rounds.",[315,19171,19172,19175],{},[23,19173,19174],{},"Generative AI has increased design complexity, making EDA automation (Agentic AI) a new investment theme."," With lower manufacturing risk and faster iteration, design automation is an attractive early-stage semiconductor play.",[315,19177,19178,19181],{},[23,19179,19180],{},"New compute (analog/quantum) remains a long-term theme, but capital flows to companies that can connect research to product to market."," Technical feasibility alone is insufficient — customer, cloud, and government partnerships are now evaluation criteria.",[205,19183],{},[32,19185,19187],{"id":19186},"technology-stack-investment-mapping","Technology Stack × Investment Mapping",[54,19189,19190,19208],{},[57,19191,19192],{},[60,19193,19194,19197,19200,19202,19205],{},[63,19195,19196],{},"Tech Domain",[63,19198,19199],{},"Representative Co. (VC)",[63,19201,17573],{},[63,19203,19204],{},"Amount",[63,19206,19207],{},"Role",[70,19209,19210,19224,19238,19252,19266,19280,19294,19309,19324,19340,19355,19369],{},[60,19211,19212,19214,19217,19219,19221],{},[75,19213,17727],{},[75,19215,19216],{},"Ricursive (Sequoia)",[75,19218,15238],{},[75,19220,18082],{},[75,19222,19223],{},"Accelerate chip design cycles",[60,19225,19226,19228,19231,19233,19235],{},[75,19227,17727],{},[75,19229,19230],{},"ChipAgents (Bessemer)",[75,19232,16856],{},[75,19234,18097],{},[75,19236,19237],{},"Agentic AI for design/verification",[60,19239,19240,19242,19245,19247,19249],{},[75,19241,16740],{},[75,19243,19244],{},"Akeana (Kleiner Perkins)",[75,19246,17693],{},[75,19248,18112],{},[75,19250,19251],{},"Open CPU instruction set",[60,19253,19254,19256,19259,19261,19263],{},[75,19255,16705],{},[75,19257,19258],{},"Kandou (Bessemer)",[75,19260,16681],{},[75,19262,18127],{},[75,19264,19265],{},"High-speed chip-to-chip PHY",[60,19267,19268,19270,19273,19275,19277],{},[75,19269,17772],{},[75,19271,19272],{},"Retym (Kleiner Perkins)",[75,19274,16874],{},[75,19276,18142],{},[75,19278,19279],{},"Signal processing for optical links",[60,19281,19282,19284,19287,19289,19291],{},[75,19283,17802],{},[75,19285,19286],{},"Ayar Labs (Sequoia related)",[75,19288,16909],{},[75,19290,18157],{},[75,19292,19293],{},"Silicon photonics to bypass copper limits",[60,19295,19296,19299,19302,19304,19306],{},[75,19297,19298],{},"AI Switch",[75,19300,19301],{},"Nexthop AI (a16z)",[75,19303,16769],{},[75,19305,18157],{},[75,19307,19308],{},"Networking for large-scale AI clusters",[60,19310,19311,19314,19317,19319,19321],{},[75,19312,19313],{},"Analog compute",[75,19315,19316],{},"Unconventional AI (a16z)",[75,19318,15238],{},[75,19320,18186],{},[75,19322,19323],{},"Probabilistic computing for power efficiency",[60,19325,19326,19329,19332,19334,19337],{},[75,19327,19328],{},"AI chip (M&A)",[75,19330,19331],{},"Graphcore (SoftBank)",[75,19333,16804],{},[75,19335,19336],{},"Undisclosed",[75,19338,19339],{},"UK graph processor",[60,19341,19342,19345,19348,19350,19352],{},[75,19343,19344],{},"CPU/Arm (M&A)",[75,19346,19347],{},"Ampere (SoftBank)",[75,19349,16804],{},[75,19351,18217],{},[75,19353,19354],{},"Arm-based cloud server CPU",[60,19356,19357,19359,19362,19364,19366],{},[75,19358,17639],{},[75,19360,19361],{},"Rigetti (Bessemer)",[75,19363,16681],{},[75,19365,18232],{},[75,19367,19368],{},"Quantum cloud + hardware",[60,19370,19371,19373,19376,19378,19380],{},[75,19372,17639],{},[75,19374,19375],{},"Quantum Circuits (Sequoia)",[75,19377,16769],{},[75,19379,18247],{},[75,19381,19382],{},"Fault-tolerant quantum",[205,19384],{},[32,19386,19388],{"id":19387},"domain-designipeda-betting-on-development-lead-time-reduction","Domain①: Design/IP/EDA — Betting on Development Lead Time Reduction",[122,19390,19392],{"id":19391},"why-vcs-are-entering-design-automation-now","Why VCs Are Entering Design Automation Now",[19,19394,19395],{},"Advanced chip design complexity grows with each process node. At 7nm→3nm→2nm, design rule complexity increases exponentially, and a single tapeout can cost tens of millions of dollars. Verification, simulation, and debugging now consume the majority of the design cycle. The thesis: apply AI agents to automate the most time-intensive parts of this workflow.",[122,19397,19399],{"id":19398},"case-study-ricursive-intelligence-35m-2025-seed","Case Study: Ricursive Intelligence ($35M, 2025 Seed)",[19,19401,19402],{},"Sequoia's lead on this $35M seed is the clearest signal that a top-tier VC is entering chip design automation at the earliest stage.",[19,19404,19405,19406,19409],{},"The key advantage: ",[23,19407,19408],{},"high software content",". Without owning fabrication, the approach of embedding AI into design flows carries relatively lower manufacturing risk. The main competitive risk is from Synopsys and Cadence — EDA giants with deep customer relationships and M&A firepower. The winning position is likely \"differentiated workflow integration\" or \"dominant capability in a specific design step (e.g., verification, static timing analysis).\"",[122,19411,19413],{"id":19412},"case-study-chipagents-21m-2025-series-a","Case Study: ChipAgents ($21M, 2025 Series A)",[19,19415,19416],{},"Bessemer-led, ChipAgents automates the verification and debug loop using agentic AI.",[19,19418,19419],{},"The framing: \"a significant fraction of design engineering time is spent in verification and bug-fixing cycles.\" An agentic AI system that can define tasks, execute, verify, and iterate autonomously in the EDA toolchain reduces that cycle. The goal is not reducing code written, but automating the verification-to-fix loop itself.",[19,19421,19422],{},[23,19423,19424],{},"EDA Automation Investment Logic:",[54,19426,19427,19435],{},[57,19428,19429],{},[60,19430,19431,19433],{},[63,19432,8964],{},[63,19434,861],{},[70,19436,19437,19444,19452,19460,19468],{},[60,19438,19439,19441],{},[75,19440,17139],{},[75,19442,19443],{},"Global chip designers × design cycle cost reduction",[60,19445,19446,19449],{},[75,19447,19448],{},"Manufacturing risk",[75,19450,19451],{},"Low (primarily software)",[60,19453,19454,19457],{},[75,19455,19456],{},"Competitive risk",[75,19458,19459],{},"M&A target for EDA incumbents (Synopsys/Cadence)",[60,19461,19462,19465],{},[75,19463,19464],{},"Exit paths",[75,19466,19467],{},"M&A (EDA giants or chip design companies) or IPO",[60,19469,19470,19473],{},[75,19471,19472],{},"Timeline",[75,19474,19475],{},"Relatively short (2–5 years)",[205,19477],{},[32,19479,19481],{"id":19480},"domain-interconnectdata-movement-connecting-gpus-is-the-main-arena","Domain②: Interconnect/Data Movement — \"Connecting GPUs\" is the Main Arena",[122,19483,19485],{"id":19484},"the-copper-bottleneck-and-the-rise-of-silicon-photonics","The Copper Bottleneck and the Rise of Silicon Photonics",[19,19487,19488],{},"In large-scale AI clusters (thousands to tens of thousands of GPUs), data movement between chips, racks, and data centers is the limiting factor across bandwidth, latency, and power simultaneously.",[134,19490,19493],{"className":19491,"code":19492,"language":139},[137],"Copper interconnect: bandwidth ↑ → power ↑↑↑, distance → signal loss\nSilicon photonics: optical transmission → high bandwidth, low power, long distance\nCo-Packaged Optics (CPO): integrate optical I/O into the chip package\n        → bypass the physical limits of copper\n",[141,19494,19492],{"__ignoreMap":143},[122,19496,19498],{"id":19497},"case-study-ayar-labs-500m-2026-series-e","Case Study: Ayar Labs ($500M, 2026 Series E)",[19,19500,19501],{},"This round, valuing Ayar Labs at $3.75B, is one of the largest fundraises in the CPO space.",[19,19503,19504],{},"The presence of NVIDIA and AMD as strategic investors is critical. Their participation is not purely financial — it signals that CPO technology will be integrated into next-generation product roadmaps. The participation of a Sequoia-related vehicle validates that pure financial returns are also available at this valuation.",[19,19506,19507,19510],{},[23,19508,19509],{},"Investment logic",": The physical limits of copper (bandwidth density, power) are fundamentally hard to engineer around. If CPO becomes the standard packaging architecture for AI compute, companies with the manufacturing technology, IP, and test capability will command significant value.",[122,19512,19514],{"id":19513},"case-study-retym-75m-2025-series-d","Case Study: Retym ($75M, 2025 Series D)",[19,19516,19517],{},"Kleiner Perkins-backed Retym specializes in coherent DSP optimized for cloud and AI workloads.",[19,19519,19520],{},"The differentiator: tailoring the DSP architecture to specific \"distance ranges\" — intra-datacenter (tens to hundreds of meters), metro (tens of km), and long-haul (hundreds of km+). Each range has different optimal DSP design. Rather than a universal chip, Retym targets specific operational regimes with purpose-built silicon.",[122,19522,19524],{"id":19523},"case-study-kandou-923m-2020-series-c","Case Study: Kandou ($92.3M, 2020 Series C)",[19,19526,19527],{},"Bessemer-led, Kandou's Matterhorn IP has demonstrated deployment in USB4 and Thunderbolt implementations.",[19,19529,19530,19531,19534],{},"The strategy rides the ",[23,19532,19533],{},"standard upgrade cycle",": as device and server standards evolve (USB4 Gen3, etc.), implementation requirements change, and Kandou's IP captures value at each transition. A recurring, standards-cycle-driven business model with a proven IP track record.",[122,19536,19538],{"id":19537},"case-study-nexthop-ai-500m-2026-series-b","Case Study: Nexthop AI ($500M, 2026 Series B)",[19,19540,19541],{},"a16z's $500M investment in Nexthop AI frames AI-era Ethernet switching as the next infrastructure layer.",[19,19543,19544],{},"The thesis: \"AI clusters are growing in scale faster than traditional switching architectures can adapt. Existing switches are not optimized for AI communication patterns (all-to-all collective communication).\" Nexthop builds AI-native switching to address this.",[19,19546,19547],{},[23,19548,19549],{},"Interconnect Investment Common Logic:",[54,19551,19552,19560],{},[57,19553,19554],{},[60,19555,19556,19558],{},[63,19557,8964],{},[63,19559,861],{},[70,19561,19562,19570,19578,19586,19594],{},[60,19563,19564,19567],{},[75,19565,19566],{},"Technical driver",[75,19568,19569],{},"GPU/accelerator scaling makes I/O the bottleneck",[60,19571,19572,19575],{},[75,19573,19574],{},"Customer base",[75,19576,19577],{},"Hyperscalers (Google/AWS/Microsoft/Meta)",[60,19579,19580,19583],{},[75,19581,19582],{},"Capital intensity",[75,19584,19585],{},"Medium to high (manufacturing qualification required)",[60,19587,19588,19591],{},[75,19589,19590],{},"Risks",[75,19592,19593],{},"Production yield, supplier qualification, geopolitics",[60,19595,19596,19598],{},[75,19597,19464],{},[75,19599,19600],{},"M&A (major chip companies) or IPO",[205,19602],{},[32,19604,19606],{"id":19605},"domain-ai-accelerators-new-compute-betting-on-computing-paradigms","Domain③: AI Accelerators / New Compute — Betting on Computing Paradigms",[122,19608,19610],{"id":19609},"two-approaches-mega-seed-vs-vertical-integration","Two Approaches: Mega-Seed vs. Vertical Integration",[19,19612,19613],{},"Two distinct approaches exist for the \"change the computing paradigm\" thesis:",[552,19615,19616,19622],{},[315,19617,19618,19621],{},[23,19619,19620],{},"Mega-seed in a startup"," (a16z × Unconventional AI): High risk, but success redefines the market.",[315,19623,19624,19627],{},[23,19625,19626],{},"Acquisition-based vertical integration"," (SoftBank × Graphcore/Ampere): Lower technology realization risk, but integration cost and market timing risk remain.",[19,19629,19630,19631,19634],{},"Both are ",[23,19632,19633],{},"ecosystem positioning plays"," — not simply chip investments.",[122,19636,19638],{"id":19637},"case-study-unconventional-ai-475m-2025-seed","Case Study: Unconventional AI ($475M, 2025 Seed)",[19,19640,19641],{},"One of the largest seed rounds in semiconductor investment history, co-led by a16z.",[19,19643,19644],{},"The analog/mixed-signal approach for probabilistic computing differs fundamentally from current digital CMOS architectures. Key challenges:",[312,19646,19647,19653,19659],{},[315,19648,19649,19652],{},[23,19650,19651],{},"Manufacturing reproducibility",": Analog circuits are sensitive to process variation",[315,19654,19655,19658],{},[23,19656,19657],{},"Toolchain",": Design and verification tools are immature",[315,19660,19661,19664],{},[23,19662,19663],{},"Ecosystem",": Software/framework adaptation required",[19,19666,19667,19668,19671],{},"But if successful, the potential for ",[23,19669,19670],{},"orders-of-magnitude power efficiency gains"," represents an overwhelming competitive advantage over current GPU architectures.",[19,19673,19674,19677],{},[23,19675,19676],{},"Why a16z makes this bet",": With its Strategic Partnerships program institutionalizing enterprise and government deployment paths, a16z can afford the early-stage cost of locking in a category-defining architecture. The cost of not owning the option is higher than the cost of the bet.",[122,19679,19681],{"id":19680},"softbanks-vertical-integration-arm-graphcore-ampere","SoftBank's Vertical Integration: Arm × Graphcore × Ampere",[54,19683,19684,19695],{},[57,19685,19686],{},[60,19687,19688,19690,19692],{},[63,19689,1228],{},[63,19691,19207],{},[63,19693,19694],{},"SoftBank Acquisition",[70,19696,19697,19707,19717],{},[60,19698,19699,19701,19704],{},[75,19700,18566],{},[75,19702,19703],{},"CPU IP, ISA standard, ecosystem core",[75,19705,19706],{},"Acquired 2016 ($32B)",[60,19708,19709,19711,19714],{},[75,19710,16801],{},[75,19712,19713],{},"Graph-optimized AI processing",[75,19715,19716],{},"Acquired 2024",[60,19718,19719,19721,19724],{},[75,19720,16888],{},[75,19722,19723],{},"Arm-based cloud server optimization",[75,19725,19726],{},"$6.5B acquisition announced 2025",[19,19728,19729,19730,19733],{},"This is a ",[23,19731,19732],{},"business strategy",", not a fund strategy. Evaluated on group synergy IRR, not VC fund returns.",[205,19735],{},[32,19737,19739],{"id":19738},"domain-quantum-long-term-but-capital-is-returning","Domain④: Quantum — Long-Term, But Capital Is Returning",[122,19741,19743],{"id":19742},"the-shift-to-commercialization-phase-quantum-investment","The Shift to \"Commercialization Phase\" Quantum Investment",[19,19745,19746,19747,849],{},"The quantum boom of the early 2020s saw many hardware startups emerge, then face market mismatch. However, 2024–2025 has seen large rounds return — now structured around ",[23,19748,19749],{},"platform competition + cloud delivery",[19,19751,19752,19753],{},"The evaluation shift: from \"can quantum achieve quantum advantage?