[{"data":1,"prerenderedAt":3674},["ShallowReactive",2],{"category-data-フィジカルAI":3},[4,1461],{"id":5,"title":6,"body":7,"description":1439,"extension":1440,"meta":1441,"navigation":1456,"ogImage":1443,"path":1457,"seo":1458,"stem":1459,"__hash__":1460},"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":8,"value":9,"toc":1389},"minimark",[10,715],[11,12,14,19,24,28,32,40,47,50,54,122,127,134,145,156,159,162,216,218,222,225,281,284,295,299,302,322,329,331,335,342,346,389,396,398,402,405,502,506,512,519,521,525,528,531,612,615,641,644,647,674,676,680,686,693,696,701,707,709],"lang-block",{"lang":13},"ja",[15,16,18],"h1",{"id":17},"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",[20,21,23],"h2",{"id":22},"ami-labsが示す世界モデルの可能性","——AMI Labsが示す「世界モデル」の可能性",[20,25,27],{"id":26},"はじめにllmの次に来るもの","はじめに：LLMの次に来るもの",[29,30,31],"p",{},"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への投資ブームはひと段落し、投資家たちの視線は「その次」へと向かっています。",[29,33,34,35,39],{},"キーワードは ",[36,37,38],"strong",{},"フィジカルAI（Physical 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です。",[29,41,42,43,46],{},"この潮流の象徴として世界が注目しているのが、チューリング賞受賞者・Yann LeCunが2026年初頭に創業した ",[36,44,45],{},"AMI Labs（Advanced Machine Intelligence Labs）"," です。",[48,49],"hr",{},[20,51,53],{"id":52},"ami-labsとは何か","AMI Labsとは何か",[55,56,57,70],"table",{},[58,59,60],"thead",{},[61,62,63,67],"tr",{},[64,65,66],"th",{},"項目",[64,68,69],{},"内容",[71,72,73,82,90,98,106,114],"tbody",{},[61,74,75,79],{},[76,77,78],"td",{},"正式名称",[76,80,81],{},"Advanced Machine Intelligence Labs（AMI Labs）",[61,83,84,87],{},[76,85,86],{},"創業",[76,88,89],{},"2026年（Yann LeCun × Alexandre Lebrun）",[61,91,92,95],{},[76,93,94],{},"本社",[76,96,97],{},"パリ（フランス）",[61,99,100,103],{},[76,101,102],{},"調達額",[76,104,105],{},"10.3億ドル（欧州史上最大のシードラウンド）",[61,107,108,111],{},[76,109,110],{},"バリュエーション",[76,112,113],{},"約35億ドル（pre-money）",[61,115,116,119],{},[76,117,118],{},"主要投資家",[76,120,121],{},"Cathay Innovation、Greycroft、HV Capital、Toyota Ventures、NVIDIA、Bezos Expeditions、Temasek ほか",[123,124,126],"h3",{"id":125},"技術の核心世界モデルworld-models","技術の核心：世界モデル（World Models）",[29,128,129,130,133],{},"AMI Labsが開発するのは、",[36,131,132],{},"世界の物理的な仕組みを学習するAI","——「世界モデル」です。",[29,135,136,137,140,141,144],{},"LeCunが2022年に提唱した ",[36,138,139],{},"JEPA（Joint Embedding Predictive Architecture）"," をベースとしており、LLMのようにテキストを生成するのではなく、",[36,142,143],{},"現実世界の抽象的な表現を学習し、行動の結果を予測・計画する"," ことが特徴です。",[146,147,152],"pre",{"className":148,"code":150,"language":151},[149],"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",[153,154,150],"code",{"__ignoreMap":155},"",[29,157,158],{},"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」と主張してきました。AMI Labsはその哲学を具現化した挑戦です。",[123,160,161],{"id":161},"主な応用分野",[55,163,164,174],{},[58,165,166],{},[61,167,168,171],{},[64,169,170],{},"分野",[64,172,173],{},"具体例",[71,175,176,184,192,200,208],{},[61,177,178,181],{},[76,179,180],{},"産業ロボティクス",[76,182,183],{},"工場での複雑な組み立て・検査作業",[61,185,186,189],{},[76,187,188],{},"自律走行",[76,190,191],{},"予測不能な道路環境への適応",[61,193,194,197],{},[76,195,196],{},"医療・ヘルスケア",[76,198,199],{},"手術支援ロボット、診断補助",[61,201,202,205],{},[76,203,204],{},"ウェアラブル",[76,206,207],{},"状況を理解して支援するデバイス",[61,209,210,213],{},[76,211,212],{},"宇宙・航空宇宙",[76,214,215],{},"遠隔環境での自律オペレーション",[48,217],{},[20,219,221],{"id":220},"なぜ今フィジカルaiなのか","なぜ今「フィジカルAI」なのか",[123,223,224],{"id":224},"市場規模の爆発的成長",[55,226,227,246],{},[58,228,229],{},[61,230,231,234,237,240,243],{},[64,232,233],{},"指標",[64,235,236],{},"2025年",[64,238,239],{},"2030年予測",[64,241,242],{},"2033年予測",[64,244,245],{},"CAGR",[71,247,248,265],{},[61,249,250,253,256,259,262],{},[76,251,252],{},"フィジカルAI市場",[76,254,255],{},"52億ドル",[76,257,258],{},"—",[76,260,261],{},"497億ドル",[76,263,264],{},"32.5%",[61,266,267,270,273,276,278],{},[76,268,269],{},"身体化AI（Embodied AI）",[76,271,272],{},"44億ドル",[76,274,275],{},"231億ドル",[76,277,258],{},[76,279,280],{},"39.0%",[123,282,283],{"id":283},"ヒューマノイドロボットの量産化",[29,285,286,287,290,291,294],{},"Goldman Sachsは2026年のヒューマノイドロボット世界出荷台数を ",[36,288,289],{},"5〜10万台"," と予測。製造コストも最終的には ",[36,292,293],{},"1台あたり1.5〜2万ドル"," まで低下すると見込まれており、大規模普及の条件が整いつつあります。",[123,296,298],{"id":297},"llmの次の壁","LLMの「次の壁」",[29,300,301],{},"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います：",[303,304,305,312,317],"ul",{},[306,307,308,311],"li",{},[36,309,310],{},"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」）",[306,313,314],{},[36,315,316],{},"長期的な計画・行動が苦手",[306,318,319],{},[36,320,321],{},"ロボット操作などリアルタイム制御に不向き",[29,323,324,325,328],{},"世界モデルはこれらを克服し、",[36,326,327],{},"「考えて動く」AIを実現する鍵","として注目されています。",[48,330],{},[20,332,334],{"id":333},"scrum-venturesが注目するフィジカルai企業群","Scrum Venturesが注目するフィジカルAI企業群",[29,336,337,338,341],{},"サンフランシスコ・東京を拠点とするベンチャーキャピタル ",[36,339,340],{},"Scrum Ventures"," は、AIとロボティクスを主要投資テーマの一つに据え、フィジカルAI分野での先行投資を積極的に行っています。",[123,343,345],{"id":344},"注目ポートフォリオapptronik","注目ポートフォリオ：Apptronik",[55,347,348,356],{},[58,349,350],{},[61,351,352,354],{},[64,353,66],{},[64,355,69],{},[71,357,358,366,373,381],{},[61,359,360,363],{},[76,361,362],{},"企業名",[76,364,365],{},"Apptronik",[61,367,368,370],{},[76,369,170],{},[76,371,372],{},"ヒューマノイドロボット",[61,374,375,378],{},[76,376,377],{},"直近調達",[76,379,380],{},"3.5億ドル（B Capital・Googleが主導）",[61,382,383,386],{},[76,384,385],{},"特徴",[76,387,388],{},"産業用途向け二足歩行ロボット「Apollo」を開発。