\" to ",[23,19754,19755],{},"\"can the company commercialize error correction, cloud delivery, and software stacks as an integrated system?\"",[122,19757,19759],{"id":19758},"case-study-quantum-circuits-series-b-60m-2024","Case Study: Quantum Circuits (Series B >$60M, 2024)",[19,19761,19762],{},"Sequoia's participation in this round signals conviction in fault-tolerant quantum architecture over near-term NISQ approaches. The bet is on the architecture that will scale — not the first to market.",[122,19764,19766],{"id":19765},"case-study-rigetti-computing-79m-2020-series-c","Case Study: Rigetti Computing ($79M, 2020 Series C)",[19,19768,19769],{},"Bessemer-led, with Quantum Cloud Services as the vertical integration model. Subsequently NASDAQ-listed via SPAC (2022). Having a cloud API entry point maintains enterprise access even before hardware reaches quantum advantage — a critical commercialization design choice.",[19,19771,19772],{},[23,19773,19774],{},"Quantum Investment Evaluation Criteria (Current):",[54,19776,19777,19786],{},[57,19778,19779],{},[60,19780,19781,19784],{},[63,19782,19783],{},"Axis",[63,19785,861],{},[70,19787,19788,19795,19803,19810],{},[60,19789,19790,19792],{},[75,19791,1236],{},[75,19793,19794],{},"Error rates, qubit count, coherence time",[60,19796,19797,19800],{},[75,19798,19799],{},"Commercialization",[75,19801,19802],{},"Cloud API, SDK, customer deployments",[60,19804,19805,19807],{},[75,19806,19663],{},[75,19808,19809],{},"Algorithm and software partners",[60,19811,19812,19815],{},[75,19813,19814],{},"Exit",[75,19816,19817],{},"Acquisition by major cloud (AWS/Azure/GCP) or IPO",[205,19819],{},[32,19821,19823],{"id":19822},"complete-investment-reference-table","Complete Investment Reference Table",[54,19825,19826,19844],{},[57,19827,19828],{},[60,19829,19830,19832,19834,19836,19838,19840,19842],{},[63,19831,16344],{},[63,19833,17570],{},[63,19835,17136],{},[63,19837,17573],{},[63,19839,17576],{},[63,19841,19196],{},[63,19843,17279],{},[70,19845,19846,19862,19878,19894,19910,19926,19942,19958,19974,19990,20006,20023,20039,20056,20072,20089],{},[60,19847,19848,19850,19852,19854,19856,19858,19860],{},[75,19849,16363],{},[75,19851,16766],{},[75,19853,16216],{},[75,19855,16769],{},[75,19857,16772],{},[75,19859,17639],{},[75,19861,14368],{},[60,19863,19864,19866,19868,19870,19872,19874,19876],{},[75,19865,16363],{},[75,19867,16732],{},[75,19869,16727],{},[75,19871,15238],{},[75,19873,16737],{},[75,19875,16740],{},[75,19877,12867],{},[60,19879,19880,19882,19884,19886,19888,19890,19892],{},[75,19881,16363],{},[75,19883,16749],{},[75,19885,16727],{},[75,19887,15238],{},[75,19889,16754],{},[75,19891,16757],{},[75,19893,12867],{},[60,19895,19896,19898,19900,19902,19904,19906,19908],{},[75,19897,16363],{},[75,19899,16819],{},[75,19901,1018],{},[75,19903,15238],{},[75,19905,16824],{},[75,19907,17727],{},[75,19909,14368],{},[60,19911,19912,19914,19916,19918,19920,19922,19924],{},[75,19913,17793],{},[75,19915,16906],{},[75,19917,16284],{},[75,19919,16909],{},[75,19921,16912],{},[75,19923,17802],{},[75,19925,14368],{},[60,19927,19928,19930,19932,19934,19936,19938,19940],{},[75,19929,16380],{},[75,19931,16836],{},[75,19933,1018],{},[75,19935,15238],{},[75,19937,16841],{},[75,19939,17742],{},[75,19941,14368],{},[60,19943,19944,19946,19948,19950,19952,19954,19956],{},[75,19945,16380],{},[75,19947,16924],{},[75,19949,16284],{},[75,19951,16769],{},[75,19953,16912],{},[75,19955,17817],{},[75,19957,14368],{},[60,19959,19960,19962,19964,19966,19968,19970,19972],{},[75,19961,16396],{},[75,19963,16783],{},[75,19965,16216],{},[75,19967,17693],{},[75,19969,16789],{},[75,19971,16740],{},[75,19973,14368],{},[60,19975,19976,19978,19980,19982,19984,19986,19988],{},[75,19977,16396],{},[75,19979,16871],{},[75,19981,1018],{},[75,19983,16874],{},[75,19985,16665],{},[75,19987,17772],{},[75,19989,14368],{},[60,19991,19992,19994,19996,19998,20000,20002,20004],{},[75,19993,16412],{},[75,19995,16853],{},[75,19997,1018],{},[75,19999,16856],{},[75,20001,16859],{},[75,20003,17757],{},[75,20005,14368],{},[60,20007,20008,20010,20012,20014,20016,20018,20020],{},[75,20009,16412],{},[75,20011,16697],{},[75,20013,16692],{},[75,20015,16681],{},[75,20017,16702],{},[75,20019,16705],{},[75,20021,20022],{},"Switzerland",[60,20024,20025,20027,20029,20031,20033,20035,20037],{},[75,20026,16412],{},[75,20028,18928],{},[75,20030,16692],{},[75,20032,16681],{},[75,20034,16719],{},[75,20036,17639],{},[75,20038,14368],{},[60,20040,20041,20043,20045,20047,20049,20051,20053],{},[75,20042,16412],{},[75,20044,16659],{},[75,20046,16654],{},[75,20048,17591],{},[75,20050,16665],{},[75,20052,17596],{},[75,20054,20055],{},"Israel",[60,20057,20058,20060,20062,20064,20066,20068,20070],{},[75,20059,16429],{},[75,20061,16678],{},[75,20063,16673],{},[75,20065,16681],{},[75,20067,16684],{},[75,20069,16687],{},[75,20071,14368],{},[60,20073,20074,20076,20078,20080,20082,20084,20086],{},[75,20075,16798],{},[75,20077,16801],{},[75,20079,16216],{},[75,20081,16804],{},[75,20083,19336],{},[75,20085,17712],{},[75,20087,20088],{},"UK",[60,20090,20091,20093,20095,20097,20099,20101,20103],{},[75,20092,16798],{},[75,20094,16888],{},[75,20096,1018],{},[75,20098,16804],{},[75,20100,16893],{},[75,20102,16896],{},[75,20104,14368],{},[205,20106],{},[32,20108,20110],{"id":20109},"scenario-based-domain-allocation","Scenario-Based Domain Allocation",[19,20112,20113],{},"Putting allocation against the three scenarios:",[54,20115,20116,20134],{},[57,20117,20118],{},[60,20119,20120,20122,20124,20126,20129,20132],{},[63,20121,1389],{},[63,20123,1392],{},[63,20125,17727],{},[63,20127,20128],{},"Interconnect",[63,20130,20131],{},"New compute",[63,20133,17639],{},[70,20135,20136,20151,20167],{},[60,20137,20138,20140,20142,20144,20146,20148],{},[75,20139,17853],{},[75,20141,624],{},[75,20143,1415],{},[75,20145,1415],{},[75,20147,1426],{},[75,20149,20150],{},"Wait",[60,20152,20153,20156,20158,20161,20163,20165],{},[75,20154,20155],{},"Main — packaging-rate-limited",[75,20157,646],{},[75,20159,20160],{},"Hold",[75,20162,1418],{},[75,20164,1426],{},[75,20166,20150],{},[60,20168,20169,20172,20174,20176,20178,20180],{},[75,20170,20171],{},"Tail — slowdown + investment chill",[75,20173,665],{},[75,20175,20160],{},[75,20177,1426],{},[75,20179,1454],{},[75,20181,1454],{},[19,20183,20184,20185,20188],{},"In the main scenario, ",[23,20186,20187],{},"CPO and AI switching"," are where I'd expect the highest returns. From operational experience: copper interconnect's physical limits have moved past the \"fix it in design\" zone — the fix has to be at the Co-Package layer. That's a rare spot where VC momentum and engineering necessity both line up.",[205,20190],{},[32,20192,17921],{"id":17920},[19,20194,1558,20195,849],{},[23,20196,20197],{},"a large round doesn't mean high probability — it means high asymmetry, and that's why the check is large",[19,20199,20200],{},"The $475M seed into Unconventional AI, and the $500M rounds into Nexthop and Ayar Labs, aren't bets on certainty. They're bets where \"if wrong, lose the check; if right, create a category worth hundreds of billions.\" The asymmetry justifies the size.",[19,20202,20203],{},"The single area I find most technically compelling is CPO. Co-packaging optical I/O with silicon sits at the intersection of semiconductors, photonics, and precision packaging — three industries where Japan has globally competitive players. This is one of the few \"cutting-edge intersections\" where Japanese industry has a shot at participating at the frontier.",[19,20205,20206],{},"Part 3 synthesizes the strategic implications — how to construct a portfolio across these themes, and how to think about risk-reward across short, medium, and long time horizons.",[205,20208],{},[19,20210,20211],{},[802,20212,1603],{},{"title":143,"searchDepth":1605,"depth":1605,"links":20214},[20215,20216,20217,20218,20223,20230,20235,20240,20241,20242,20243,20244,20245,20246,20251,20258,20263,20268,20269,20270],{"id":18008,"depth":1605,"text":18009},{"id":16166,"depth":1605,"text":16166},{"id":18047,"depth":1605,"text":18048},{"id":18255,"depth":1605,"text":18256,"children":20219},[20220,20221,20222],{"id":18259,"depth":1610,"text":18260},{"id":18266,"depth":1610,"text":18267},{"id":18280,"depth":1610,"text":18281},{"id":18350,"depth":1605,"text":18351,"children":20224},[20225,20226,20227,20228,20229],{"id":18354,"depth":1610,"text":18354},{"id":18366,"depth":1610,"text":18367},{"id":18382,"depth":1610,"text":18383},{"id":18392,"depth":1610,"text":18393},{"id":18402,"depth":1610,"text":18403},{"id":18470,"depth":1605,"text":18471,"children":20231},[20232,20233,20234],{"id":18474,"depth":1610,"text":18475},{"id":18502,"depth":1610,"text":18503},{"id":18541,"depth":1610,"text":18542},{"id":18600,"depth":1605,"text":18601,"children":20236},[20237,20238,20239],{"id":18604,"depth":1610,"text":18605},{"id":18618,"depth":1610,"text":18619},{"id":18628,"depth":1610,"text":18629},{"id":18689,"depth":1605,"text":18690},{"id":19019,"depth":1605,"text":19019},{"id":17031,"depth":1605,"text":17032},{"id":19139,"depth":1605,"text":19140},{"id":17100,"depth":1605,"text":17101},{"id":19186,"depth":1605,"text":19187},{"id":19387,"depth":1605,"text":19388,"children":20247},[20248,20249,20250],{"id":19391,"depth":1610,"text":19392},{"id":19398,"depth":1610,"text":19399},{"id":19412,"depth":1610,"text":19413},{"id":19480,"depth":1605,"text":19481,"children":20252},[20253,20254,20255,20256,20257],{"id":19484,"depth":1610,"text":19485},{"id":19497,"depth":1610,"text":19498},{"id":19513,"depth":1610,"text":19514},{"id":19523,"depth":1610,"text":19524},{"id":19537,"depth":1610,"text":19538},{"id":19605,"depth":1605,"text":19606,"children":20259},[20260,20261,20262],{"id":19609,"depth":1610,"text":19610},{"id":19637,"depth":1610,"text":19638},{"id":19680,"depth":1610,"text":19681},{"id":19738,"depth":1605,"text":19739,"children":20264},[20265,20266,20267],{"id":19742,"depth":1610,"text":19743},{"id":19758,"depth":1610,"text":19759},{"id":19765,"depth":1610,"text":19766},{"id":19822,"depth":1605,"text":19823},{"id":20109,"depth":1605,"text":20110},{"id":17920,"depth":1605,"text":17921},"GPU性能が伸びるほど設計・I/O・実装がAIの限界要因になる。CPO・SerDes・EDA自動化・量子の代表案件を、半導体プロセス開発出身のCVCの視点で深掘り。$500M超の大型ラウンドが示す投資ロジックを読み解く。",{"date":17979,"image":20273,"alt":20274,"tags":20275,"tagsEn":20276,"published":1666},"/blogs-img/blog-semiconductor-vc-2.png","半導体投資の技術領域別ケーススタディ",[17983,17984,1657,44,1660],[17986,17987,1662,837,1665],"/blogs/7-semiconductor-vc-series-2",{"title":17994,"description":20271},"blogs/7-semiconductor-vc-series-2","Y9LfzeMj7U2z5bHwMoEBJduUvPfvW6N0Nqq5OfYJ0X8",{"id":20282,"title":20283,"body":20284,"description":21867,"extension":1651,"meta":21868,"navigation":1666,"ogImage":21869,"path":21873,"seo":21874,"stem":21875,"__hash__":21876},"content/blogs/8-semiconductor-vc-series-3.md","半導体×VCの勝ち筋は「ボラティリティを設計する」こと｜連載③戦略的示唆と投資提案 | Semiconductor VC Series③: Designing for Volatility",{"type":7,"value":20285,"toc":21821},[20286,21044,21046],[10,20287,20288,20292,20299,20303,20306,20313,20316,20319,20333,20335,20337,20345,20353,20361,20363,20367,20483,20487,20493,20495,20499,20503,20564,20570,20572,20576,20648,20654,20656,20660,20732,20738,20740,20744,20863,20866,20872,20874,20877,20880,20886,20889,20892,20918,20922,20936,20938,20941,20945,20989,20993,21025,21027,21029,21032,21035,21038,21040],{"lang":12},[14,20289,20291],{"id":20290},"半導体vcの勝ち筋はボラティリティを設計すること","半導体×VCの勝ち筋は「ボラティリティを設計する」こと",[19,20293,20294,20295,20298],{},"私は半導体プロセス開発の現場でEEPROMの量産化と歩留まり改善を主導してきた人間です。