Nasaとの共同研究実績も持つ",[29,390,391,392,395],{},"Scrum Venturesはこのような、",[36,393,394],{},"現実の物理環境で動作するAIロボット企業","への投資を通じて、フィジカルAI時代のインフラ構築を支援しています。",[48,397],{},[20,399,401],{"id":400},"フィジカルai主要プレイヤー比較","フィジカルAI主要プレイヤー比較",[29,403,404],{},"AMI Labsと同様に、世界モデル・ロボット基盤モデルを開発する主要企業を比較します。",[55,406,407,422],{},[58,408,409],{},[61,410,411,414,417,419],{},[64,412,413],{},"企業",[64,415,416],{},"調達総額",[64,418,110],{},[64,420,421],{},"技術的特徴",[71,423,424,440,456,471,487],{},[61,425,426,431,434,437],{},[76,427,428],{},[36,429,430],{},"AMI Labs",[76,432,433],{},"10.3億ドル",[76,435,436],{},"約35億ドル",[76,438,439],{},"世界モデル（JEPA）、LeCun主導",[61,441,442,447,450,453],{},[76,443,444],{},[36,445,446],{},"Physical Intelligence（π）",[76,448,449],{},"11億ドル",[76,451,452],{},"約56億ドル",[76,454,455],{},"ロボット基盤モデル（π0）、家庭・産業用",[61,457,458,462,465,468],{},[76,459,460],{},[36,461,365],{},[76,463,464],{},"3.5億ドル以上",[76,466,467],{},"非公開",[76,469,470],{},"ヒューマノイドロボット「Apollo」、NASA連携",[61,472,473,478,481,484],{},[76,474,475],{},[36,476,477],{},"Figure AI",[76,479,480],{},"6.75億ドル",[76,482,483],{},"約26億ドル",[76,485,486],{},"OpenAIとの提携、Figure 01/02",[61,488,489,494,497,499],{},[76,490,491],{},[36,492,493],{},"1X Technologies",[76,495,496],{},"1億ドル以上",[76,498,467],{},[76,500,501],{},"OpenAIバックアップの人型ロボット",[123,503,505],{"id":504},"ami-labs-vs-physical-intelligence-の違い","AMI Labs vs. Physical Intelligence の違い",[146,507,510],{"className":508,"code":509,"language":151},[149],"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",[153,511,509],{"__ignoreMap":155},[29,513,514,515,518],{},"両社は補完的な関係にあり、",[36,516,517],{},"世界モデル（AMI）× ロボット操作モデル（π）"," の組み合わせが、将来的な汎用ロボットを実現する可能性があります。",[48,520],{},[20,522,524],{"id":523},"投資家視点フィジカルaiをどう捉えるか","投資家視点：フィジカルAIをどう捉えるか",[123,526,527],{"id":527},"バリューチェーンで考える",[29,529,530],{},"フィジカルAI産業は以下のレイヤーに分解できます：",[55,532,533,545],{},[58,534,535],{},[61,536,537,540,542],{},[64,538,539],{},"レイヤー",[64,541,69],{},[64,543,544],{},"代表企業・銘柄",[71,546,547,560,573,586,599],{},[61,548,549,554,557],{},[76,550,551],{},[36,552,553],{},"基盤モデル",[76,555,556],{},"世界モデル・ロボット基盤AI",[76,558,559],{},"AMI Labs、Physical Intelligence",[61,561,562,567,570],{},[76,563,564],{},[36,565,566],{},"ハードウェア",[76,568,569],{},"センサー、チップ、アクチュエーター",[76,571,572],{},"NVIDIA（NVDA）、Mobileye",[61,574,575,580,583],{},[76,576,577],{},[36,578,579],{},"ロボット本体",[76,581,582],{},"ヒューマノイド・産業ロボット",[76,584,585],{},"Apptronik、Figure AI、Boston Dynamics",[61,587,588,593,596],{},[76,589,590],{},[36,591,592],{},"プラットフォーム",[76,594,595],{},"データ収集・シミュレーション基盤",[76,597,598],{},"Foxglove（40M調達）、Dyna Robotics",[61,600,601,606,609],{},[76,602,603],{},[36,604,605],{},"エンドユーザー",[76,607,608],{},"製造・物流・医療への導入",[76,610,611],{},"Amazon、Tesla、Toyota",[123,613,614],{"id":614},"リスク要因",[303,616,617,623,629,635],{},[306,618,619,622],{},[36,620,621],{},"技術的不確実性","：世界モデルの実用化は研究段階のものが多い",[306,624,625,628],{},[36,626,627],{},"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",[306,630,631,634],{},[36,632,633],{},"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",[306,636,637,640],{},[36,638,639],{},"長い開発サイクル","：ソフトウェアAIと異なり、実証・量産に時間がかかる",[123,642,643],{"id":643},"個人投資家の現実的アプローチ",[29,645,646],{},"フィジカルAIのスタートアップへの直接投資は上場前企業が多く困難です。現実的なアプローチ：",[648,649,650,656,662,668],"ol",{},[306,651,652,655],{},[36,653,654],{},"NVIDIA（NVDA）","：フィジカルAIの「ピッケルと鍬」。GPUに加え、Isaacsim（ロボットシミュレーター）でエコシステムを構築",[306,657,658,661],{},[36,659,660],{},"ロボティクス関連ETF","：ROBO、ARKQ など",[306,663,664,667],{},[36,665,666],{},"トヨタ・ホンダ等の自動車メーカー株","：自律走行・工場自動化への先行投資",[306,669,670,673],{},[36,671,672],{},"間接的受益企業","：センサーメーカー、半導体設計会社",[48,675],{},[20,677,679],{"id":678},"まとめフィジカルaiは10年に一度の転換点","まとめ：フィジカルAIは「10年に一度」の転換点",[29,681,682,683,46],{},"LLMが「情報処理革命」だとすれば、フィジカルAIは ",[36,684,685],{},"「物理世界との相互作用革命」",[29,687,688,689,692],{},"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、",[36,690,691],{},"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","です。",[29,694,695],{},"Scrum Venturesのような先見性のある投資家が、この波に早期からベットしている事実は示唆に富みます。",[29,697,698],{},[36,699,700],{},"ZYL0 の視点：",[702,703,704],"blockquote",{},[29,705,706],{},"フィジカルAIは「SFの世界」ではなく、今まさに資本が流れ込んでいる「現実の投資テーマ」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。",[48,708],{},[29,710,711],{},[712,713,714],"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。",[11,716,718,722,726,730,733,740,747,749,753,815,819,826,836,842,845,849,903,905,909,913,965,969,980,984,987,1005,1011,1013,1017,1022,1026,1068,1075,1077,1081,1173,1177,1183,1190,1192,1196,1200,1203,1284,1288,1314,1318,1321,1347,1349,1353,1363,1369,1372,1377,1382,1384],{"lang":717},"en",[15,719,721],{"id":720},"physical-ai-the-next-investment-frontier","Physical AI: The Next Investment Frontier",[20,723,725],{"id":724},"what-ami-labs-reveals-about-the-world-model-era","——What AMI Labs Reveals About the World Model Era",[20,727,729],{"id":728},"introduction-what-comes-after-llms","Introduction: What Comes After LLMs?",[29,731,732],{},"Large language models (LLMs) like ChatGPT have dominated the technology landscape for the past several years. Now, as the initial generative AI investment frenzy settles, the smart money is asking a new question: what comes next?",[29,734,735,736,739],{},"The answer is increasingly clear: ",[36,737,738],{},"Physical AI"," — the technology domain where AI moves beyond language and images to understand and act within the real, physical world.",[29,741,742,743,746],{},"The defining signal of this shift came in early 2026, when Turing Award winner Yann LeCun launched ",[36,744,745],{},"AMI Labs (Advanced Machine Intelligence Labs)",", securing the largest seed round in European history and placing a very large bet on a fundamentally different kind of AI.",[48,748],{},[20,750,752],{"id":751},"what-is-ami-labs","What Is AMI Labs?",[55,754,755,765],{},[58,756,757],{},[61,758,759,762],{},[64,760,761],{},"Item",[64,763,764],{},"Details",[71,766,767,775,783,791,799,807],{},[61,768,769,772],{},[76,770,771],{},"Full Name",[76,773,774],{},"Advanced Machine Intelligence Labs (AMI Labs)",[61,776,777,780],{},[76,778,779],{},"Founded",[76,781,782],{},"2026 (Yann LeCun × Alexandre Lebrun)",[61,784,785,788],{},[76,786,787],{},"Headquarters",[76,789,790],{},"Paris, France",[61,792,793,796],{},[76,794,795],{},"Funding",[76,797,798],{},"$1.03B (Europe's largest seed round ever)",[61,800,801,804],{},[76,802,803],{},"Valuation",[76,805,806],{},"~$3.5B (pre-money)",[61,808,809,812],{},[76,810,811],{},"Key Investors",[76,813,814],{},"Cathay Innovation, Greycroft, HV Capital, Toyota Ventures, NVIDIA, Bezos Expeditions, Temasek, and others",[123,816,818],{"id":817},"the-core-technology-world-models","The Core Technology: World Models",[29,820,821,822,825],{},"AMI Labs is building AI that learns how the physical world works — so-called ",[36,823,824],{},"world models",".",[29,827,828,829,832,833,825],{},"The approach is grounded in ",[36,830,831],{},"JEPA (Joint Embedding Predictive Architecture)",", a framework LeCun proposed in 2022 as an alternative to the LLM paradigm. Rather than generating text, JEPA-based systems learn abstract representations of real-world sensory data and use them to ",[36,834,835],{},"predict consequences and plan sequences of actions",[146,837,840],{"className":838,"code":839,"language":151},[149],"LLM approach:    Text → Generate more text\nWorld Model:     Sensor data → Understand physical world → Plan actions\n",[153,841,839],{"__ignoreMap":155},[29,843,844],{},"LeCun has long argued that LLMs alone cannot reach artificial general intelligence. AMI Labs is his answer to that gap.",[123,846,848],{"id":847},"target-applications","Target Applications",[55,850,851,861],{},[58,852,853],{},[61,854,855,858],{},[64,856,857],{},"Domain",[64,859,860],{},"Examples",[71,862,863,871,879,887,895],{},[61,864,865,868],{},[76,866,867],{},"Industrial robotics",[76,869,870],{},"Complex assembly, quality inspection in factories",[61,872,873,876],{},[76,874,875],{},"Autonomous vehicles",[76,877,878],{},"Adapting to unpredictable road environments",[61,880,881,884],{},[76,882,883],{},"Healthcare",[76,885,886],{},"Surgical assist robots, diagnostic support",[61,888,889,892],{},[76,890,891],{},"Wearables",[76,893,894],{},"Context-aware assistive devices",[61,896,897,900],{},[76,898,899],{},"Aerospace",[76,901,902],{},"Autonomous operations in remote environments",[48,904],{},[20,906,908],{"id":907},"why-physical-ai-why-now","Why Physical AI, Why Now?",[123,910,912],{"id":911},"market-size-explosive-growth-ahead","Market Size: Explosive Growth Ahead",[55,914,915,933],{},[58,916,917],{},[61,918,919,922,925,928,931],{},[64,920,921],{},"Metric",[64,923,924],{},"2025",[64,926,927],{},"2030 Forecast",[64,929,930],{},"2033 Forecast",[64,932,245],{},[71,934,935,950],{},[61,936,937,940,943,945,948],{},[76,938,939],{},"Physical AI Market",[76,941,942],{},"$5.2B",[76,944,258],{},[76,946,947],{},"$49.7B",[76,949,264],{},[61,951,952,955,958,961,963],{},[76,953,954],{},"Embodied AI Market",[76,956,957],{},"$4.4B",[76,959,960],{},"$23.1B",[76,962,258],{},[76,964,280],{},[123,966,968],{"id":967},"humanoid-robots-are-entering-mass-production","Humanoid Robots Are Entering Mass Production",[29,970,971,972,975,976,979],{},"Goldman Sachs forecasts global humanoid robot shipments of ",[36,973,974],{},"50,000–100,000 units in 2026",", with per-unit costs eventually declining to ",[36,977,978],{},"$15,000–$20,000",". The conditions for mass adoption are falling into place.",[123,981,983],{"id":982},"the-wall-llms-cannot-climb","The Wall LLMs Cannot Climb",[29,985,986],{},"LLMs have been extraordinary at language and image tasks, but they have clear structural limitations:",[303,988,989,995,1000],{},[306,990,991,994],{},[36,992,993],{},"No physical common sense"," (they don't \"know\" that a cup left on a ledge will fall)",[306,996,997],{},[36,998,999],{},"Weak at long-horizon planning and physical