半導体産業の「サイクリカル性」は教科書の話ではなく、4製品以上の量産立ち上げを通じて骨で覚えた肌感覚です。",[23,20296,20297],{},"サイクルの波に乗るのは難しいが、サイクルを無効化する構造的ポジションは取れる","——これがTier1 VCの戦略を10年見続けて私が至った結論で、連載最終回はその視点で全体を統合します。",[32,20300,20302],{"id":20301},"ボラティリティを設計するとは何か","「ボラティリティを設計する」とは何か",[19,20304,20305],{},"最初に、本連載の問題意識をフラットに置きます。",[19,20307,20308,20309,20312],{},"半導体産業は本質的にサイクリカルです。需要は爆発的に膨らみ、供給は2〜3年後に追いつき、価格は暴落する——半導体の歴史はその繰り返しでした。しかし今、Tier1 VCが半導体スタートアップに数億ドルを投じるとき、彼らは「サイクルの波に乗る」のではなく、「",[23,20310,20311],{},"サイクル自体を無効化する構造的ポジション","」を狙っています。",[19,20314,20315],{},"それが「ボラティリティを設計する（designing for volatility）」という概念です。チップの価格が上下しても需要が安定する「標準化されたインターフェース技術」（CPO、UCIeなど）、チップ世代交代のたびに価値が高まる「設計インフラ」（EDA自動化、IP再利用）、特定のハードウェアベンダーに依存しない「ソフトウェア定義の計算基盤」——いずれも、価格競争に巻き込まれにくいレイヤーへの賭けです。",[19,20317,20318],{},"本連載の最終回では、連載①②で整理したVCの戦略・技術・案件データを踏まえ、以下を体系化します。",[552,20320,20321,20324,20327,20330],{},[315,20322,20323],{},"Tier1 VC 5社の戦略比較と差異の本質",[315,20325,20326],{},"短期（〜18ヶ月）・中期（2〜4年）・長期（5年〜）の投資機会とリスク",[315,20328,20329],{},"マクロ環境（市場規模・装置投資・政策）のサマリ",[315,20331,20332],{},"個人投資家・機関投資家・事業会社それぞれへの示唆",[205,20334],{},[32,20336,16166],{"id":16166},[792,20338,20339],{},[19,20340,20341,20344],{},[23,20342,20343],{},"①「AI需要→半導体ボトルネック→スタートアップ機会」という構造は2026年以降も継続する","\nGPU性能のスケーリングが続く限り、設計・データ移動・パッケージングというボトルネックは解消されない。Tier1 VCはこの構造的需要を見越し、すでに2023〜2025年に大型投資を実行済みなんですよね。",[792,20346,20347],{},[19,20348,20349,20352],{},[23,20350,20351],{},"②VC各社の\"賭け方\"は大きく異なる：エコシステム型 vs. 垂直統合型","\nSequoia・a16zがスタートアップ群への分散投資でエコシステムを形成するのに対し、SoftBankはAmpere（$6.5B評価額）など垂直統合型の大型単独案件を選好します。Kleiner PerkinsはIPO経路を重視した米国製造復活テーマに集中し、Bessemerはデータセンターインフラとのシナジーを軸に据える。",[792,20354,20355],{},[19,20356,20357,20360],{},[23,20358,20359],{},"③長期は「新計算原理」が最大の非線形リターン源だが、商用化リスクは高い","\nアナログ計算・量子・フォトニクスはいずれも現在の投資期待値が低い一方、商用化に成功した場合のリターンは既存手法と非線形に差がつく。VCはこれらをポートフォリオの「オプション」として保有し、中期の確実な案件でポートフォリオを安定させる戦略をとっています。",[205,20362],{},[32,20364,20366],{"id":20365},"tier1-vc戦略比較差異の本質","Tier1 VC戦略比較：差異の本質",[54,20368,20369,20387],{},[57,20370,20371],{},[60,20372,20373,20375,20378,20381,20384],{},[63,20374,16344],{},[63,20376,20377],{},"投資の\"型\"",[63,20379,20380],{},"代表案件",[63,20382,20383],{},"提携・エコシステム構造",[63,20385,20386],{},"想定出口",[70,20388,20389,20408,20427,20445,20464],{},[60,20390,20391,20396,20399,20402,20405],{},[75,20392,20393],{},[23,20394,20395],{},"Sequoia Capital",[75,20397,20398],{},"分散・エコシステム型。設計自動化・新計算原理を中心に複数スタートアップへ初期投資",[75,20400,20401],{},"Ricursive ($35M)、Unconventional Computing ($475M)",[75,20403,20404],{},"NVIDIA・Synopsys・Coreweaveとのネットワークで被投資企業を繋ぐ",[75,20406,20407],{},"M&A（EDA大手）またはIPO（計算インフラ）",[60,20409,20410,20415,20418,20421,20424],{},[75,20411,20412],{},[23,20413,20414],{},"Andreessen Horowitz (a16z)",[75,20416,20417],{},"テーマ型・大型集中。「AIファクトリー」インフラを横断するデータ移動と新世代メモリ",[75,20419,20420],{},"Ayar Labs ($500M)、Nexthop AI ($500M)",[75,20422,20423],{},"クラウドプロバイダー（AWS・Azure・GCP）とのコマーシャルパイプライン形成",[75,20425,20426],{},"データセンター事業者へのM&AまたはIPO",[60,20428,20429,20433,20436,20439,20442],{},[75,20430,20431],{},[23,20432,16396],{},[75,20434,20435],{},"政策・製造復活テーマ型。CHIPS法補助金の受益企業、米国製造回帰",[75,20437,20438],{},"Retym ($75M)、Kandou Technologies ($92.3M)",[75,20440,20441],{},"SEMI協会・米国防総省DARPAプログラムとの連携",[75,20443,20444],{},"IPO（CHIPS法関連企業への市場期待）",[60,20446,20447,20452,20455,20458,20461],{},[75,20448,20449],{},[23,20450,20451],{},"Bessemer Venture Partners",[75,20453,20454],{},"データセンターインフラ連動型。既存クラウド投資との シナジーを重視",[75,20456,20457],{},"Quantum Circuits ($60M+)、ChipAgents ($21M)",[75,20459,20460],{},"既存クラウド・SaaS投資先との顧客共有",[75,20462,20463],{},"M&A（クラウド大手による内製化）",[60,20465,20466,20471,20474,20477,20480],{},[75,20467,20468],{},[23,20469,20470],{},"SoftBank Vision Fund",[75,20472,20473],{},"垂直統合・メガラウンド型。単一企業に大規模資本を集中、グループ内シナジー優先",[75,20475,20476],{},"Ampere Computing ($6.5B評価額)、Graphcore買収",[75,20478,20479],{},"SBグループ（ARM・T-Mobile・WeWork残存網）との垂直統合",[75,20481,20482],{},"長期保有またはIPO（ARM上場の先例）",[122,20484,20486],{"id":20485},"戦略の本質的差異エコシステム型vs垂直統合型","戦略の本質的差異：「エコシステム型」vs.「垂直統合型」",[134,20488,20491],{"className":20489,"code":20490,"language":139},[137],"エコシステム型（Sequoia / a16z / KP / Bessemer）\n────────────────────────────────────────────\nVC\n ├── スタートアップA（設計）\n ├── スタートアップB（インターコネクト）\n ├── スタートアップC（パッケージング）\n └── スタートアップD（新計算）\n        │\n        ▼\n 各社が相互に顧客・サプライヤー関係を形成\n → VC自身がエコシステムのハブになる\n\n垂直統合型（SoftBank）\n────────────────────────────────────────────\nSoftBank\n └── Ampere Computing（独自Arm CPU設計）\n      └── ARM（SB傘下）からIPライセンス\n           └── データセンター向け直販モデル\n                → グループ内でValue Chainを完結\n",[141,20492,20490],{"__ignoreMap":143},[205,20494],{},[32,20496,20498],{"id":20497},"投資機会とリスク3つの時間軸","投資機会とリスク：3つの時間軸",[122,20500,20502],{"id":20501},"短期18ヶ月設計効率化とeda自動化","短期（〜18ヶ月）：設計効率化とEDA自動化",[54,20504,20505,20521],{},[57,20506,20507],{},[60,20508,20509,20512,20515,20518],{},[63,20510,20511],{},"機会領域",[63,20513,20514],{},"投資テーマ",[63,20516,20517],{},"代表スタートアップ",[63,20519,20520],{},"主なリスク",[70,20522,20523,20536,20550],{},[60,20524,20525,20527,20530,20533],{},[75,20526,16862],{},[75,20528,20529],{},"LLMによる回路検証・バグ修正の自動化で設計期間を30〜50%短縮",[75,20531,20532],{},"Ricursive（Seq.）、ChipAgents（Bessemer）",[75,20534,20535],{},"Synopsys/Cadenceがin-house LLM開発を加速、M&Aで吸収する可能性",[60,20537,20538,20541,20544,20547],{},[75,20539,20540],{},"IP再利用・モジュール化",[75,20542,20543],{},"既存IP資産のAI検索・カスタマイズにより新チップ設計コスト削減",[75,20545,20546],{},"複数の非公開ステルス案件",[75,20548,20549],{},"大手ファブ（TSMC・Samsung）が独自IPライブラリを強化、差別化困難",[60,20551,20552,20555,20558,20561],{},[75,20553,20554],{},"サプライチェーン可視化",[75,20556,20557],{},"地政学リスクにより部品調達の不確実性が増大、可視化SaaSへの需要急増",[75,20559,20560],{},"（非公開多数）",[75,20562,20563],{},"景気後退局面ではソフトウェア支出が最初に削減される",[19,20565,20566,20569],{},[23,20567,20568],{},"短期の投資判断基準","：収益化期間が短い（18ヶ月以内に初期顧客獲得可能）か、規制の影響を受けにくいソフトウェアファーストモデルかどうかが鍵。EDA自動化はこの条件を満たすが、競合（Cadence・Synopsys）の大型R&D予算には要注意。",[205,20571],{},[122,20573,20575],{"id":20574},"中期24年データ移動cpo先端パッケージング","中期（2〜4年）：データ移動・CPO・先端パッケージング",[54,20577,20578,20590],{},[57,20579,20580],{},[60,20581,20582,20584,20586,20588],{},[63,20583,20511],{},[63,20585,20514],{},[63,20587,20517],{},[63,20589,20520],{},[70,20591,20592,20606,20620,20634],{},[60,20593,20594,20597,20600,20603],{},[75,20595,20596],{},"CPO（コパッケージド光学）",[75,20598,20599],{},"GPU-メモリ間の銅配線を光配線に置換、帯域幅を10倍・消費電力を60〜80%削減",[75,20601,20602],{},"Ayar Labs（a16z、$500M）",[75,20604,20605],{},"量産歩留まり（光コンポーネントの製造難易度）、TSMC/IntelのCPO内製化リスク",[60,20607,20608,20611,20614,20617],{},[75,20609,20610],{},"高速SerDes・電気インターコネクト",[75,20612,20613],{},"銅配線の延命技術としてデータセンター内配線を高効率化",[75,20615,20616],{},"Kandou Technologies（KP、$92.3M）",[75,20618,20619],{},"光学への移行が予想より早まった場合のレガシー化リスク",[60,20621,20622,20625,20628,20631],{},[75,20623,20624],{},"先端パッケージング（Chiplet/3D IC）",[75,20626,20627],{},"Chiplet標準規格（UCIe）による異種チップ統合",[75,20629,20630],{},"Nexthop AI（a16z、$500M）",[75,20632,20633],{},"ファブレスとファブの間のIP所有権交渉の複雑化",[60,20635,20636,20639,20642,20645],{},[75,20637,20638],{},"次世代メモリインターフェース",[75,20640,20641],{},"HBM3/4の後継として帯域幅と電力効率のトレードオフを再設計",[75,20643,20644],{},"（SK Hynix・Micronとの協業スタートアップ）",[75,20646,20647],{},"DRAMメーカーの自社開発能力向上でサードパーティ不要論",[19,20649,20650,20653],{},[23,20651,20652],{},"中期の投資判断基準","：TSMC・Samsung・Intelの先端パッケージング拡張計画と競合するかどうか。競合するなら提携関係（JDA）を確立できているかが生死を分ける。Ayar Labsがa16zの支援のもとTSMCと共同開発体制を構築しているのはこの典型例。",[205,20655],{},[122,20657,20659],{"id":20658},"長期5年新計算原理と量子","長期（5年〜）：新計算原理と量子",[54,20661,20662,20674],{},[57,20663,20664],{},[60,20665,20666,20668,20670,20672],{},[63,20667,20511],{},[63,20669,20514],{},[63,20671,20517],{},[63,20673,20520],{},[70,20675,20676,20690,20704,20718],{},[60,20677,20678,20681,20684,20687],{},[75,20679,20680],{},"アナログ計算",[75,20682,20683],{},"デジタル変換コストを排除し推論電力を100分の1以下に削減、エッジAI向け",[75,20685,20686],{},"Unconventional Computing（Seq.、$475M）",[75,20688,20689],{},"プログラマビリティ（汎用性）の制限、開発者エコシステムの未成熟",[60,20691,20692,20695,20698,20701],{},[75,20693,20694],{},"量子コンピューティング",[75,20696,20697],{},"素因数分解・量子化学シミュレーション・最適化問題での超越性",[75,20699,20700],{},"Quantum Circuits（Bessemer、$60M+）、Rigetti（$79M）",[75,20702,20703],{},"エラー訂正技術の成熟度、「量子優位性」の商用ユースケース特定",[60,20705,20706,20709,20712,20715],{},[75,20707,20708],{},"フォトニック計算",[75,20710,20711],{},"光を用いた行列演算でAI推論の消費電力を抜本的に削減",[75,20713,20714],{},"（非公開複数社）",[75,20716,20717],{},"シリコンフォトニクスの製造インフラ未成熟、スケールアップコスト",[60,20719,20720,20723,20726,20729],{},[75,20721,20722],{},"ニューロモルフィック",[75,20724,20725],{},"脳の神経回路を模倣した非同期計算、超低電力センサー処理",[75,20727,20728],{},"（学術スピンアウト複数）",[75,20730,20731],{},"汎用計算への転用困難、ニッチ市場からの脱出経路不明確",[19,20733,20734,20737],{},[23,20735,20736],{},"長期の投資判断基準","：「論文→プロダクト→市場」の接続性。量子の場合、論文数と特許出願は世界トップクラスだが、商用ユースケースへの接続（特にエラー率の商用許容水準到達）が見えていない企業は評価を保留すべきなんですよね。一方、アナログ計算はエッジ推論という明確な市場セグメントが存在し、中期以降の商用化シナリオが描きやすい。",[205,20739],{},[32,20741,20743],{"id":20742},"マクロ環境市場規模装置投資政策","マクロ環境：市場規模・装置投資・政策",[54,20745,20746,20767],{},[57,20747,20748],{},[60,20749,20750,20752,20755,20758,20761,20764],{},[63,20751,225],{},[63,20753,20754],{},"2024実績",[63,20756,20757],{},"2025予測",[63,20759,20760],{},"2026予測",[63,20762,20763],{},"2027予測",[63,20765,20766],{},"出典",[70,20768,20769,20787,20805,20825,20844],{},[60,20770,20771,20774,20777,20779,20781,20784],{},[75,20772,20773],{},"世界半導体市場規模",[75,20775,20776],{},"$627B",[75,20778,16231],{},[75,20780,16150],{},[75,20782,20783],{},"$1.1T+",[75,20785,20786],{},"WSTS 2025年秋季予測",[60,20788,20789,20792,20795,20798,20800,20802],{},[75,20790,20791],{},"300mm装置支出",[75,20793,20794],{},"$107B",[75,20796,20797],{},"$119B",[75,20799,16287],{},[75,20801,16298],{},[75,20803,20804],{},"SEMI 2026年予測",[60,20806,20807,20810,20813,20816,20819,20822],{},[75,20808,20809],{},"AI半導体（GPU/NPU/ASIC）",[75,20811,20812],{},"$130B",[75,20814,20815],{},"$200B+",[75,20817,20818],{},"$280B+",[75,20820,20821],{},"$380B+",[75,20823,20824],{},"IDC/Gartner各社推計",[60,20826,20827,20830,20833,20835,20838,20841],{},[75,20828,20829],{},"米CHIPS法補助金執行額",[75,20831,20832],{},"$8B",[75,20834,14544],{},[75,20836,20837],{},"$52B（総額上限）",[75,20839,20840],{},"〃",[75,20842,20843],{},"米商務省",[60,20845,20846,20849,20852,20855,20858,20860],{},[75,20847,20848],{},"中国半導体輸出規制強化",[75,20850,20851],{},"HBM・EDA規制",[75,20853,20854],{},"EUV輸出制限継続",[75,20856,20857],{},"さらなる強化見込み",[75,20859,20840],{},[75,20861,20862],{},"BIS/経産省",[122,20864,20865],{"id":20865},"政策環境が投資判断に与える影響",[134,20867,20870],{"className":20868,"code":20869,"language":139},[137],"CHIPS法・EU半導体法\n    │\n    ├── 補助金の受益企業（Intel Foundry・TSMC米国拠点・Samsung Texas）\n    │        └── → 建設・装置・材料スタートアップに間接需要\n    │\n    └── 地政学リスクの優先度上昇\n             └── → 米国内設計・製造のスタートアップへのVC資金流入加速\n\n中国規制（BIS 輸出規制）\n    │\n    ├── 中国向けAI半導体（H800→H20）の販売制限\n    │        └── → 中国内半導体自給自足投資の爆発（华为/寒武纪等）\n    │\n    └── 先進EDA・製造装置の中国輸出禁止\n             └── → 欧米スタートアップへの反射的恩恵（競合中国企業の能力制限）\n",[141,20871,20869],{"__ignoreMap":143},[205,20873],{},[32,20875,20876],{"id":20876},"ポートフォリオ構成の考え方",[19,20878,20879],{},"Tier1 VCの実際のポートフォリオ配分から読み取れるのは、以下のような「時間軸別の3層構造」だ：",[134,20881,20884],{"className":20882,"code":20883,"language":139},[137],"ポートフォリオの理想的な時間軸配分（イメージ）\n─────────────────────────────────────\n短期（〜18ヶ月）  ████████████████░░░░  40%\n  └── EDA自動化・設計SaaS・サプライチェーン可視化\n\n中期（2〜4年）    ██████████████████░░  45%\n  └── CPO・インターコネクト・先端パッケージング・AIアクセラレータ\n\n長期（5年〜）     ██████░░░░░░░░░░░░░░  15%\n  └── アナログ計算・量子・フォトニクス・ニューロモルフィック\n─────────────────────────────────────\n",[141,20885,20883],{"__ignoreMap":143},[122,20887,20888],{"id":20888},"個人投資家への示唆",[19,20890,20891],{},"個人投資家が直接VCファンドに参加できるケースは限られるが、以下の方法で間接的な半導体スタートアップエクスポージャーを取得できる：",[552,20893,20894,20900,20906,20912],{},[315,20895,20896,20899],{},[23,20897,20898],{},"上場半導体企業（NVIDIA・ASML・Lam Research・TSMC ADR）","：スタートアップの顧客であり、スタートアップが解決するボトルネックから直接利益を得る",[315,20901,20902,20905],{},[23,20903,20904],{},"半導体特化ETF（SOXX・SMH）","：分散効果はあるが、大型株中心でスタートアップの成長性は反映されにくい",[315,20907,20908,20911],{},[23,20909,20910],{},"SPAC・二次流通市場","：Rigetti（量子）のようにSPAC経由で上場した銘柄は、高リスク・高ボラティリティで個人投資家がアクセスしやすい",[315,20913,20914,20917],{},[23,20915,20916],{},"CVCを持つ事業会社株","：SoftBankグループ・Intel Capital・Qualcomm Ventures母体企業への投資は、VCポートフォリオへの間接アクセスとなる",[122,20919,20921],{"id":20920},"機関投資家事業会社への示唆","機関投資家・事業会社への示唆",[312,20923,20924,20930],{},[315,20925,20926,20929],{},[23,20927,20928],{},"LP（大学基金・年金）","：Sequoia / a16z / Bessemerへのコミットメントは半導体の長期テーマと合致するが、J-Curveリスク（投資後3〜5年はリターンがマイナス）を許容できる長期資金が前提",[315,20931,20932,20935],{},[23,20933,20934],{},"事業会社CVC","：自社バリューチェーンのボトルネックを特定し、そこに対応するスタートアップへ戦略的投資することで、M&Aオプションと市場情報を同時に取得できる",[205,20937],{},[32,20939,20940],{"id":20940},"連載まとめと次の注目ポイント",[122,20942,20944],{"id":20943},"_3回の連載で明らかになったこと","3回の連載で明らかになったこと",[54,20946,20947,20957],{},[57,20948,20949],{},[60,20950,20951,20954],{},[63,20952,20953],{},"連載",[63,20955,20956],{},"核心メッセージ",[70,20958,20959,20969,20979],{},[60,20960,20961,20966],{},[75,20962,20963],{},[23,20964,20965],{},"連載①：概観と主要プレイヤー",[75,20967,20968],{},"Tier1 VCは「AI需要のボトルネック」という構造的欠如部分に集中している。