action",[306,1001,1002],{},[36,1003,1004],{},"Ill-suited for real-time robotic control",[29,1006,1007,1008,825],{},"World models are designed to overcome these limitations — enabling AI that can ",[36,1009,1010],{},"reason about and act within the physical world",[48,1012],{},[20,1014,1016],{"id":1015},"physical-ai-companies-on-scrum-ventures-radar","Physical AI Companies on Scrum Ventures' Radar",[29,1018,1019,1021],{},[36,1020,340],{},", a seed-stage VC with offices in San Francisco and Tokyo, has identified AI and robotics as a core investment thesis. The firm is among those actively deploying capital into the Physical AI space.",[123,1023,1025],{"id":1024},"portfolio-spotlight-apptronik","Portfolio Spotlight: Apptronik",[55,1027,1028,1036],{},[58,1029,1030],{},[61,1031,1032,1034],{},[64,1033,761],{},[64,1035,764],{},[71,1037,1038,1045,1052,1060],{},[61,1039,1040,1043],{},[76,1041,1042],{},"Company",[76,1044,365],{},[61,1046,1047,1049],{},[76,1048,857],{},[76,1050,1051],{},"Humanoid robotics",[61,1053,1054,1057],{},[76,1055,1056],{},"Recent Funding",[76,1058,1059],{},"$350M (led by B Capital and Google)",[61,1061,1062,1065],{},[76,1063,1064],{},"Technology",[76,1066,1067],{},"Apollo biped robot for industrial environments; NASA collaboration track record",[29,1069,1070,1071,1074],{},"Scrum Ventures' involvement in companies like Apptronik reflects a thesis that ",[36,1072,1073],{},"the AI-native robotic stack"," — hardware + software + world models — will define the next computing platform.",[48,1076],{},[20,1078,1080],{"id":1079},"competitive-landscape-key-physical-ai-players","Competitive Landscape: Key Physical AI Players",[55,1082,1083,1096],{},[58,1084,1085],{},[61,1086,1087,1089,1092,1094],{},[64,1088,1042],{},[64,1090,1091],{},"Total Raised",[64,1093,803],{},[64,1095,1064],{},[71,1097,1098,1113,1129,1144,1159],{},[61,1099,1100,1104,1107,1110],{},[76,1101,1102],{},[36,1103,430],{},[76,1105,1106],{},"$1.03B",[76,1108,1109],{},"~$3.5B",[76,1111,1112],{},"World models (JEPA), Yann LeCun",[61,1114,1115,1120,1123,1126],{},[76,1116,1117],{},[36,1118,1119],{},"Physical Intelligence (π)",[76,1121,1122],{},"$1.1B",[76,1124,1125],{},"~$5.6B",[76,1127,1128],{},"Robot foundation models (π0)",[61,1130,1131,1135,1138,1141],{},[76,1132,1133],{},[36,1134,365],{},[76,1136,1137],{},"$350M+",[76,1139,1140],{},"Undisclosed",[76,1142,1143],{},"Humanoid robot \"Apollo\", NASA partner",[61,1145,1146,1150,1153,1156],{},[76,1147,1148],{},[36,1149,477],{},[76,1151,1152],{},"$675M",[76,1154,1155],{},"~$2.6B",[76,1157,1158],{},"OpenAI partnership, Figure 01/02",[61,1160,1161,1165,1168,1170],{},[76,1162,1163],{},[36,1164,493],{},[76,1166,1167],{},"$100M+",[76,1169,1140],{},[76,1171,1172],{},"OpenAI-backed humanoid robots",[123,1174,1176],{"id":1175},"ami-labs-vs-physical-intelligence-a-key-distinction","AMI Labs vs. Physical Intelligence: A Key Distinction",[146,1178,1181],{"className":1179,"code":1180,"language":151},[149],"AMI Labs:\n  └── \"Understanding the world\" is the goal; robots are one application\n  └── Research-led. JEPA learns abstract physical representations\n\nPhysical Intelligence (π):\n  └── Focused specifically on \"making robots act\"\n  └── Provides π0 model to robotics manufacturers (OpenAI-style positioning)\n",[153,1182,1180],{"__ignoreMap":155},[29,1184,1185,1186,1189],{},"These two companies are more complementary than competitive. The combination of ",[36,1187,1188],{},"world models (AMI) × robot action models (π)"," may be the path to truly general-purpose robots.",[48,1191],{},[20,1193,1195],{"id":1194},"investment-framework-how-to-think-about-physical-ai","Investment Framework: How to Think About Physical AI",[123,1197,1199],{"id":1198},"the-value-chain","The Value Chain",[29,1201,1202],{},"Physical AI breaks down into distinct investment layers:",[55,1204,1205,1217],{},[58,1206,1207],{},[61,1208,1209,1212,1215],{},[64,1210,1211],{},"Layer",[64,1213,1214],{},"Description",[64,1216,860],{},[71,1218,1219,1232,1245,1258,1271],{},[61,1220,1221,1226,1229],{},[76,1222,1223],{},[36,1224,1225],{},"Foundation Models",[76,1227,1228],{},"World models, robot action AI",[76,1230,1231],{},"AMI Labs, Physical Intelligence",[61,1233,1234,1239,1242],{},[76,1235,1236],{},[36,1237,1238],{},"Hardware",[76,1240,1241],{},"Sensors, chips, actuators",[76,1243,1244],{},"NVIDIA (NVDA), Mobileye",[61,1246,1247,1252,1255],{},[76,1248,1249],{},[36,1250,1251],{},"Robots",[76,1253,1254],{},"Humanoid, industrial machines",[76,1256,1257],{},"Apptronik, Figure AI, Boston Dynamics",[61,1259,1260,1265,1268],{},[76,1261,1262],{},[36,1263,1264],{},"Platforms",[76,1266,1267],{},"Data infra, simulation tools",[76,1269,1270],{},"Foxglove ($40M), Dyna Robotics",[61,1272,1273,1278,1281],{},[76,1274,1275],{},[36,1276,1277],{},"End Users",[76,1279,1280],{},"Manufacturing, logistics, healthcare",[76,1282,1283],{},"Amazon, Tesla, Toyota",[123,1285,1287],{"id":1286},"key-risk-factors","Key Risk Factors",[303,1289,1290,1296,1302,1308],{},[306,1291,1292,1295],{},[36,1293,1294],{},"Technical