チップ単体ではなく、設計・移動・実装・供給網という4層への分解投資が共通パターン",[60,20970,20971,20976],{},[75,20972,20973],{},[23,20974,20975],{},"連載②：技術別ケーススタディ",[75,20977,20978],{},"EDA自動化・CPO・先端パッケージング・新計算原理という4領域で約$2.5B規模の投資が2022〜2025年に集中執行された。技術的成熟度と商用化パスが投資規模を決定している",[60,20980,20981,20986],{},[75,20982,20983],{},[23,20984,20985],{},"連載③：戦略的示唆（本稿）",[75,20987,20988],{},"VCの「型」はエコシステム型と垂直統合型に二分される。短期はソフトウェア収益性、中期はCPO/パッケージング、長期はアナログ・量子がそれぞれのリターン源",[122,20990,20992],{"id":20991},"今後の注目ポイント2026年後半以降","今後の注目ポイント（2026年後半以降）",[552,20994,20995,21001,21007,21013,21019],{},[315,20996,20997,21000],{},[23,20998,20999],{},"Ayar Labs量産開始タイミング","：a16zが$500M投じたCPOの量産が2026〜2027年に本格化するか。歩留まりが商用水準（90%以上）に達するかが業界全体のCPO移行速度を決定する",[315,21002,21003,21006],{},[23,21004,21005],{},"Ampere ComputingのIPO計画","：SoftBankが2026年中に示唆しているAmpereのIPOが実現すれば、Arm系CPUのデータセンター向け市場シェアの公開データが初めて入手可能になる",[315,21008,21009,21012],{},[23,21010,21011],{},"CHIPS法補助金の第2フェーズ","：2026年後半に予定される米商務省の第2ラウンド補助金決定（ファウンドリー以外の設計・材料・EDA企業への配分拡大）が、スタートアップへの資金流入を加速させる可能性がある",[315,21014,21015,21018],{},[23,21016,21017],{},"中国の先進ファブリケーション突破口","：HuaweiのKirin/Ascend系チップが7nm相当に到達した実績を踏まえ、5nm以下への突破タイミングが欧米規制強化の速度と「技術ギャップ」の行方を決める",[315,21020,21021,21024],{},[23,21022,21023],{},"量子エラー訂正の商用許容水準到達","：Google（Willow）・IBM・Microsoftが競う論理量子ビットの実装が2027〜2028年に商用ユースケース（金融最適化・創薬）に接続できるか",[205,21026],{},[32,21028,17032],{"id":17031},[19,21030,21031],{},"半導体投資の本質は**「インフラのロックイン」**です。GPUもメモリも、単体では差別化できない。しかし、その間をつなぐデータ移動技術（CPO・SerDes）、上流の設計効率（EDA自動化）、下流の実装技術（先端パッケージング）は、一度採用されたらすぐには切り替えられない粘着性（スティッキネス）を持ちます。",[19,21033,21034],{},"Tier1 VCがこの領域に$2.5B以上を集中させているのは、「チップの価格競争に巻き込まれたくない」という合理的判断の結果です。価格競争は製造業の宿命だが、インフラはソフトウェア的なネットワーク効果を持てる——これは、私が現場で量産プロセスのコモディティ化と「装置・データ・プロセスIP」の固有価値を同時に経験してきたからこそ、強く同意できる構造です。",[19,21036,21037],{},"連載を通じて改めて感じたのは、**「半導体は地政学と分離できない」**という事実。CHIPS法・BIS輸出規制・日本の経済安保法——これらの政策変数が、純粋な技術評価と同等以上に投資判断を左右します。個人投資家として半導体セクターに向き合うなら、技術だけでなく政策カレンダーを同時に追うことが不可欠だと考えています。",[205,21039],{},[19,21041,21042],{},[802,21043,804],{},[205,21045],{},[10,21047,21048,21052,21059,21063,21066,21069,21084,21087,21101,21103,21105,21113,21121,21129,21131,21135,21247,21251,21257,21259,21263,21267,21329,21335,21337,21341,21413,21419,21421,21425,21497,21503,21505,21509,21618,21622,21628,21630,21634,21640,21646,21650,21653,21679,21683,21697,21699,21703,21707,21751,21755,21787,21789,21791,21798,21805,21812,21815,21817],{"lang":809},[14,21049,21051],{"id":21050},"semiconductor-vc-designing-for-volatility","Semiconductor VC: Designing for Volatility",[19,21053,21054,21055,21058],{},"I came up through the floor — leading EEPROM volume production and yield improvement on the process side. Semiconductor cyclicality isn't a textbook abstraction for me; it's something I felt in my bones across more than four product ramps. ",[23,21056,21057],{},"You can't reliably surf the cycle, but you can take structural positions that are immune to it."," That's the conclusion I've arrived at after watching Tier-1 VCs for a decade, and that's the lens that ties this final installment together.",[32,21060,21062],{"id":21061},"what-designing-for-volatility-means","What \"Designing for Volatility\" Means",[19,21064,21065],{},"Let me reset the question flat first.",[19,21067,21068],{},"The semiconductor industry is inherently cyclical. Demand explodes, supply catches up 2–3 years later, prices collapse — the history of semiconductors has always been one of boom and bust. But when Tier-1 VCs commit hundreds of millions of dollars to semiconductor startups today, they are not trying to \"ride the cycle.\" They are targeting structural positions that are immune to the cycle itself.",[19,21070,21071,21072,21075,21076,21079,21080,21083],{},"This is what \"designing for volatility\" means: investing in ",[23,21073,21074],{},"standardized interface technologies"," (CPO, UCIe) whose demand remains stable even when chip prices swing wildly; ",[23,21077,21078],{},"design infrastructure"," (EDA automation, IP reuse) that becomes more valuable with every chip generation transition; and ",[23,21081,21082],{},"software-defined computing foundations"," that are independent of any specific hardware vendor — bets on layers where price competition doesn't dominate.",[19,21085,21086],{},"In this final installment, I synthesize the strategy, technology, and deal data from Parts 1 and 2 into four key frameworks:",[552,21088,21089,21092,21095,21098],{},[315,21090,21091],{},"Comparative analysis of five Tier1 VCs and the essence of their strategic differences",[315,21093,21094],{},"Investment opportunities and risks across short (≤18 months), medium (2–4 years), and long (5+ years) time horizons",[315,21096,21097],{},"Macro environment summary: market size, equipment investment, and policy",[315,21099,21100],{},"Implications for individual investors, institutional investors, and corporate strategists",[205,21102],{},[32,21104,17101],{"id":17100},[792,21106,21107],{},[19,21108,21109,21112],{},[23,21110,21111],{},"① The \"AI demand → semiconductor bottleneck → startup opportunity\" structure will persist beyond 2026","\nAs long as GPU performance scaling continues, bottlenecks in design, data movement, and packaging will remain unsolved. Tier1 VCs have already anticipated this structural demand with major investments executed in 2023–2025.",[792,21114,21115],{},[19,21116,21117,21120],{},[23,21118,21119],{},"② VC firms' investment \"styles\" diverge sharply: ecosystem model vs. vertical integration model","\nWhile Sequoia and a16z build ecosystems through diversified investments in startup clusters, SoftBank concentrates capital in large single-company vertical integration bets (e.g., Ampere at $6.5B valuation). Kleiner Perkins focuses on the U.S. manufacturing revival thesis with clear IPO paths, while Bessemer emphasizes synergies with its existing data center infrastructure portfolio.",[792,21122,21123],{},[19,21124,21125,21128],{},[23,21126,21127],{},"③ Long-term, \"new computing paradigms\" offer the highest non-linear return potential — but commercialization risk is high","\nAnalog computing, quantum, and photonics all carry low near-term expected values, but successful commercialization creates non-linear return gaps versus conventional approaches. VCs hold these as \"options\" within diversified portfolios, using medium-term certainties to stabilize overall returns.",[205,21130],{},[32,21132,21134],{"id":21133},"tier1-vc-strategy-comparison-the-essence-of-differences","Tier1 VC Strategy Comparison: The Essence of Differences",[54,21136,21137,21155],{},[57,21138,21139],{},[60,21140,21141,21143,21146,21149,21152],{},[63,21142,16344],{},[63,21144,21145],{},"Investment Style",[63,21147,21148],{},"Signature Deals",[63,21150,21151],{},"Ecosystem/Partnership Structure",[63,21153,21154],{},"Expected Exit",[70,21156,21157,21175,21193,21211,21229],{},[60,21158,21159,21163,21166,21169,21172],{},[75,21160,21161],{},[23,21162,20395],{},[75,21164,21165],{},"Diversified ecosystem. Early-stage bets across design automation and novel computing paradigms",[75,21167,21168],{},"Ricursive ($35M), Unconventional Computing ($475M)",[75,21170,21171],{},"Network bridges portfolio companies to NVIDIA, Synopsys, Coreweave",[75,21173,21174],{},"M&A (EDA majors) or IPO (compute infra)",[60,21176,21177,21181,21184,21187,21190],{},[75,21178,21179],{},[23,21180,20414],{},[75,21182,21183],{},"Thematic, large concentrated bets on \"AI factory\" infrastructure spanning data movement and next-gen memory",[75,21185,21186],{},"Ayar Labs ($500M), Nexthop AI ($500M)",[75,21188,21189],{},"Commercial pipeline formation with cloud providers (AWS, Azure, GCP)",[75,21191,21192],{},"M&A by data center operators or IPO",[60,21194,21195,21199,21202,21205,21208],{},[75,21196,21197],{},[23,21198,16396],{},[75,21200,21201],{},"Policy/manufacturing revival theme. CHIPS Act beneficiaries, U.S. onshoring thesis",[75,21203,21204],{},"Retym ($75M), Kandou Technologies ($92.3M)",[75,21206,21207],{},"SEMI association, U.S. DoD/DARPA program linkages",[75,21209,21210],{},"IPO (market enthusiasm for CHIPS Act beneficiaries)",[60,21212,21213,21217,21220,21223,21226],{},[75,21214,21215],{},[23,21216,20451],{},[75,21218,21219],{},"Data center infrastructure synergy. Prioritizes fit with existing cloud investment portfolio",[75,21221,21222],{},"Quantum Circuits ($60M+), ChipAgents ($21M)",[75,21224,21225],{},"Customer sharing with existing cloud/SaaS portfolio companies",[75,21227,21228],{},"M&A (cloud majors internalizing capabilities)",[60,21230,21231,21235,21238,21241,21244],{},[75,21232,21233],{},[23,21234,20470],{},[75,21236,21237],{},"Vertical integration, mega-rounds. Concentrated capital in single companies; group synergy prioritized",[75,21239,21240],{},"Ampere Computing ($6.5B valuation), Graphcore acquisition",[75,21242,21243],{},"Vertical integration with SB group (Arm, T-Mobile, residual WeWork network)",[75,21245,21246],{},"Long-term hold or IPO (Arm IPO as precedent)",[122,21248,21250],{"id":21249},"the-essential-divide-ecosystem-vs-vertical-integration","The Essential Divide: Ecosystem vs. Vertical Integration",[134,21252,21255],{"className":21253,"code":21254,"language":139},[137],"Ecosystem Model (Sequoia / a16z / KP / Bessemer)\n────────────────────────────────────────────────\nVC\n ├── Startup A (Design Automation)\n ├── Startup B (Interconnect/CPO)\n ├── Startup C (Advanced Packaging)\n └── Startup D (New Computing Paradigm)\n        │\n        ▼\n Portfolio companies form mutual customer-supplier relationships\n → VC itself becomes the hub of a self-reinforcing ecosystem\n\nVertical Integration Model (SoftBank)\n────────────────────────────────────────────────\nSoftBank\n └── Ampere Computing (proprietary Arm CPU design)\n      └── IP licensed from Arm (SB subsidiary)\n           └── Direct-to-data-center sales model\n                → Full value chain captured within the group\n",[141,21256,21254],{"__ignoreMap":143},[205,21258],{},[32,21260,21262],{"id":21261},"investment-opportunities-and-risks-three-time-horizons","Investment Opportunities and Risks: Three Time Horizons",[122,21264,21266],{"id":21265},"short-term-18-months-design-efficiency-and-eda-automation","Short-Term (≤18 Months): Design Efficiency and EDA Automation",[54,21268,21269,21285],{},[57,21270,21271],{},[60,21272,21273,21276,21279,21282],{},[63,21274,21275],{},"Opportunity",[63,21277,21278],{},"Investment Theme",[63,21280,21281],{},"Representative Startups",[63,21283,21284],{},"Key Risks",[70,21286,21287,21301,21315],{},[60,21288,21289,21292,21295,21298],{},[75,21290,21291],{},"EDA Automation",[75,21293,21294],{},"LLM-powered circuit verification and bug correction reducing design time by 30–50%",[75,21296,21297],{},"Ricursive (Seq.), ChipAgents (Bessemer)",[75,21299,21300],{},"Synopsys/Cadence accelerating in-house LLM development; acquisition risk",[60,21302,21303,21306,21309,21312],{},[75,21304,21305],{},"IP Reuse/Modularization",[75,21307,21308],{},"AI-powered search and customization of existing IP assets reducing new chip design cost",[75,21310,21311],{},"Multiple undisclosed stealth-stage companies",[75,21313,21314],{},"Leading fabs (TSMC, Samsung) building proprietary IP libraries, crowding out differentiation",[60,21316,21317,21320,21323,21326],{},[75,21318,21319],{},"Supply Chain Visibility",[75,21321,21322],{},"Geopolitical risk driving demand for SaaS platforms that map procurement uncertainty",[75,21324,21325],{},"(Numerous undisclosed companies)",[75,21327,21328],{},"In a recession, software spend is the first to be cut",[19,21330,21331,21334],{},[23,21332,21333],{},"Short-term investment criteria",": Can the company acquire initial customers within 18 months? Is the model software-first and insulated from regulatory risk? EDA automation meets these conditions, but the large R&D budgets of Cadence and Synopsys demand constant vigilance.",[205,21336],{},[122,21338,21340],{"id":21339},"medium-term-24-years-data-movement-cpo-and-advanced-packaging","Medium-Term (2–4 Years): Data Movement, CPO, and Advanced Packaging",[54,21342,21343,21355],{},[57,21344,21345],{},[60,21346,21347,21349,21351,21353],{},[63,21348,21275],{},[63,21350,21278],{},[63,21352,21281],{},[63,21354,21284],{},[70,21356,21357,21371,21385,21399],{},[60,21358,21359,21362,21365,21368],{},[75,21360,21361],{},"CPO (Co-Packaged Optics)",[75,21363,21364],{},"Replacing copper GPU-to-memory interconnect with optical fiber; 10× bandwidth, 60–80% power reduction",[75,21366,21367],{},"Ayar Labs (a16z, $500M)",[75,21369,21370],{},"Production yield (optical component manufacturing complexity); TSMC/Intel CPO internalization risk",[60,21372,21373,21376,21379,21382],{},[75,21374,21375],{},"High-Speed SerDes and Electrical Interconnect",[75,21377,21378],{},"Extending copper's lifespan by improving intra-data-center wiring efficiency",[75,21380,21381],{},"Kandou Technologies (KP, $92.3M)",[75,21383,21384],{},"Legacy risk if optical migration happens faster than expected",[60,21386,21387,21390,21393,21396],{},[75,21388,21389],{},"Advanced Packaging (Chiplet/3D IC)",[75,21391,21392],{},"Heterogeneous chip integration via Chiplet standards (UCIe)",[75,21394,21395],{},"Nexthop AI (a16z, $500M)",[75,21397,21398],{},"IP ownership negotiation complexity between fabless companies and foundries",[60,21400,21401,21404,21407,21410],{},[75,21402,21403],{},"Next-Gen Memory Interfaces",[75,21405,21406],{},"Redesigning bandwidth vs. power tradeoffs as HBM3/4 successors",[75,21408,21409],{},"(Startups co-developing with SK Hynix, Micron)",[75,21411,21412],{},"DRAM vendors' growing in-house development capacity making third parties redundant",[19,21414,21415,21418],{},[23,21416,21417],{},"Medium-term investment criteria",": Does the company compete with TSMC's, Samsung's, or Intel's advanced packaging expansion plans? If so, has it established a JDA (Joint Development Agreement) partnership? This is a matter of survival. Ayar Labs' co-development structure with TSMC, backed by a16z, is the canonical example.",[205,21420],{},[122,21422,21424],{"id":21423},"long-term-5-years-new-computing-paradigms-and-quantum","Long-Term (5+ Years): New Computing Paradigms and Quantum",[54,21426,21427,21439],{},[57,21428,21429],{},[60,21430,21431,21433,21435,21437],{},[63,21432,21275],{},[63,21434,21278],{},[63,21436,21281],{},[63,21438,21284],{},[70,21440,21441,21455,21469,21483],{},[60,21442,21443,21446,21449,21452],{},[75,21444,21445],{},"Analog Computing",[75,21447,21448],{},"Eliminating digital conversion overhead; 100× inference power reduction for edge AI",[75,21450,21451],{},"Unconventional Computing (Seq., $475M)",[75,21453,21454],{},"Programmability (generality) limitations; underdeveloped developer ecosystem",[60,21456,21457,21460,21463,21466],{},[75,21458,21459],{},"Quantum Computing",[75,21461,21462],{},"Exceeding classical computation in prime factorization, quantum chemistry simulation, optimization",[75,21464,21465],{},"Quantum Circuits (Bessemer, $60M+), Rigetti ($79M)",[75,21467,21468],{},"Error correction maturity; identifying commercial use cases that achieve \"quantum advantage\"",[60,21470,21471,21474,21477,21480],{},[75,21472,21473],{},"Photonic Computing",[75,21475,21476],{},"Using light for matrix operations to dramatically reduce AI inference power consumption",[75,21478,21479],{},"(Multiple undisclosed companies)",[75,21481,21482],{},"Immature silicon photonics manufacturing infrastructure; scale-up cost",[60,21484,21485,21488,21491,21494],{},[75,21486,21487],{},"Neuromorphic Computing",[75,21489,21490],{},"Asynchronous computation mimicking neural circuits; ultra-low-power sensor processing",[75,21492,21493],{},"(Multiple academic spinouts)",[75,21495,21496],{},"Difficult to adapt for general-purpose compute; unclear path from niche markets",[19,21498,21499,21502],{},[23,21500,21501],{},"Long-term investment criteria",": Is there a clear \"paper → product → market\" connection? For quantum, the volume of papers and patents is world-class, but companies that cannot show a path to commercial error rates should be approached with caution. Analog computing, by contrast, has a well-defined target market in edge inference, making medium-to-long-term commercialization scenarios more legible.",[205,21504],{},[32,21506,21508],{"id":21507},"macro-environment-market-size-equipment-investment-and-policy","Macro Environment: Market Size, Equipment Investment, and Policy",[54,21510,21511,21532],{},[57,21512,21513],{},[60,21514,21515,21517,21520,21523,21526,21529],{},[63,21516,1015],{},[63,21518,21519],{},"2024 Actual",[63,21521,21522],{},"2025 Forecast",[63,21524,21525],{},"2026 Forecast",[63,21527,21528],{},"2027 Forecast",[63,21530,21531],{},"Source",[70,21533,21534,21550,21566,21582,21599],{},[60,21535,21536,21539,21541,21543,21545,21547],{},[75,21537,21538],{},"Global Semiconductor Market",[75,21540,20776],{},[75,21542,16231],{},[75,21544,16150],{},[75,21546,20783],{},[75,21548,21549],{},"WSTS Autumn 2025",[60,21551,21552,21555,21557,21559,21561,21563],{},[75,21553,21554],{},"300mm Equipment Spending",[75,21556,20794],{},[75,21558,20797],{},[75,21560,16287],{},[75,21562,16298],{},[75,21564,21565],{},"SEMI 2026 Forecast",[60,21567,21568,21571,21573,21575,21577,21579],{},[75,21569,21570],{},"AI Semiconductors (GPU/NPU/ASIC)",[75,21572,20812],{},[75,21574,20815],{},[75,21576,20818],{},[75,21578,20821],{},[75,21580,21581],{},"IDC/Gartner estimates",[60,21583,21584,21587,21589,21591,21594,21596],{},[75,21585,21586],{},"U.S. CHIPS Act Grants Executed",[75,21588,20832],{},[75,21590,14544],{},[75,21592,21593],{},"$52B (cap)",[75,21595,250],{},[75,21597,21598],{},"U.S. Commerce Dept.",[60,21600,21601,21604,21607,21610,21613,21615],{},[75,21602,21603],{},"China Semiconductor Export Controls",[75,21605,21606],{},"HBM + EDA",[75,21608,21609],{},"EUV export ban continues",[75,21611,21612],{},"Further tightening expected",[75,21614,250],{},[75,21616,21617],{},"BIS/METI",[122,21619,21621],{"id":21620},"how-policy-environment-affects-investment-decisions","How Policy Environment Affects Investment Decisions",[134,21623,21626],{"className":21624,"code":21625,"language":139},[137],"CHIPS Act / EU Chips Act\n    │\n    ├── Direct beneficiaries (Intel Foundry, TSMC Arizona, Samsung Texas)\n    │        └── → Indirect demand for construction/equipment/materials startups\n    │\n    └── Elevated priority on geopolitical risk management\n             └── → Accelerated VC capital flow into U.S.-designed/manufactured startups\n\nChina Export Controls (BIS)\n    │\n    ├── Sales restrictions on AI chips to China (H800 → H20)\n    │        └── → China's domestic semiconductor self-sufficiency investment explodes (Huawei/Cambricon)\n    │\n    └── Advanced EDA/equipment export ban to China\n             └── → Reflexive benefit for Western startups (Chinese competitors' capability constrained)\n",[141,21627,21625],{"__ignoreMap":143},[205,21629],{},[32,21631,21633],{"id":21632},"portfolio-construction-framework","Portfolio Construction Framework",[19,21635,21636,21637,8210],{},"Reading across the actual portfolio allocations of Tier1 VCs reveals a consistent ",[23,21638,21639],{},"three-layer structure by time horizon",[134,21641,21644],{"className":21642,"code":21643,"language":139},[137],"Ideal Portfolio Allocation by Time Horizon (Illustrative)\n──────────────────────────────────────────────────────\nShort-term (≤18 months)   ████████████████░░░░  40%\n  └── EDA automation · Design SaaS · Supply chain visibility\n\nMedium-term (2–4 years)   ██████████████████░░  45%\n  └── CPO · Interconnect · Advanced packaging · AI accelerators\n\nLong-term (5+ years)      ██████░░░░░░░░░░░░░░  15%\n  └── Analog computing · Quantum · Photonics · Neuromorphic\n──────────────────────────────────────────────────────\n",[141,21645,21643],{"__ignoreMap":143},[122,21647,21649],{"id":21648},"implications-for-individual-investors","Implications for Individual Investors",[19,21651,21652],{},"Individual investors rarely have direct access to VC funds, but indirect semiconductor startup exposure is available through:",[552,21654,21655,21661,21667,21673],{},[315,21656,21657,21660],{},[23,21658,21659],{},"Listed semiconductor companies (NVIDIA, ASML, Lam Research, TSMC ADR)",": Direct customers of startups, benefitting directly from the bottlenecks that startups are solving",[315,21662,21663,21666],{},[23,21664,21665],{},"Semiconductor ETFs (SOXX, SMH)",": Diversification benefit, but large-cap weighted — limited startup growth reflection",[315,21668,21669,21672],{},[23,21670,21671],{},"SPACs and secondary markets",": Companies like Rigetti (quantum), which went public via SPAC, offer individual investors high-risk, high-volatility access",[315,21674,21675,21678],{},[23,21676,21677],{},"Corporate parents of active CVCs",": Investing in SoftBank Group, Intel (Intel Capital parent), or Qualcomm (Qualcomm Ventures parent) provides indirect VC portfolio exposure",[122,21680,21682],{"id":21681},"implications-for-institutional-investors-and-corporates","Implications for Institutional Investors and Corporates",[312,21684,21685,21691],{},[315,21686,21687,21690],{},[23,21688,21689],{},"LPs (endowments, pension funds)",": Commitments to Sequoia / a16z / Bessemer align well with long-term semiconductor themes, but require capital that can tolerate the J-Curve (returns typically negative for 3–5 years post-commitment)",[315,21692,21693,21696],{},[23,21694,21695],{},"Corporate CVC",": Identify the bottleneck in your own value chain and strategically invest in the startup addressing it — simultaneously acquiring M&A optionality and competitive intelligence",[205,21698],{},[32,21700,21702],{"id":21701},"series-wrap-up-and-next-focal-points","Series Wrap-Up and Next Focal Points",[122,21704,21706],{"id":21705},"what-the-three-part-series-revealed","What the Three-Part Series Revealed",[54,21708,21709,21719],{},[57,21710,21711],{},[60,21712,21713,21716],{},[63,21714,21715],{},"Part",[63,21717,21718],{},"Core Message",[70,21720,21721,21731,21741],{},[60,21722,21723,21728],{},[75,21724,21725],{},[23,21726,21727],{},"Part 1: Overview and Key Players",[75,21729,21730],{},"Tier1 VCs are concentrating on the \"structural absence\" at AI infrastructure bottlenecks. The common pattern is decomposing investment across four layers — design, movement, packaging, and supply chain — rather than betting on chips themselves",[60,21732,21733,21738],{},[75,21734,21735],{},[23,21736,21737],{},"Part 2: Technology Deep Dives",[75,21739,21740],{},"Approximately $2.5B in investment was concentrated across EDA automation, CPO, advanced packaging, and new computing paradigms from 2022–2025. Technical maturity and commercial path clarity determine deal size",[60,21742,21743,21748],{},[75,21744,21745],{},[23,21746,21747],{},"Part 3: Strategic Implications (this piece)",[75,21749,21750],{},"VC styles bifurcate into ecosystem vs. vertical integration models. Short-term returns come from software profitability; medium-term from CPO/packaging; long-term from analog/quantum",[122,21752,21754],{"id":21753},"focal-points-for-h2-2026-and-beyond","Focal Points for H2 2026 and Beyond",[552,21756,21757,21763,21769,21775,21781],{},[315,21758,21759,21762],{},[23,21760,21761],{},"Ayar Labs volume production timeline",": Will a16z's $500M CPO bet reach volume manufacturing in 2026–2027? Achieving commercial yield rates (90%+) will determine the pace of CPO adoption across the entire industry",[315,21764,21765,21768],{},[23,21766,21767],{},"Ampere Computing IPO",": If SoftBank's hinted 2026 IPO for Ampere materializes, it will provide the first publicly available data on Arm-based CPU market share in data centers",[315,21770,21771,21774],{},[23,21772,21773],{},"CHIPS Act Phase 2 grants",": U.S. Commerce Department's second-round grant decisions in H2 2026 (expanding beyond foundries to design, materials, and EDA companies) could accelerate capital flows to startups",[315,21776,21777,21780],{},[23,21778,21779],{},"China's advanced fabrication breakthrough",": Given Huawei's demonstrated 7nm-equivalent Kirin/Ascend chips, the timing of any sub-5nm breakthrough will determine both Western regulatory response speed and the persistence of the \"technology gap\"",[315,21782,21783,21786],{},[23,21784,21785],{},"Quantum error correction hitting commercial tolerances",": Whether Google (Willow), IBM, and Microsoft's logical qubit implementations can connect to commercial use cases (financial optimization, drug discovery) by 2027–2028",[205,21788],{},[32,21790,17921],{"id":17920},[19,21792,21793,21794,21797],{},"The essence of semiconductor investment is ",[23,21795,21796],{},"infrastructure lock-in",". GPUs alone, memory alone — neither creates sustainable differentiation. But the data-movement technology bridging them (CPO, SerDes), the upstream design-efficiency layer (EDA automation), and the downstream assembly layer (advanced packaging) all have the kind of stickiness that resists easy substitution once adopted.",[19,21799,21800,21801,21804],{},"Tier-1 VCs concentrating $2.5B+ in these areas is the rational outcome of a simple observation: ",[23,21802,21803],{},"they don't want to be dragged into chip price competition",". Price wars are manufacturing's destiny, but infrastructure can achieve software-like network effects. From having watched the same dynamic play out in process IP, equipment data, and yield-improvement know-how on the floor, I think this framing is right.",[19,21806,21807,21808,21811],{},"What this series reinforced for me is that ",[23,21809,21810],{},"semiconductors are inseparable from geopolitics",". The CHIPS Act, BIS export controls, Japan's economic security legislation — these policy variables influence investment decisions at least as much as pure technical assessment.",[19,21813,21814],{},"For individual investors engaging with the semiconductor sector: tracking the policy calendar alongside the technology roadmap is no longer optional. It's table stakes.",[205,21816],{},[19,21818,21819],{},[802,21820,1603],{},{"title":143,"searchDepth":1605,"depth":1605,"links":21822},[21823,21824,21825,21828,21833,21836,21840,21844,21845,21846,21847,21850,21855,21858,21862,21866],{"id":20301,"depth":1605,"text":20302},{"id":16166,"depth":1605,"text":16166},{"id":20365,"depth":1605,"text":20366,"children":21826},[21827],{"id":20485,"depth":1610,"text":20486},{"id":20497,"depth":1605,"text":20498,"children":21829},[21830,21831,21832],{"id":20501,"depth":1610,"text":20502},{"id":20574,"depth":1610,"text":20575},{"id":20658,"depth":1610,"text":20659},{"id":20742,"depth":1605,"text":20743,"children":21834},[21835],{"id":20865,"depth":1610,"text":20865},{"id":20876,"depth":1605,"text":20876,"children":21837},[21838,21839],{"id":20888,"depth":1610,"text":20888},{"id":20920,"depth":1610,"text":20921},{"id":20940,"depth":1605,"text":20940,"children":21841},[21842,21843],{"id":20943,"depth":1610,"text":20944},{"id":20991,"depth":1610,"text":20992},{"id":17031,"depth":1605,"text":17032},{"id":21061,"depth":1605,"text":21062},{"id":17100,"depth":1605,"text":17101},{"id":21133,"depth":1605,"text":21134,"children":21848},[21849],{"id":21249,"depth":1610,"text":21250},{"id":21261,"depth":1605,"text":21262,"children":21851},[21852,21853,21854],{"id":21265,"depth":1610,"text":21266},{"id":21339,"depth":1610,"text":21340},{"id":21423,"depth":1610,"text":21424},{"id":21507,"depth":1605,"text":21508,"children":21856},[21857],{"id":21620,"depth":1610,"text":21621},{"id":21632,"depth":1605,"text":21633,"children":21859},[21860,21861],{"id":21648,"depth":1610,"text":21649},{"id":21681,"depth":1610,"text":21682},{"id":21701,"depth":1605,"text":21702,"children":21863},[21864,21865],{"id":21705,"depth":1610,"text":21706},{"id":21753,"depth":1610,"text":21754},{"id":17920,"depth":1605,"text":17921},"Tier1 VCがチップ単体ではなく「設計生産性・データ移動・実装・供給網」に分解して投資する理由を、半導体プロセス開発出身のCVCの視点で読み解く。