uncertainty",": World models are still largely in research stages",[306,1297,1298,1301],{},[36,1299,1300],{},"Hardware bottlenecks",": Robot body cost, durability, reliability",[306,1303,1304,1307],{},[36,1305,1306],{},"Regulatory landscape",": Autonomous machines in healthcare and public spaces face evolving rules",[306,1309,1310,1313],{},[36,1311,1312],{},"Long development cycles",": Unlike software AI, physical systems require real-world validation and manufacturing scale-up",[123,1315,1317],{"id":1316},"realistic-strategies-for-individual-investors","Realistic Strategies for Individual Investors",[29,1319,1320],{},"Direct investment in Physical AI startups is largely inaccessible pre-IPO. Practical approaches:",[648,1322,1323,1329,1335,1341],{},[306,1324,1325,1328],{},[36,1326,1327],{},"NVIDIA (NVDA)",": The \"picks and shovels\" of Physical AI. Beyond GPUs, Isaac Sim (robot simulator) makes NVIDIA the ecosystem hub",[306,1330,1331,1334],{},[36,1332,1333],{},"Robotics ETFs",": ROBO, ARKQ",[306,1336,1337,1340],{},[36,1338,1339],{},"Automotive OEMs",": Toyota, Honda — investing heavily in autonomous manufacturing and mobility",[306,1342,1343,1346],{},[36,1344,1345],{},"Indirect beneficiaries",": Sensor manufacturers, chip design companies",[48,1348],{},[20,1350,1352],{"id":1351},"conclusion-physical-ai-is-a-once-in-a-decade-shift","Conclusion: Physical AI Is a Once-in-a-Decade Shift",[29,1354,1355,1356,1359,1360,825],{},"If LLMs represented a revolution in ",[36,1357,1358],{},"information processing",", Physical AI represents a revolution in ",[36,1361,1362],{},"interaction with the physical world",[29,1364,1365,1366,825],{},"AMI Labs' $1B raise is not just one startup's success story — it is a ",[36,1367,1368],{},"signal that the center of gravity of the AI industry is shifting from digital space to physical space",[29,1370,1371],{},"The fact that forward-thinking investors like Scrum Ventures are placing bets on this thesis early is worth noting.",[29,1373,1374],{},[36,1375,1376],{},"ZYL0's View:",[702,1378,1379],{},[29,1380,1381],{},"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.",[48,1383],{},[29,1385,1386],{},[712,1387,1388],{},"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":155,"searchDepth":1390,"depth":1390,"links":1391},2,[1392,1393,1394,1399,1404,1407,1410,1415,1416,1417,1418,1422,1427,1430,1433,1438],{"id":22,"depth":1390,"text":23},{"id":26,"depth":1390,"text":27},{"id":52,"depth":1390,"text":53,"children":1395},[1396,1398],{"id":125,"depth":1397,"text":126},3,{"id":161,"depth":1397,"text":161},{"id":220,"depth":1390,"text":221,"children":1400},[1401,1402,1403],{"id":224,"depth":1397,"text":224},{"id":283,"depth":1397,"text":283},{"id":297,"depth":1397,"text":298},{"id":333,"depth":1390,"text":334,"children":1405},[1406],{"id":344,"depth":1397,"text":345},{"id":400,"depth":1390,"text":401,"children":1408},[1409],{"id":504,"depth":1397,"text":505},{"id":523,"depth":1390,"text":524,"children":1411},[1412,1413,1414],{"id":527,"depth":1397,"text":527},{"id":614,"depth":1397,"text":614},{"id":643,"depth":1397,"text":643},{"id":678,"depth":1390,"text":679},{"id":724,"depth":1390,"text":725},{"id":728,"depth":1390,"text":729},{"id":751,"depth":1390,"text":752,"children":1419},[1420,1421],{"id":817,"depth":1397,"text":818},{"id":847,"depth":1397,"text":848},{"id":907,"depth":1390,"text":908,"children":1423},[1424,1425,1426],{"id":911,"depth":1397,"text":912},{"id":967,"depth":1397,"text":968},{"id":982,"depth":1397,"text":983},{"id":1015,"depth":1390,"text":1016,"children":1428},[1429],{"id":1024,"depth":1397,"text":1025},{"id":1079,"depth":1390,"text":1080,"children":1431},[1432],{"id":1175,"depth":1397,"text":1176},{"id":1194,"depth":1390,"text":1195,"children":1434},[1435,1436,1437],{"id":1198,"depth":1397,"text":1199},{"id":1286,"depth":1397,"text":1287},{"id":1316,"depth":1397,"text":1317},{"id":1351,"depth":1390,"text":1352},"Yann LeCunが創業した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解説。Scrum Venturesが注目する具体的な企業群と投資戦略を日英バイリンガルで紹介。","md",{"date":1442,"image":1443,"alt":1444,"tags":1445,"tagsEn":1451,"published":1456},"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",[1446,1447,1448,1449,1450],"フィジカルAI","AI投資","ロボティクス","世界モデル","スタートアップ",[738,1452,1453,1454,1455],"AI Investment","Robotics","World Models","Startups",true,"/blogs/2-physical-ai-next-investment-theme",{"title":6,"description":1439},"blogs/2-physical-ai-next-investment-theme","pWGz4PbzyzsFQIpJeKfMvMPVb977qCF-ovg98P1t_aU",{"id":1462,"title":1463,"body":1464,"description":3658,"extension":1440,"meta":3659,"navigation":1456,"ogImage":3661,"path":3670,"seo":3671,"stem":3672,"__hash__":3673},"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":8,"value":1465,"toc":3600},[1466,2581],[11,1467,1468,1472,1476,1479,1482,1485,1488,1490,1493,1513,1515,1519,1739,1741,1745,1749,1752,1756,1759,1766,1770,1773,1776,1781,1834,1836,1840,1843,1846,1852,1856,1859,1862,1868,1872,1875,1878,1882,1885,1888,1892,1895,1898,1903,1954,1956,1960,1964,1967,1981,1988,1992,1995,1998,2018,2021,2027,2031,2034,2084,2087,2089,2093,2097,2100,2107,2111,2114,2117,2121,2124,2127,2132,2176,2178,2182,2559,2561,2565],{"lang":13},[15,1469,1471],{"id":1470},"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",[20,1473,1475],{"id":1474},"連載技術別深掘りとケーススタディ","——連載②：技術別深掘りとケーススタディ",[20,1477,1478],{"id":1478},"はじめに",[29,1480,1481],{},"連載第2回は技術領域の深掘りだ。