短期・中期・長期の投資機会とリスクを体系化する。",{"date":17979,"image":21869,"alt":21870,"tags":21871,"tagsEn":21872,"published":1666},"/blogs-img/blog-semiconductor-vc-3.png","半導体VC投資の戦略フレームワーク",[17983,17984,1657,4240,1660],[17986,17987,1662,4245,1665],"/blogs/8-semiconductor-vc-series-3",{"title":20283,"description":21867},"blogs/8-semiconductor-vc-series-3","r1Kamt0c9nXMtrYyBd1fSNeMGA8aOHIWVqekFf93F18",{"id":21878,"title":21879,"body":21880,"description":23586,"extension":1651,"meta":23587,"navigation":1666,"ogImage":23589,"path":23599,"seo":23600,"stem":23601,"__hash__":23602},"content/blogs/9-claude-gemini-chatgpt-comparison.md","最新Claude・Gemini・ChatGPTを徹底比較｜VCの僕がたどり着いた使い分けの結論 | Claude vs Gemini vs ChatGPT: A VC's Guide to the Latest Models",{"type":7,"value":21881,"toc":23532},[21882,22709,22711],[10,21883,21884,21888,21890,21893,21906,21913,21916,21918,21921,21924,21928,21931,21942,21945,21949,21960,21963,21967,21970,21987,21993,21996,21998,22001,22004,22007,22212,22215,22220,22227,22230,22235,22241,22246,22255,22260,22267,22272,22279,22284,22287,22292,22299,22301,22304,22307,22311,22329,22344,22348,22373,22380,22384,22401,22404,22406,22409,22416,22420,22425,22442,22446,22451,22475,22479,22484,22505,22508,22514,22531,22538,22540,22544,22547,22550,22639,22650,22653,22664,22671,22674,22700,22703,22705],{"lang":12},[14,21885,21887],{"id":21886},"最新claudegeminichatgptを徹底比較vcの僕がたどり着いた使い分けの結論","最新Claude・Gemini・ChatGPTを徹底比較｜VCの僕がたどり着いた使い分けの結論",[205,21889],{},[32,21891,21892],{"id":21892},"概要",[19,21894,21895,21896,21902,21903,21905],{},"ここ1〜2ヶ月で3大AIラボのフラッグシップが立て続けにアップデートされました。OpenAIが2026年3月5日に",[23,21897,21898,21899,4438],{},"GPT-5.4（Pro / Thinking）",[23,21900,21901],{},"をリリースし、Anthropicは4月16日に","を投入、GoogleもGemini 3から",[23,21904,4441],{},"へと更新。この3モデルが現時点のトップ層です。",[19,21907,21908,21909,21912],{},"僕はVCとしてスタートアップの評価書を書いたり、決算短信を要約したり、コードを読んだりと1日中AIを触っているわけですが、正直「1つで全部済ませる」時代はもう終わったと思っています。",[23,21910,21911],{},"モデルごとに「尖っている部分」と「鈍い部分」がハッキリ分かれてきている","というのが2026年4月時点の実感です。",[19,21914,21915],{},"この記事では、最新3モデルを公式ベンチマークとVC実務での使用感の両面からざっくばらんに比較して、最後に僕自身の使い分けまで公開します。数字マニアックな話というよりは、「で、結局どれをどう使えばいいの？」という実務寄りの話です。",[205,21917],{},[32,21919,21920],{"id":21920},"これまでのトレンド",[19,21922,21923],{},"まずはここ2年の流れをざっくり振り返ります。",[122,21925,21927],{"id":21926},"_2024年gpt-4時代の完成とclaudeの台頭","2024年：GPT-4時代の完成とClaudeの台頭",[19,21929,21930],{},"2024年はGPT-4oが事実上のデファクトスタンダードでしたよね。マルチモーダルも綺麗にまとまっていて、「とりあえずChatGPT Plus契約しとけ」が正解だった時代です。",[19,21932,21933,21934,21937,21938,21941],{},"そこにAnthropicが",[23,21935,21936],{},"Claude 3.5 Sonnet","で殴り込んできて、特に",[23,21939,21940],{},"コーディング性能","で「あれ、これGPT-4oより明らかに上手くない？」と業界がざわつき始めます。Cursor、Cline、Aider等のAIコーディングツールが一気にClaudeを標準モデルに据え始めたのがこの頃です。",[19,21943,21944],{},"Geminiは……正直、2024年前半は微妙でした。Gemini 1.5でコンテキスト100万トークンという「量で殴る」戦略は面白かったけど、肝心の回答品質がGPT-4やClaudeに一歩及ばない印象でした。",[122,21946,21948],{"id":21947},"_2025年推論モデルreasoningの時代","2025年：推論モデル（Reasoning）の時代",[19,21950,21951,21952,21955,21956,21959],{},"2025年はとにかく「推論」がキーワードでした。OpenAIが",[23,21953,21954],{},"o1 → o3 → o4","と推論モデルを連発し、「考えてから答える」パラダイムが定着。Anthropicも",[23,21957,21958],{},"Claude 3.7 Sonnet","で「Extended Thinking」を搭載、GoogleもGemini 2.5 Proで「Deep Think」モードを用意します。",[19,21961,21962],{},"この流れで、**数学・科学・コードといった「正解のある問題」**については、どのフラッグシップモデルも博士課程レベルに到達したと言っていい状況になりました。",[122,21964,21966],{"id":21965},"_2026年マルチモーダルとエージェントの統合","2026年：マルチモーダルとエージェントの統合",[19,21968,21969],{},"そして2026年、3社の最新フラッグシップが出揃いました。",[312,21971,21972,21977,21982],{},[315,21973,21974,21976],{},[23,21975,4747],{},": GPT-5.4 Pro（2026年3月5日）",[315,21978,21979,21981],{},[23,21980,4763],{},": Claude Opus 4.7（2026年4月16日）",[315,21983,21984,21986],{},[23,21985,4780],{},": Gemini 3.1 Pro（Deep Think搭載）",[19,21988,21989,21992],{},[23,21990,21991],{},"「推論するかどうかをユーザーが切り替える」時代は終わり、モデルが勝手に判断する","というのが大きな流れです。ユーザーは質問するだけ。簡単な質問は速く、難しい質問はじっくり——これが標準になりました。",[19,21994,21995],{},"もう一つの潮流は**「コンピュータ操作」の実用化**。GPT-5.4はOSWorldベンチで75%を記録し、人間の専門家（72.4%）を超えました。「AIがブラウザやPCを動かして仕事をする」段階に突入した年、というのが2026年です。",[205,21997],{},[32,21999,22000],{"id":22000},"最新モデルの性能比較",[19,22002,22003],{},"それでは本題。2026年4月時点の公開ベンチマークと実運用の感触を整理します。",[122,22005,22006],{"id":22006},"ざっくり比較表",[54,22008,22009,22022],{},[57,22010,22011],{},[60,22012,22013,22015,22017,22019],{},[63,22014,65],{},[63,22016,4438],{},[63,22018,4441],{},[63,22020,22021],{},"GPT-5.4 Pro",[70,22023,22024,22038,22052,22066,22081,22093,22107,22121,22135,22151,22167,22182,22198],{},[60,22025,22026,22029,22032,22035],{},[75,22027,22028],{},"リリース",[75,22030,22031],{},"2026年4月16日",[75,22033,22034],{},"2026年前半（3.1世代）",[75,22036,22037],{},"2026年3月5日",[60,22039,22040,22042,22047,22049],{},[75,22041,4448],{},[75,22043,22044],{},[23,22045,22046],{},"87.6%",[75,22048,4459],{},[75,22050,22051],{},"約82%",[60,22053,22054,22056,22060,22063],{},[75,22055,4464],{},[75,22057,22058],{},[23,22059,4473],{},[75,22061,22062],{},"約53%",[75,22064,22065],{},"約55〜58%",[60,22067,22068,22071,22074,22079],{},[75,22069,22070],{},"GPQA Diamond（科学推論）",[75,22072,22073],{},"高水準",[75,22075,22076],{},[23,22077,22078],{},"94.3%",[75,22080,22073],{},[60,22082,22083,22086,22088,22091],{},[75,22084,22085],{},"Humanity's Last Exam",[75,22087,250],{},[75,22089,22090],{},"44.4%",[75,22092,250],{},[60,22094,22095,22098,22100,22102],{},[75,22096,22097],{},"ARC-AGI-2（抽象推論）",[75,22099,250],{},[75,22101,4404],{},[75,22103,22104],{},[23,22105,22106],{},"83.3%",[60,22108,22109,22112,22114,22116],{},[75,22110,22111],{},"BrowseComp（Web操作）",[75,22113,250],{},[75,22115,250],{},[75,22117,22118],{},[23,22119,22120],{},"89.3%",[60,22122,22123,22126,22128,22130],{},[75,22124,22125],{},"OSWorld（PC操作）",[75,22127,250],{},[75,22129,250],{},[75,22131,22132],{},[23,22133,22134],{},"75%",[60,22136,22137,22140,22143,22148],{},[75,22138,22139],{},"MMMU-Pro（マルチモーダル）",[75,22141,22142],{},"◯",[75,22144,22145],{},[23,22146,22147],{},"80.5%",[75,22149,22150],{},"◎",[60,22152,22153,22156,22159,22164],{},[75,22154,22155],{},"コンテキスト長",[75,22157,22158],{},"200K（1M版あり）",[75,22160,22161],{},[23,22162,22163],{},"1M",[75,22165,22166],{},"1M〜1.1M",[60,22168,22169,22172,22175,22180],{},[75,22170,22171],{},"入力価格（$/1M tokens）",[75,22173,22174],{},"$5",[75,22176,22177],{},[23,22178,22179],{},"最安水準（約1/5）",[75,22181,4633],{},[60,22183,22184,22187,22190,22195],{},[75,22185,22186],{},"出力価格（$/1M tokens）",[75,22188,22189],{},"$25",[75,22191,22192],{},[23,22193,22194],{},"最安水準",[75,22196,22197],{},"$15",[60,22199,22200,22203,22206,22210],{},[75,22201,22202],{},"動画・音声ネイティブ",[75,22204,22205],{},"△",[75,22207,22208],{},[23,22209,22150],{},[75,22211,22142],{},[122,22213,22214],{"id":22214},"各軸の詳細",[19,22216,22217],{},[23,22218,22219],{},"① コーディング：Claudeが王座奪還",[19,22221,22222,22223,22226],{},"Opus 4.6→4.7で**SWE-bench Verifiedが80.8%→87.6%",[23,22224,22225],{},"と7ポイント近くジャンプし、Gemini 3.1 Pro（80.6%）とGPT-5.4（約82%）を抜き返しました。難度の高い","SWE-bench Proでは64.3%**と、GPT-5.4（約57.7%）・Gemini（約54.2%）を10ポイント以上引き離しています。",[19,22228,22229],{},"CursorBenchも58%→70%と大幅改善。実務で「このリポジトリ全体を読んで設計の問題点を指摘して」系のタスクをやると、体感でもClaudeが一段抜けているのが分かります。",[19,22231,22232],{},[23,22233,22234],{},"② 科学推論：Geminiが頭ひとつリード",[19,22236,22237,22240],{},[23,22238,22239],{},"GPQA Diamond 94.3%","、**MMMLU 92.6%**など、研究レベルの知識・推論ベンチでGemini 3.1 Proが最強クラス。Google Researchの基礎研究基盤が効いている印象で、数式や論文読解の厳密性で安定感があります。",[19,22242,22243],{},[23,22244,22245],{},"③ エージェント・PC操作：GPT-5.4が独走",[19,22247,22248,7758,22251,22254],{},[23,22249,22250],{},"BrowseComp 89.3%",[23,22252,22253],{},"OSWorld 75%","（人間エキスパート72.4%を超える）と、「ブラウザやPCを自動操作する」系のベンチでGPT-5.4 Proが独走しています。ARC-AGI-2でも**83.3%**と、抽象推論のトップです。",[19,22256,22257],{},[23,22258,22259],{},"④ 長文コンテキスト：Geminiの1Mが実用性で先行",[19,22261,22262,22263,22266],{},"Claudeも1Mコンテキスト版を提供していますが、",[23,22264,22265],{},"ネイティブで1Mを全モデルに提供しているGemini","が量の面で有利。PDF数十本を一気読みさせるような用途では現時点でGeminiが最もコストパフォーマンスが良いです。",[19,22268,22269],{},[23,22270,22271],{},"⑤ マルチモーダル（動画・音声）：Geminiが設計思想で勝つ",[19,22273,22274,22275,22278],{},"Geminiは最初からテキスト・画像・動画・音声・コードを統合設計している強みが効いていて、YouTube動画を直接渡して要約させたり、長尺の会議録音を議事録化するといった用途では",[23,22276,22277],{},"Gemini一択","の状況です。",[19,22280,22281],{},[23,22282,22283],{},"⑥ 創作・文章表現：Claudeが依然として好まれる",[19,22285,22286],{},"ブラインドの人間評価で、創作系タスクはClaudeが47%、GPT-5.4が29%、Geminiが24%で選ばれるという結果が出ています。投資メモや編集系の下書きを任せるならClaudeが無難、というのが実感と一致します。",[19,22288,22289],{},[23,22290,22291],{},"⑦ コスト：Geminiが圧倒的",[19,22293,22294,22295,22298],{},"Gemini 3.1 Proは",[23,22296,22297],{},"Claude Opus 4.7のおよそ1/5、GPT-5.4の約1/4","の価格帯。バッチ処理や大量データを流すワークロードではGeminiが経済合理性で圧勝します。",[205,22300],{},[32,22302,22303],{"id":22303},"各サービスの特徴",[19,22305,22306],{},"性能の話だけだと見えてこない部分として、「サービスとしての使いやすさ・エコシステム」の違いが実はかなり大きいです。",[122,22308,22310],{"id":22309},"claude-opus-47anthropic","Claude Opus 4.7（Anthropic）",[312,22312,22313,22318,22324],{},[315,22314,22315,22317],{},[23,22316,13766],{},": コーディング・エージェント特化、創作力、Artifacts、MCP（Model Context Protocol）によるエコシステム拡張、Finance Agent評価トップ水準",[315,22319,22320,22323],{},[23,22321,22322],{},"弱み",": 画像生成なし、動画・音声理解はGeminiに劣る、API価格が最も高い（$5/$25）",[315,22325,22326,22328],{},[23,22327,403],{},": Claude.ai（Web）、Claude Code（CLI）、API、Amazon Bedrock、Google Cloud Vertex AI",[19,22330,22331,22332,22335,22336,22339,22340,22343],{},"Anthropicは明確に「",[23,22333,22334],{},"開発者・エンタープライズ特化","」に振り切っているのが特徴です。",[23,22337,22338],{},"Opus 4.7は「前モデルが緩く解釈していた指示を文字通り実行する」ように調整","されていて、プロンプトの設計力が以前より問われるようになりました。Claude Codeは2025年に登場して以降、",[23,22341,22342],{},"今では僕のメインIDE","になっています。",[122,22345,22347],{"id":22346},"gemini-31-progoogle","Gemini 3.1 Pro（Google）",[312,22349,22350,22359,22364],{},[315,22351,22352,22354,22355,22358],{},[23,22353,13766],{},": Google検索統合、Workspace連携（Gmail、Drive、Docs）、ネイティブマルチモーダル（動画・音声込み）、1Mコンテキスト標準、",[23,22356,22357],{},"圧倒的な低価格","、科学・数式ベンチの強さ",[315,22360,22361,22363],{},[23,22362,22322],{},": コーディングでの細かい詰めが依然として劣る、ブランド的に「とりあえず試す」のハードルが高い",[315,22365,22366,22368,22369,22372],{},[23,22367,403],{},": gemini.google.com、AI Studio、Vertex AI、Googleアプリ内統合、",[23,22370,22371],{},"Gemini 3.1 Deep Think","（Google AI Ultra加入者向け）",[19,22374,22375,22376,22379],{},"Geminiの真骨頂は",[23,22377,22378],{},"Googleエコシステムとの統合","。会社メールを全部読んで要約してくれる、Driveのドキュメントを横断検索できる、といった「日常業務」での強さが圧倒的です。Deep Thinkは科学・エンジニアリングの高難度問題向けに別途用意されている専用推論モードで、高度な分析が必要な場面で切り替えて使えます。",[122,22381,22383],{"id":22382},"gpt-54-proopenai","GPT-5.4 Pro（OpenAI）",[312,22385,22386,22391,22396],{},[315,22387,22388,22390],{},[23,22389,13766],{},": エージェント（BrowseComp・OSWorld圧倒的）、抽象推論（ARC-AGI-2 83.3%）、プラグイン・GPTsエコシステム、DALL-E/Sora統合、Advanced Voice Mode、Canvas、知名度・ユーザー数",[315,22392,22393,22395],{},[23,22394,22322],{},": First-token時間が長い（推論モデルなので）、Knowledge cutoffが2025年8月",[315,22397,22398,22400],{},[23,22399,403],{},": ChatGPT（Web/アプリ）、API、Microsoft Copilot統合、Pro・Enterpriseプラン限定",[19,22402,22403],{},"GPT-5.4 Proは**「コンピュータ操作エージェント」としての完成度が頭ひとつ抜けています**。OSWorldで人間エキスパートを超えたのは象徴的で、「ブラウザで調査→スプレッドシート作成→Slackで送信」みたいな複数ステップ業務の自動化はGPT-5.4 Proが最強候補です。音声対話の自然さも依然として業界トップクラス。",[205,22405],{},[32,22407,22408],{"id":22408},"記事作成者の使い分け",[19,22410,22411,22412,22415],{},"最後に、VCとしての僕自身のリアルな使い分けを公開します。",[23,22413,22414],{},"3つ全部有料プラン契約しています","（経費です）。",[122,22417,22419],{"id":22418},"claude-opus-47メイン50","Claude Opus 4.7（メイン50%）",[19,22421,22422,22424],{},[23,22423,7810],{},": コーディング、決算書読解、投資メモ執筆、契約書レビュー、創作・編集系",[312,22426,22427,22430,22433,22436,22439],{},[315,22428,22429],{},"この記事もClaudeで下書きを作っています",[315,22431,22432],{},"Claude Code経由でこのブログのNuxt 3実装をメンテナンス",[315,22434,22435],{},"投資先のデューデリ資料（PDF100本超）を1Mコンテキスト版で一気読み",[315,22437,22438],{},"SWE-bench Pro 64.3%の地力は、大規模コードベースの読解でそのまま体感できる",[315,22440,22441],{},"「指示を文字通り実行する」方向の調整が自分のワークフローと相性が良い",[122,22443,22445],{"id":22444},"gemini-31-proサブ30","Gemini 3.1 Pro（サブ30%）",[19,22447,22448,22450],{},[23,22449,7810],{},": リサーチ、Google Workspace連携、動画・音声処理、多言語翻訳、大量バッチ処理",[312,22452,22453,22456,22459,22462,22468],{},[315,22454,22455],{},"カンファレンスの録画動画を丸ごと渡して要点抽出",[315,22457,22458],{},"会社メールに届く大量のピッチメール（英語・日本語・中国語混在）の仕分け",[315,22460,22461],{},"Google検索グラウンディングによる「最新情報リサーチ」",[315,22463,22464,22465],{},"コスパが異常に良いので、",[23,22466,22467],{},"量で殴るタスクはGeminiで回す",[315,22469,22470,22471,22474],{},"科学系スタートアップのDDでは",[23,22472,22473],{},"Deep Think","に切り替え",[122,22476,22478],{"id":22477},"gpt-54-proサブ20","GPT-5.4 Pro（サブ20%）",[19,22480,22481,22483],{},[23,22482,7810],{},": Web自動操作、画像生成、音声対話、ブレスト、GPTs定型業務",[312,22485,22486,22489,22496,22499,22502],{},[315,22487,22488],{},"BrowseCompの強さが効く**「10社のIRページを巡回して決算データを抽出」**系タスク",[315,22490,22491,22492,22495],{},"OSWorldが強いので",[23,22493,22494],{},"ExcelやGoogle Sheetsへの自動記入","も任せられる",[315,22497,22498],{},"投資先向けプレゼン用の画像生成（DALL-E/Sora統合）",[315,22500,22501],{},"散歩中の音声対話でアイデアの壁打ち",[315,22503,22504],{},"「ChatGPTで試したいから」という依頼者が多いので検証用にも必須",[122,22506,22507],{"id":22507},"僕の結論",[19,22509,22510,22513],{},[23,22511,22512],{},"「モデル戦争」はもう終わっていて、今は「得意分野の棲み分け」フェーズ","だと思います。2026年4月の状況を要約するとこうです。",[312,22515,22516,22521,22526],{},[315,22517,22518],{},[23,22519,22520],{},"コーディング・長文読解・創作 → Claude Opus 4.7",[315,22522,22523],{},[23,22524,22525],{},"科学・マルチモーダル・大量処理・コスト → Gemini 3.1 Pro",[315,22527,22528],{},[23,22529,22530],{},"Web/PC操作・抽象推論・エコシステム → GPT-5.4 Pro",[19,22532,22533,22534,22537],{},"VCとして1つアドバイスするなら、",[23,22535,22536],{},"「月20〜30ドル×3社」をケチらないこと","。ワークロードごとに最適モデルが違うので、使い分けた方が圧倒的に生産性が上がります。",[205,22539],{},[32,22541,22543],{"id":22542},"投資シナリオaiラボ三つ巴の今後","投資シナリオ：AIラボ三つ巴の今後",[19,22545,22546],{},"ここからが本記事の投資パートです。「どのモデルが勝つか」ではなく、**「3社の優劣の振れ方が、関連銘柄にどう波及するか」**が投資家にとっての論点です。",[122,22548,22549],{"id":22549},"シナリオ別の関連銘柄インパクト",[54,22551,22552,22572],{},[57,22553,22554],{},[60,22555,22556,22558,22560,22563,22566,22569],{},[63,22557,601],{},[63,22559,604],{},[63,22561,22562],{},"OpenAI/MSFT",[63,22564,22565],{},"Anthropic/AMZN-GOOGL",[63,22567,22568],{},"Google/GOOGL",[63,22570,22571],{},"NVDA",[70,22573,22574,22590,22607,22623],{},[60,22575,22576,22579,22581,22583,22585,22588],{},[75,22577,22578],{},"A. 三つ巴継続（ベース）",[75,22580,624],{},[75,22582,638],{},[75,22584,638],{},[75,22586,22587],{},"やや上方",[75,22589,627],{},[60,22591,22592,22595,22598,22600,22602,22604],{},[75,22593,22594],{},"B. Anthropic先行（Claudeコーディング独走）",[75,22596,22597],{},"25%",[75,22599,638],{},[75,22601,657],{},[75,22603,638],{},[75,22605,22606],{},"コア維持",[60,22608,22609,22612,22614,22617,22619,22621],{},[75,22610,22611],{},"C. Google逆転（Gemini統合勝ち）",[75,22613,3434],{},[75,22615,22616],{},"やや下方",[75,22618,638],{},[75,22620,657],{},[75,22622,22606],{},[60,22624,22625,22628,22631,22633,22635,22637],{},[75,22626,22627],{},"D. OpenAIエージェント独走",[75,22629,22630],{},"10%",[75,22632,657],{},[75,22634,638],{},[75,22636,638],{},[75,22638,22606],{},[19,22640,22641,22642,22645,22646,22649],{},"私の現状の見立ては",[23,22643,22644],{},"シナリオA","で、向こう12ヶ月は3社が交互にリードを奪い合う展開が続くと見ています。CVCの感覚で言うと、",[23,22647,22648],{},"こういう時期にいちばん効くのは「三社いずれが勝っても恩恵を受ける銘柄」","——具体的にはNVDA（GPU）、ASML（露光装置）、TSMC（ファブ）、ARM（IP）です。",[122,22651,22652],{"id":22652},"投資判断と含意",[19,22654,22655,22656,22659,22660,22663],{},"スタートアップ投資の観点では——",[23,22657,22658],{},"アプリケーション層で「特定モデルに依存したビジネス」を作るのは危険","。2026年だけでもGPT-5.4→Opus 4.7と",[23,22661,22662],{},"2ヶ月で王座が入れ替わっている","わけで、複数モデルを抽象化して切り替えられる設計のスタートアップが長期では生き残る、というのが私の仮説です。",[19,22665,22666,22667,22670],{},"公開市場の観点では、",[23,22668,22669],{},"3社のラボ間競争が激化するほどNVDA・ASML・TSMCのGPU/ファブ系へのキャペックスは増える","——つまり「誰が勝つか」を当てにいくよりも、「全員が走り続ける限り恩恵を受けるレイヤー」に寄せる方が、リスク調整後リターンは高くなる構造です。",[122,22672,22673],{"id":22673},"ウォッチポイント",[312,22675,22676,22682,22688,22694],{},[315,22677,22678,22681],{},[23,22679,22680],{},"Claude/GPT/Geminiの次の主要バージョン"," — 王座交代が継続するか、誰かが固定的にリードを取るか",[315,22683,22684,22687],{},[23,22685,22686],{},"NVDA決算のデータセンター売上"," — モデル競争激化局面ではキャペックス連動性が高い",[315,22689,22690,22693],{},[23,22691,22692],{},"OpenAI/Anthropicの一次資金調達バリュエーション"," — シードからの倍率が市場心理の温度計",[315,22695,22696,22699],{},[23,22697,22698],{},"規制動向（EU AI Act、米AI Bill of Rights実装）"," — 大規模モデル開発のコスト構造を変える",[19,22701,22702],{},"それでは、今回はここまで。皆さんの使い分けも教えてもらえると嬉しいです。",[205,22704],{},[19,22706,22707],{},[802,22708,804],{},[205,22710],{},[10,22712,22713,22717,22719,22723,22736,22743,22746,22748,22752,22755,22759,22762,22768,22771,22775,22781,22788,22792,22795,22812,22819,22826,22828,22832,22835,22839,23028,23032,23037,23048,23051,23056,23064,23069,23080,23085,23092,23097,23103,23108,23111,23116,23123,23125,23129,23132,23136,23155,23170,23174,23198,23205,23209,23226,23233,23235,23239,23242,23246,23252,23269,23273,23278,23300,23304,23309,23332,23336,23342,23359,23366,23368,23372,23378,23382,23465,23475,23479,23486,23493,23497,23523,23526,23528],{"lang":809},[14,22714,22716],{"id":22715},"claude-vs-gemini-vs-chatgpt-a-vcs-guide-to-the-latest-models","Claude vs Gemini vs ChatGPT: A VC's Guide to the Latest Models",[205,22718],{},[32,22720,22722],{"id":22721},"overview","Overview",[19,22724,22725,22726,22729,22730,22732,22733,22735],{},"In the past two months, all three major AI labs have refreshed their flagships in rapid succession. OpenAI released ",[23,22727,22728],{},"GPT-5.4 (Pro / Thinking)"," on March 5, 2026. Anthropic shipped ",[23,22731,4438],{}," on April 16. Google updated Gemini 3 to ",[23,22734,4441],{},". These three are the current top tier.",[19,22737,22738,22739,22742],{},"As a VC who spends all day writing investment memos, summarizing earnings reports, and reading code with AI assistance, my honest take is: ",[23,22740,22741],{},"the era of \"just use one model for everything\" is over",". Each flagship has sharpened its edges, and the differences matter in practice.",[19,22744,22745],{},"This article compares the three flagships as of April 2026, blending public benchmarks with real working-VC usage, and ends with my personal usage split.",[205,22747],{},[32,22749,22751],{"id":22750},"recent-trends","Recent Trends",[19,22753,22754],{},"A quick look back at the last two years:",[122,22756,22758],{"id":22757},"_2024-gpt-4-era-peaks-claude-emerges","2024: GPT-4 Era Peaks, Claude Emerges",[19,22760,22761],{},"GPT-4o was effectively the default in 2024. Multimodal was polished; \"just subscribe to ChatGPT Plus\" was the right answer.",[19,22763,22764,22765,22767],{},"Then Anthropic's ",[23,22766,21936],{}," arrived with coding quality that clearly exceeded GPT-4o. Cursor, Cline, Aider — the AI coding tool ecosystem standardized on Claude almost overnight.",[19,22769,22770],{},"Gemini, frankly, was underwhelming in early 2024. Gemini 1.5's 1M-token context was interesting, but response quality lagged GPT-4 and Claude.",[122,22772,22774],{"id":22773},"_2025-the-reasoning-era","2025: The Reasoning Era",[19,22776,22777,22778,22780],{},"Reasoning was the 2025 theme. OpenAI shipped ",[23,22779,21954],{}," in rapid succession. Anthropic added \"Extended Thinking\" to Claude 3.7 Sonnet. Google countered with Gemini 2.5 Pro's \"Deep Think.\"",[19,22782,22783,22784,22787],{},"For ",[23,22785,22786],{},"math, science, and coding — problems with correct answers"," — every flagship reached PhD-level competence.",[122,22789,22791],{"id":22790},"_2026-multimodal-agent-integration","2026: Multimodal + Agent Integration",[19,22793,22794],{},"Now the current lineup:",[312,22796,22797,22802,22807],{},[315,22798,22799,22801],{},[23,22800,4747],{},": GPT-5.4 Pro (March 5, 2026)",[315,22803,22804,22806],{},[23,22805,4763],{},": Claude Opus 4.7 (April 16, 2026)",[315,22808,22809,22811],{},[23,22810,4780],{},": Gemini 3.1 Pro (with Deep Think)",[19,22813,22814,22815,22818],{},"The major shift: ",[23,22816,22817],{},"users no longer toggle \"reasoning mode\" — the model decides",". Simple questions are fast, hard ones take longer. That's the new normal.",[19,22820,22821,22822,22825],{},"Another trend: ",[23,22823,22824],{},"computer-use agents are now practical",". GPT-5.4 scored 75% on OSWorld, beating human experts (72.4%). 2026 is the year AI actually started driving browsers and PCs.",[205,22827],{},[32,22829,22831],{"id":22830},"performance-comparison","Performance Comparison",[19,22833,22834],{},"Key benchmarks and real-world feel as of April 2026:",[122,22836,22838],{"id":22837},"quick-comparison-table","Quick Comparison Table",[54,22840,22841,22853],{},[57,22842,22843],{},[60,22844,22845,22847,22849,22851],{},[63,22846,8964],{},[63,22848,4438],{},[63,22850,4441],{},[63,22852,22021],{},[70,22854,22855,22869,22882,22896,22910,22920,22933,22946,22959,22972,22987,23001,23015],{},[60,22856,22857,22860,22863,22866],{},[75,22858,22859],{},"Release date",[75,22861,22862],{},"April 16, 2026",[75,22864,22865],{},"Early 2026 (3.1 gen)",[75,22867,22868],{},"March 5, 2026",[60,22870,22871,22873,22877,22879],{},[75,22872,4448],{},[75,22874,22875],{},[23,22876,22046],{},[75,22878,4459],{},[75,22880,22881],{},"~82%",[60,22883,22884,22886,22890,22893],{},[75,22885,4464],{},[75,22887,22888],{},[23,22889,4473],{},[75,22891,22892],{},"~53%",[75,22894,22895],{},"~55–58%",[60,22897,22898,22901,22904,22908],{},[75,22899,22900],{},"GPQA Diamond (science)",[75,22902,22903],{},"high",[75,22905,22906],{},[23,22907,22078],{},[75,22909,22903],{},[60,22911,22912,22914,22916,22918],{},[75,22913,22085],{},[75,22915,250],{},[75,22917,22090],{},[75,22919,250],{},[60,22921,22922,22925,22927,22929],{},[75,22923,22924],{},"ARC-AGI-2 (abstract reasoning)",[75,22926,250],{},[75,22928,4404],{},[75,22930,22931],{},[23,22932,22106],{},[60,22934,22935,22938,22940,22942],{},[75,22936,22937],{},"BrowseComp (web ops)",[75,22939,250],{},[75,22941,250],{},[75,22943,22944],{},[23,22945,22120],{},[60,22947,22948,22951,22953,22955],{},[75,22949,22950],{},"OSWorld (PC ops)",[75,22952,250],{},[75,22954,250],{},[75,22956,22957],{},[23,22958,22134],{},[60,22960,22961,22964,22966,22970],{},[75,22962,22963],{},"MMMU-Pro (multimodal)",[75,22965,22142],{},[75,22967,22968],{},[23,22969,22147],{},[75,22971,22150],{},[60,22973,22974,22977,22980,22984],{},[75,22975,22976],{},"Context length",[75,22978,22979],{},"200K (1M variant)",[75,22981,22982],{},[23,22983,22163],{},[75,22985,22986],{},"1M–1.1M",[60,22988,22989,22992,22994,22999],{},[75,22990,22991],{},"Input price ($/1M tokens)",[75,22993,22174],{},[75,22995,22996],{},[23,22997,22998],{},"~1/5 of Claude",[75,23000,4633],{},[60,23002,23003,23006,23008,23013],{},[75,23004,23005],{},"Output price ($/1M tokens)",[75,23007,22189],{},[75,23009,23010],{},[23,23011,23012],{},"lowest tier",[75,23014,22197],{},[60,23016,23017,23020,23022,23026],{},[75,23018,23019],{},"Native video/audio",[75,23021,22205],{},[75,23023,23024],{},[23,23025,22150],{},[75,23027,22142],{},[122,23029,23031],{"id":23030},"detail-per-axis","Detail Per Axis",[19,23033,23034],{},[23,23035,23036],{},"① Coding: Claude reclaims the crown",[19,23038,23039,23040,23043,23044,23047],{},"Opus 4.6 → 4.7 jumped ",[23,23041,23042],{},"SWE-bench Verified from 80.8% to 87.6%",", a ~7-point gain that leapfrogs Gemini 3.1 Pro (80.6%) and GPT-5.4 (~82%). On the harder ",[23,23045,23046],{},"SWE-bench Pro, Opus 4.7 scores 64.3%",", beating GPT-5.4 (~57.7%) and Gemini (~54.2%) by 10+ points.",[19,23049,23050],{},"CursorBench jumped 58% → 70%. In practice, \"read this entire repo and point out the design flaws\"-style prompts clearly land better on Claude.",[19,23052,23053],{},[23,23054,23055],{},"② Science Reasoning: Gemini leads",[19,23057,23058,10338,23060,23063],{},[23,23059,22239],{},[23,23061,23062],{},"MMMLU 92.6%"," put Gemini 3.1 Pro at the top of research-level knowledge and reasoning benchmarks. Google Research's foundational work shows up in rigor around symbolic math and paper comprehension.",[19,23065,23066],{},[23,23067,23068],{},"③ Agentic / PC Control: GPT-5.4 dominates",[19,23070,23071,10338,23073,23075,23076,23079],{},[23,23072,22250],{},[23,23074,22253],{}," (beats the 72.4% human expert baseline) make GPT-5.4 Pro the clear winner for \"drive a browser and a PC\" tasks. It also tops ",[23,23077,23078],{},"ARC-AGI-2 at 83.3%"," for abstract reasoning.",[19,23081,23082],{},[23,23083,23084],{},"④ Long Context: Gemini's 1M wins on practicality",[19,23086,23087,23088,23091],{},"Claude offers a 1M context variant, but ",[23,23089,23090],{},"Gemini ships 1M as standard",". For \"dump 50 PDFs in one context\" workloads, Gemini is the most cost-effective today.",[19,23093,23094],{},[23,23095,23096],{},"⑤ Multimodal (video/audio): Gemini by design",[19,23098,23099,23100,849],{},"Gemini was architected multimodal from day one — text, image, video, audio, code all unified. For summarizing a YouTube URL or turning long meeting recordings into minutes, ",[23,23101,23102],{},"Gemini is essentially the only choice",[19,23104,23105],{},[23,23106,23107],{},"⑥ Creative Writing: Claude still preferred",[19,23109,23110],{},"In blind human evaluations, Claude is picked 47% of the time vs. 29% for GPT-5.4 and 24% for Gemini. For investment memos and editorial drafts, Claude remains the safe choice.",[19,23112,23113],{},[23,23114,23115],{},"⑦ Cost: Gemini crushes it",[19,23117,23118,23119,23122],{},"Gemini 3.1 Pro is roughly ",[23,23120,23121],{},"1/5 of Claude Opus 4.7 and 1/4 of GPT-5.4",". For batch processing and high-volume workloads, Gemini wins on pure economics.",[205,23124],{},[32,23126,23128],{"id":23127},"service-characteristics","Service Characteristics",[19,23130,23131],{},"Performance alone misses the \"service + ecosystem\" dimension.",[122,23133,23135],{"id":23134},"claude-opus-47-anthropic","Claude Opus 4.7 (Anthropic)",[312,23137,23138,23144,23150],{},[315,23139,23140,23143],{},[23,23141,23142],{},"Strengths",": Coding/agentic focus, creative writing, Artifacts, MCP ecosystem, top-tier finance-agent performance",[315,23145,23146,23149],{},[23,23147,23148],{},"Weaknesses",": No image generation, weaker video/audio than Gemini, highest API price ($5/$25)",[315,23151,23152,23154],{},[23,23153,1196],{},": Claude.ai, Claude Code (CLI), API, AWS Bedrock, Google Vertex AI",[19,23156,23157,23158,23161,23162,23165,23166,23169],{},"Anthropic is clearly ",[23,23159,23160],{},"developer-and-enterprise-first",". Opus 4.7 is specifically tuned to ",[23,23163,23164],{},"\"execute instructions literally where prior models interpreted them loosely\""," — prompt craft now matters more. Claude Code has become ",[23,23167,23168],{},"my primary IDE"," since its 2025 launch.",[122,23171,23173],{"id":23172},"gemini-31-pro-google","Gemini 3.1 Pro (Google)",[312,23175,23176,23185,23190],{},[315,23177,23178,23180,23181,23184],{},[23,23179,23142],{},": Google Search integration, Workspace (Gmail/Drive/Docs), native multimodal including video/audio, 1M context as standard, ",[23,23182,23183],{},"aggressively low pricing",", strong science/math benchmarks",[315,23186,23187,23189],{},[23,23188,23148],{},": Coding polish still trails, less \"first try\" mindshare",[315,23191,23192,23194,23195,23197],{},[23,23193,1196],{},": gemini.google.com, AI Studio, Vertex AI, Google app integration, ",[23,23196,22371],{}," (Google AI Ultra subscribers)",[19,23199,23200,23201,23204],{},"Gemini's real advantage is ",[23,23202,23203],{},"Google ecosystem integration",". Reading an entire corporate email inbox, cross-searching Drive docs — it dominates \"daily work\" surface area. Deep Think is a separate high-reasoning mode for science and engineering problems, switchable when you need it.",[122,23206,23208],{"id":23207},"gpt-54-pro-openai","GPT-5.4 Pro (OpenAI)",[312,23210,23211,23216,23221],{},[315,23212,23213,23215],{},[23,23214,23142],{},": Agentic capability (BrowseComp/OSWorld dominant), abstract reasoning (ARC-AGI-2 83.3%), plugins/GPTs ecosystem, DALL-E/Sora integration, Advanced Voice Mode, Canvas, brand scale",[315,23217,23218,23220],{},[23,23219,23148],{},": Long time-to-first-token (reasoning model), knowledge cutoff August 2025",[315,23222,23223,23225],{},[23,23224,1196],{},": ChatGPT (web/app), API, Microsoft Copilot, Pro/Enterprise plans only",[19,23227,23228,23229,23232],{},"GPT-5.4 Pro is a class apart as a ",[23,23230,23231],{},"computer-use agent",". Beating human experts on OSWorld is symbolic — for multi-step workflows like \"research in browser → update spreadsheet → post to Slack,\" GPT-5.4 Pro is the top candidate. Voice mode remains best-in-class.",[205,23234],{},[32,23236,23238],{"id":23237},"my-personal-usage-split","My Personal Usage Split",[19,23240,23241],{},"Here's how I actually divide my time across all three (yes, I pay for all three — business expense).",[122,23243,23245],{"id":23244},"claude-opus-47-primary-50","Claude Opus 4.7 (Primary, ~50%)",[19,23247,23248,23251],{},[23,23249,23250],{},"Use cases",": Coding, earnings-report analysis, investment memos, contract review, creative/editorial drafts",[312,23253,23254,23257,23260,23263,23266],{},[315,23255,23256],{},"Drafted this article in Claude",[315,23258,23259],{},"Maintain this Nuxt 3 blog via Claude Code",[315,23261,23262],{},"Load 100+ PDFs in the 1M-context variant for due diligence",[315,23264,23265],{},"SWE-bench Pro 64.3% translates directly into superior large-codebase comprehension",[315,23267,23268],{},"\"Execute instructions literally\" tuning fits my workflow",[122,23270,23272],{"id":23271},"gemini-31-pro-secondary-30","Gemini 3.1 Pro (Secondary, ~30%)",[19,23274,23275,23277],{},[23,23276,23250],{},": Research, Google Workspace integration, video/audio, multilingual translation, high-volume batch",[312,23279,23280,23283,23286,23289,23295],{},[315,23281,23282],{},"Full conference recordings → key-point extraction",[315,23284,23285],{},"Sorting mixed EN/JA/ZH pitch emails from my corporate inbox",[315,23287,23288],{},"Latest-info research with Google Search grounding",[315,23290,23291,23292],{},"Absurdly cheap — my ",[23,23293,23294],{},"volume workhorse",[315,23296,23297,23299],{},[23,23298,22473],{}," for DD on science/deep-tech startups",[122,23301,23303],{"id":23302},"gpt-54-pro-secondary-20","GPT-5.4 Pro (Secondary, ~20%)",[19,23305,23306,23308],{},[23,23307,23250],{},": Web automation, image generation, voice, brainstorming, custom GPTs",[312,23310,23311,23317,23323,23326,23329],{},[315,23312,23313,23314],{},"BrowseComp strength shines on ",[23,23315,23316],{},"\"crawl 10 IR pages and extract earnings data\"",[315,23318,23319,23320],{},"OSWorld strength means I can trust it to ",[23,23321,23322],{},"fill Excel/Sheets automatically",[315,23324,23325],{},"Visuals for portfolio-company decks (DALL-E/Sora)",[315,23327,23328],{},"Voice brainstorming while walking",[315,23330,23331],{},"Founders still say \"I tried it on ChatGPT…\" — I need it for verification",[122,23333,23335],{"id":23334},"my-conclusion","My Conclusion",[19,23337,23338,23341],{},[23,23339,23340],{},"The \"model war\" is over. We're in the \"specialization\" phase."," April 2026 boils down to this:",[312,23343,23344,23349,23354],{},[315,23345,23346],{},[23,23347,23348],{},"Coding / long context / creative → Claude Opus 4.7",[315,23350,23351],{},[23,23352,23353],{},"Science / multimodal / volume / cost → Gemini 3.1 Pro",[315,23355,23356],{},[23,23357,23358],{},"Web+PC automation / abstract reasoning / ecosystem → GPT-5.4 Pro",[19,23360,23361,23362,23365],{},"One piece of VC advice: ",[23,23363,23364],{},"don't skimp on $20–30/month × 3 subscriptions",". Workloads differ, and using them appropriately massively improves productivity.",[205,23367],{},[32,23369,23371],{"id":23370},"investment-scenario-the-three-way-lab-race","Investment Scenario: The Three-Way Lab Race",[19,23373,23374,23375,849],{},"This is where the piece earns its VC framing. The investor question isn't \"which model wins?\" — it's ",[23,23376,23377],{},"\"how does the back-and-forth between the three labs ripple through related stocks?\"",[122,23379,23381],{"id":23380},"scenario-related-name-impact","Scenario × related-name impact",[54,23383,23384,23401],{},[57,23385,23386],{},[60,23387,23388,23390,23392,23394,23397,23399],{},[63,23389,1389],{},[63,23391,1392],{},[63,23393,22562],{},[63,23395,23396],{},"Anthropic (AMZN/GOOGL bet)",[63,23398,22568],{},[63,23400,22571],{},[70,23402,23403,23419,23434,23450],{},[60,23404,23405,23408,23410,23412,23414,23417],{},[75,23406,23407],{},"A. Three-way continues (base)",[75,23409,624],{},[75,23411,1426],{},[75,23413,1426],{},[75,23415,23416],{},"Slight upside",[75,23418,1415],{},[60,23420,23421,23424,23426,23428,23430,23432],{},[75,23422,23423],{},"B. Anthropic pulls ahead (Claude coding dominance)",[75,23425,22597],{},[75,23427,1426],{},[75,23429,1444],{},[75,23431,1426],{},[75,23433,1415],{},[60,23435,23436,23439,23441,23444,23446,23448],{},[75,23437,23438],{},"C. Google reverses (Gemini ecosystem integration wins)",[75,23440,3434],{},[75,23442,23443],{},"Slight down",[75,23445,1426],{},[75,23447,1444],{},[75,23449,1415],{},[60,23451,23452,23455,23457,23459,23461,23463],{},[75,23453,23454],{},"D. OpenAI agentic dominance",[75,23456,22630],{},[75,23458,1444],{},[75,23460,1426],{},[75,23462,1426],{},[75,23464,1415],{},[19,23466,23467,23468,23470,23471,23474],{},"My current base case is ",[23,23469,18899],{},": the next ~12 months stay competitive, with leadership flipping between the three. From the CVC seat, the most useful position in this kind of regime is ",[23,23472,23473],{},"\"benefits no matter who wins\""," — concretely NVDA (GPUs), ASML (lithography), TSMC (fab), ARM (IP).",[122,23476,23478],{"id":23477},"investment-call-implications","Investment call & implications",[19,23480,23481,23482,23485],{},"For startup investors: ",[23,23483,23484],{},"building application-layer businesses that depend on a single model is dangerous",". Just in 2026, the crown flipped from GPT-5.4 to Opus 4.7 in two months. Startups that abstract over multiple models and swap them transparently will be the ones that survive the next two crown flips.",[19,23487,23488,23489,23492],{},"For public markets: ",[23,23490,23491],{},"the more intense the competition between the three labs, the more capex flows into NVDA/ASML/TSMC",". Trying to pick which lab wins is a worse risk-adjusted bet than positioning across the layer that benefits while everyone keeps running.",[122,23494,23496],{"id":23495},"watch-items","Watch items",[312,23498,23499,23505,23511,23517],{},[315,23500,23501,23504],{},[23,23502,23503],{},"Next major versions of Claude / GPT / Gemini"," — does the crown keep flipping, or does someone consolidate the lead?",[315,23506,23507,23510],{},[23,23508,23509],{},"NVDA earnings, datacenter segment"," — high beta to lab competition intensity",[315,23512,23513,23516],{},[23,23514,23515],{},"OpenAI / Anthropic primary fundraising valuations"," — markup multiples are the temperature gauge",[315,23518,23519,23522],{},[23,23520,23521],{},"Regulation (EU AI Act, US AI Bill of Rights implementation)"," — changes the cost structure of frontier model development",[19,23524,23525],{},"That's all for today. I'd love to hear how you split your usage.",[205,23527],{},[19,23529,23530],{},[802,23531,1603],{},{"title":143,"searchDepth":1605,"depth":1605,"links":23533},[23534,23535,23540,23544,23549,23555,23560,23561,23566,23570,23575,23581],{"id":21892,"depth":1605,"text":21892},{"id":21920,"depth":1605,"text":21920,"children":23536},[23537,23538,23539],{"id":21926,"depth":1610,"text":21927},{"id":21947,"depth":1610,"text":21948},{"id":21965,"depth":1610,"text":21966},{"id":22000,"depth":1605,"text":22000,"children":23541},[23542,23543],{"id":22006,"depth":1610,"text":22006},{"id":22214,"depth":1610,"text":22214},{"id":22303,"depth":1605,"text":22303,"children":23545},[23546,23547,23548],{"id":22309,"depth":1610,"text":22310},{"id":22346,"depth":1610,"text":22347},{"id":22382,"depth":1610,"text":22383},{"id":22408,"depth":1605,"text":22408,"children":23550},[23551,23552,23553,23554],{"id":22418,"depth":1610,"text":22419},{"id":22444,"depth":1610,"text":22445},{"id":22477,"depth":1610,"text":22478},{"id":22507,"depth":1610,"text":22507},{"id":22542,"depth":1605,"text":22543,"children":23556},[23557,23558,23559],{"id":22549,"depth":1610,"text":22549},{"id":22652,"depth":1610,"text":22652},{"id":22673,"depth":1610,"text":22673},{"id":22721,"depth":1605,"text":22722},{"id":22750,"depth":1605,"text":22751,"children":23562},[23563,23564,23565],{"id":22757,"depth":1610,"text":22758},{"id":22773,"depth":1610,"text":22774},{"id":22790,"depth":1610,"text":22791},{"id":22830,"depth":1605,"text":22831,"children":23567},[23568,23569],{"id":22837,"depth":1610,"text":22838},{"id":23030,"depth":1610,"text":23031},{"id":23127,"depth":1605,"text":23128,"children":23571},[23572,23573,23574],{"id":23134,"depth":1610,"text":23135},{"id":23172,"depth":1610,"text":23173},{"id":23207,"depth":1610,"text":23208},{"id":23237,"depth":1605,"text":23238,"children":23576},[23577,23578,23579,23580],{"id":23244,"depth":1610,"text":23245},{"id":23271,"depth":1610,"text":23272},{"id":23302,"depth":1610,"text":23303},{"id":23334,"depth":1610,"text":23335},{"id":23370,"depth":1605,"text":23371,"children":23582},[23583,23584,23585],{"id":23380,"depth":1610,"text":23381},{"id":23477,"depth":1610,"text":23478},{"id":23495,"depth":1610,"text":23496},"Claude Opus 4.7、Gemini 3.1 Pro、GPT-5.4 Pro——2026年4月時点の最新フラッグシップAIをVC実務で比較。使い分けに加え、AIラボ三つ巴の投資シナリオと関連株（NVDA/MSFT/GOOGL）への含意までを整理する。",{"date":23588,"image":23589,"alt":23590,"tags":23591,"tagsEn":23597,"published":1666},"21st Apr 2026","/blogs-img/blog-ai-comparison.png","Claude・Gemini・ChatGPTの最新モデル比較",[23592,23593,23594,23595,23596],"AI","生成AI","Claude","ChatGPT","Gemini",[23592,23598,23594,23595,23596],"Generative AI","/blogs/9-claude-gemini-chatgpt-comparison",{"title":21879,"description":23586},"blogs/9-claude-gemini-chatgpt-comparison","lBFB0XDYxBf6tc7anbCbFqGchHTZ6ST4XCshahinNi8",1777988819613]