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」だ。",[29,1483,1484],{},"なぜか。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る。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。",[29,1486,1487],{},"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く。",[48,1489],{},[20,1491,1492],{"id":1492},"エグゼクティブサマリ",[303,1494,1495,1501,1507],{},[306,1496,1497,1500],{},[36,1498,1499],{},"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る。",[306,1502,1503,1506],{},[36,1504,1505],{},"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。",[306,1508,1509,1512],{},[36,1510,1511],{},"「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る。",[48,1514],{},[20,1516,1518],{"id":1517},"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",[55,1520,1521,1540],{},[58,1522,1523],{},[61,1524,1525,1528,1531,1534,1537],{},[64,1526,1527],{},"技術領域",[64,1529,1530],{},"代表企業（VC）",[64,1532,1533],{},"主要ラウンド",[64,1535,1536],{},"投資額",[64,1538,1539],{},"技術の役割",[71,1541,1542,1559,1575,1592,1609,1626,1643,1659,1675,1692,1708,1724],{},[61,1543,1544,1547,1550,1553,1556],{},[76,1545,1546],{},"設計AI/EDA",[76,1548,1549],{},"Ricursive（Sequoia）",[76,1551,1552],{},"Seed",[76,1554,1555],{},"$35M",[76,1557,1558],{},"チップ設計サイクルの短縮",[61,1560,1561,1563,1566,1569,1572],{},[76,1562,1546],{},[76,1564,1565],{},"ChipAgents（Bessemer）",[76,1567,1568],{},"Series A",[76,1570,1571],{},"$21M",[76,1573,1574],{},"設計・検証のエージェント化",[61,1576,1577,1580,1583,1586,1589],{},[76,1578,1579],{},"RISC-V IP",[76,1581,1582],{},"Akeana（Kleiner Perkins）",[76,1584,1585],{},"累計",[76,1587,1588],{},">$100M",[76,1590,1591],{},"自由に使えるCPU命令セット",[61,1593,1594,1597,1600,1603,1606],{},[76,1595,1596],{},"SerDes/IP",[76,1598,1599],{},"Kandou（Bessemer）",[76,1601,1602],{},"Series C",[76,1604,1605],{},"$92.3M",[76,1607,1608],{},"高速チップ間通信の物理層",[61,1610,1611,1614,1617,1620,1623],{},[76,1612,1613],{},"コヒレントDSP",[76,1615,1616],{},"Retym（Kleiner Perkins）",[76,1618,1619],{},"Series D",[76,1621,1622],{},"$75M",[76,1624,1625],{},"長距離光通信の信号処理",[61,1627,1628,1631,1634,1637,1640],{},[76,1629,1630],{},"CPO/光I/O",[76,1632,1633],{},"Ayar Labs（Sequoia系）",[76,1635,1636],{},"Series E",[76,1638,1639],{},"$500M",[76,1641,1642],{},"銅配線の限界をシリコンフォトニクスで超える",[61,1644,1645,1648,1651,1654,1656],{},[76,1646,1647],{},"AIスイッチ",[76,1649,1650],{},"Nexthop AI（a16z）",[76,1652,1653],{},"Series B",[76,1655,1639],{},[76,1657,1658],{},"AI大規模クラスタのネットワーク",[61,1660,1661,1664,1667,1669,1672],{},[76,1662,1663],{},"アナログ新計算",[76,1665,1666],{},"Unconventional AI（a16z）",[76,1668,1552],{},[76,1670,1671],{},"$475M",[76,1673,1674],{},"電力効率の桁違い改善を狙う確率的計算",[61,1676,1677,1680,1683,1686,1689],{},[76,1678,1679],{},"AIチップ（買収）",[76,1681,1682],{},"Graphcore（SoftBank）",[76,1684,1685],{},"M&A",[76,1687,1688],{},"非開示",[76,1690,1691],{},"英国発グラフプロセッサ",[61,1693,1694,1697,1700,1702,1705],{},[76,1695,1696],{},"CPU/Arm（買収）",[76,1698,1699],{},"Ampere（SoftBank）",[76,1701,1685],{},[76,1703,1704],{},"$6.5B",[76,1706,1707],{},"クラウドサーバー向けArmベースCPU",[61,1709,1710,1713,1716,1718,1721],{},[76,1711,1712],{},"量子",[76,1714,1715],{},"Rigetti（Bessemer）",[76,1717,1602],{},[76,1719,1720],{},"$79M",[76,1722,1723],{},"量子クラウド提供+ハードウェア",[61,1725,1726,1728,1731,1733,1736],{},[76,1727,1712],{},[76,1729,1730],{},"Quantum Circuits（Sequoia）",[76,1732,1653],{},[76,1734,1735],{},">$60M",[76,1737,1738],{},"フォールトトレラント量子コンピュータ",[48,1740],{},[20,1742,1744],{"id":1743},"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",[123,1746,1748],{"id":1747},"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",[29,1750,1751],{},"先端チップの設計工数は年々増大している。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る。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る。",[123,1753,1755],{"id":1754},"ケーススタディricursive-intelligence35m2025年seed","ケーススタディ：Ricursive Intelligence（$35M、2025年Seed）",[29,1757,1758],{},"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だ。",[29,1760,1761,1762,1765],{},"注目点は",[36,1763,1764],{},"ソフトウェア比率が高い","ことだ。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。",[123,1767,1769],{"id":1768},"ケーススタディchipagents21m2025年series-a","ケーススタディ：ChipAgents（$21M、2025年Series A）",[29,1771,1772],{},"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だ。",[29,1774,1775],{},"「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」ことだ。",[29,1777,1778],{},[36,1779,1780],{},"EDA自動化の投資ロジックまとめ：",[55,1782,1783,1792],{},[58,1784,1785],{},[61,1786,1787,1790],{},[64,1788,1789],{},"観点",[64,1791,69],{},[71,1793,1794,1802,1810,1818,1826],{},[61,1795,1796,1799],{},[76,1797,1798],{},"市場の大きさ",[76,1800,1801],{},"全世界のチップ設計者数×設計サイクル短縮価値",[61,1803,1804,1807],{},[76,1805,1806],{},"製造リスク",[76,1808,1809],{},"低（ソフトウェア主体）",[61,1811,1812,1815],{},[76,1813,1814],{},"競合リスク",[76,1816,1817],{},"EDA大手（Synopsys/Cadence）のM&A対象になりうる",[61,1819,1820,1823],{},[76,1821,1822],{},"出口",[76,1824,1825],{},"M&A（EDA大手/チップ設計会社）またはIPO",[61,1827,1828,1831],{},[76,1829,1830],{},"タイムライン",[76,1832,1833],{},"比較的短い（2〜5年）",[48,1835],{},[20,1837,1839],{"id":1838},"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",[123,1841,1842],{"id":1842},"銅配線の限界とシリコンフォトニクスの台頭",[29,1844,1845],{},"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る。",[146,1847,1850],{"className":1848,"code":1849,"language":151},[149],"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",[153,1851,1849],{"__ignoreMap":155},[123,1853,1855],{"id":1854},"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）",[29,1857,1858],{},"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だ。",[29,1860,1861],{},"戦略投資家にNVIDIA・AMDが名を連ねる構造は重要だ。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」ためだ。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。",[29,1863,1864,1867],{},[36,1865,1866],{},"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する。",[123,1869,1871],{"id":1870},"ケーススタディretym75m2025年series-d","ケーススタディ：Retym（$75M、2025年Series D）",[29,1873,1874],{},"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が参加している。",[29,1876,1877],{},"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だ。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る。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。",[123,1879,1881],{"id":1880},"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）",[29,1883,1884],{},"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だ。",[29,1886,1887],{},"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だ。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。",[123,1889,1891],{"id":1890},"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）",[29,1893,1894],{},"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」を標榜する。",[29,1896,1897],{},"投資ロジックは明快だ。「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する。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。",[29,1899,1900],{},[36,1901,1902],{},"インターコネクト投資の共通ロジック：",[55,1904,1905,1913],{},[58,1906,1907],{},[61,1908,1909,1911],{},[64,1910,1789],{},[64,1912,69],{},[71,1914,1915,1923,1931,1939,1947],{},[61,1916,1917,1920],{},[76,1918,1919],{},"技術ドライバー",[76,1921,1922],{},"GPU/アクセラレータの性能スケールがI/Oを際立たせる",[61,1924,1925,1928],{},[76,1926,1927],{},"顧客",[76,1929,1930],{},"ハイパースケーラー（Google/AWS/Microsoft/Meta等）",[61,1932,1933,1936],{},[76,1934,1935],{},"資本集約度",[76,1937,1938],{},"中〜高（量産・認定が必要）",[61,1940,1941,1944],{},[76,1942,1943],{},"リスク",[76,1945,1946],{},"量産歩留まり、サプライヤー認定、地政学",[61,1948,1949,1951],{},[76,1950,1822],{},[76,1952,1953],{},"M&A（チップ大手）またはIPO",[48,1955],{},[20,1957,1959],{"id":1958},"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",[123,1961,1963],{"id":1962},"パラダイム賭け-vs-垂直統合","パラダイム賭け vs. 垂直統合",[29,1965,1966],{},"「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する。",[648,1968,1969,1975],{},[306,1970,1971,1974],{},[36,1972,1973],{},"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る。",[306,1976,1977,1980],{},[36,1978,1979],{},"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る。",[29,1982,1983,1984,1987],{},"どちらも「単なるチップ投資」ではなく、",[36,1985,1986],{},"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る。",[123,1989,1991],{"id":1990},"ケーススタディunconventional-ai475m2025年seed","ケーススタディ：Unconventional AI（$475M、2025年Seed）",[29,1993,1994],{},"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だ。",[29,1996,1997],{},"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る。主な課題は：",[303,1999,2000,2006,2012],{},[306,2001,2002,2005],{},[36,2003,2004],{},"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",[306,2007,2008,2011],{},[36,2009,2010],{},"ツールチェーン","：設計・検証ツールが整っていない",[306,2013,2014,2017],{},[36,2015,2016],{},"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",[29,2019,2020],{},"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。",[29,2022,2023,2026],{},[36,2024,2025],{},"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。",[123,2028,2030],{"id":2029},"softbankの垂直統合armgraphcoreampere","SoftBankの垂直統合：Arm×Graphcore×Ampere",[29,2032,2033],{},"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だ。",[55,2035,2036,2049],{},[58,2037,2038],{},[61,2039,2040,2043,2046],{},[64,2041,2042],{},"会社",[64,2044,2045],{},"役割",[64,2047,2048],{},"SoftBankの取得形態",[71,2050,2051,2062,2073],{},[61,2052,2053,2056,2059],{},[76,2054,2055],{},"Arm",[76,2057,2058],{},"CPU IP、命令セット標準、エコシステムの核",[76,2060,2061],{},"2016年買収（$32B）",[61,2063,2064,2067,2070],{},[76,2065,2066],{},"Graphcore",[76,2068,2069],{},"グラフ最適化AI処理（GC200/BOW等）",[76,2071,2072],{},"2024年買収",[61,2074,2075,2078,2081],{},[76,2076,2077],{},"Ampere Computing",[76,2079,2080],{},"Armベースクラウドサーバーへの最適化",[76,2082,2083],{},"2025年$6.5Bで買収発表",[29,2085,2086],{},"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」という事業戦略だ。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だ。",[48,2088],{},[20,2090,2092],{"id":2091},"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",[123,2094,2096],{"id":2095},"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",[29,2098,2099],{},"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る。",[29,2101,2102,2103,2106],{},"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、",[36,2104,2105],{},"「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だ。",[123,2108,2110],{"id":2109},"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）",[29,2112,2113],{},"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。",[29,2115,2116],{},"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だ。「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。",[123,2118,2120],{"id":2119},"ケーススタディrigetti-computing79m2020年series-c","ケーススタディ：Rigetti Computing（$79M、2020年Series C）",[29,2122,2123],{},"Bessemerがリードしたこの案件は、量子ハードウェア+クラウド提供（Quantum Cloud Services）という垂直統合モデルだ。",[29,2125,2126],{},"後にSPACでNASDAQ上場（2022年）し、量子スタートアップとして初期の公開市場実績を残した。クラウド提供という「ソフト的な入り口」を持つことで、企業顧客へのアクセスを維持する戦略は、量子の商用化において重要な設計判断だった。",[29,2128,2129],{},[36,2130,2131],{},"量子投資の評価基準（現在）：",[55,2133,2134,2144],{},[58,2135,2136],{},[61,2137,2138,2141],{},[64,2139,2140],{},"評価軸",[64,2142,2143],{},"詳細",[71,2145,2146,2154,2162,2169],{},[61,2147,2148,2151],{},[76,2149,2150],{},"技術",[76,2152,2153],{},"エラー率・量子ビット数・コヒーレンス時間",[61,2155,2156,2159],{},[76,2157,2158],{},"商用化",[76,2160,2161],{},"クラウドAPI・SDK・顧客導入実績",[61,2163,2164,2166],{},[76,2165,2016],{},[76,2167,2168],{},"アルゴリズム・ソフトウェアパートナー",[61,2170,2171,2173],{},[76,2172,1822],{},[76,2174,2175],{},"大手クラウド（AWS/Azure/GCP）への買収またはIPO",[48,2177],{},[20,2179,2181],{"id":2180},"代表投資案件-完全版テーブル","代表投資案件 完全版テーブル",[55,2183,2184,2211],{},[58,2185,2186],{},[61,2187,2188,2191,2194,2197,2200,2203,2205,2208],{},[64,2189,2190],{},"VC",[64,2192,2193],{},"投資先",[64,2195,2196],{},"投資年",[64,2198,2199],{},"ラウンド",[64,2201,2202],{},"金額（$M）",[64,2204,1527],{},[64,2206,2207],{},"ステージ",[64,2209,2210],{},"地理",[71,2212,2213,2237,2259,2280,2300,2323,2344,2363,2385,2406,2429,2452,2472,2495,2518,2539],{},[61,2214,2215,2218,2221,2224,2226,2229,2231,2234],{},[76,2216,2217],{},"Sequoia",[76,2219,2220],{},"Quantum Circuits",[76,2222,2223],{},"2024",[76,2225,1653],{},[76,2227,2228],{},">60",[76,2230,1712],{},[76,2232,2233],{},"B",[76,2235,2236],{},"米国",[61,2238,2239,2241,2244,2247,2249,2252,2254,2256],{},[76,2240,2217],{},[76,2242,2243],{},"InCore Semiconductors",[76,2245,2246],{},"2023",[76,2248,1552],{},[76,2250,2251],{},"3",[76,2253,1579],{},[76,2255,1552],{},[76,2257,2258],{},"インド",[61,2260,2261,2263,2266,2268,2270,2273,2276,2278],{},[76,2262,2217],{},[76,2264,2265],{},"Mindgrove Technologies",[76,2267,2246],{},[76,2269,1552],{},[76,2271,2272],{},"2.3",[76,2274,2275],{},"SoC",[76,2277,1552],{},[76,2279,2258],{},[61,2281,2282,2284,2287,2289,2291,2294,2296,2298],{},[76,2283,2217],{},[76,2285,2286],{},"Ricursive Intelligence",[76,2288,924],{},[76,2290,1552],{},[76,2292,2293],{},"35",[76,2295,1546],{},[76,2297,1552],{},[76,2299,2236],{},[61,2301,2302,2305,2308,2311,2313,2316,2318,2321],{},[76,2303,2304],{},"Sequoia（関連）",[76,2306,2307],{},"Ayar Labs",[76,2309,2310],{},"2026",[76,2312,1636],{},[76,2314,2315],{},"500",[76,2317,1630],{},[76,2319,2320],{},"E",[76,2322,2236],{},[61,2324,2325,2328,2331,2333,2335,2338,2340,2342],{},[76,2326,2327],{},"a16z",[76,2329,2330],{},"Unconventional AI",[76,2332,924],{},[76,2334,1552],{},[76,2336,2337],{},"475",[76,2339,1663],{},[76,2341,1552],{},[76,2343,2236],{},[61,2345,2346,2348,2351,2353,2355,2357,2359,2361],{},[76,2347,2327],{},[76,2349,2350],{},"Nexthop AI",[76,2352,2310],{},[76,2354,1653],{},[76,2356,2315],{},[76,2358,1647],{},[76,2360,2233],{},[76,2362,2236],{},[61,2364,2365,2368,2371,2373,2375,2378,2380,2383],{},[76,2366,2367],{},"Kleiner Perkins",[76,2369,2370],{},"Akeana",[76,2372,2223],{},[76,2374,1585],{},[76,2376,2377],{},">100",[76,2379,1579],{},[76,2381,2382],{},"Early",[76,2384,2236],{},[61,2386,2387,2389,2392,2394,2396,2399,2401,2404],{},[76,2388,2367],{},[76,2390,2391],{},"Retym",[76,2393,924],{},[76,2395,1619],{},[76,2397,2398],{},"75",[76,2400,1613],{},[76,2402,2403],{},"D",[76,2405,2236],{},[61,2407,2408,2411,2414,2416,2418,2421,2424,2427],{},[76,2409,2410],{},"Bessemer",[76,2412,2413],{},"ChipAgents",[76,2415,924],{},[76,2417,1568],{},[76,2419,2420],{},"21",[76,2422,2423],{},"EDA自動化",[76,2425,2426],{},"A",[76,2428,2236],{},[61,2430,2431,2433,2436,2439,2441,2444,2446,2449],{},[76,2432,2410],{},[76,2434,2435],{},"Kandou",[76,2437,2438],{},"2020",[76,2440,1602],{},[76,2442,2443],{},"92.3",[76,2445,1596],{},[76,2447,2448],{},"C",[76,2450,2451],{},"スイス",[61,2453,2454,2456,2459,2461,2463,2466,2468,2470],{},[76,2455,2410],{},[76,2457,2458],{},"Rigetti Computing",[76,2460,2438],{},[76,2462,1602],{},[76,2464,2465],{},"79",[76,2467,1712],{},[76,2469,2448],{},[76,2471,2236],{},[61,2473,2474,2476,2479,2482,2485,2487,2490,2492],{},[76,2475,2410],{},[76,2477,2478],{},"Habana Labs",[76,2480,2481],{},"2018",[76,2483,2484],{},"Series B（参加）",[76,2486,2398],{},[76,2488,2489],{},"AIアクセラレータ",[76,2491,2233],{},[76,2493,2494],{},"イスラエル",[61,2496,2497,2500,2503,2506,2508,2511,2514,2516],{},[76,2498,2499],{},"SoftBank",[76,2501,2502],{},"Fungible",[76,2504,2505],{},"2019",[76,2507,1602],{},[76,2509,2510],{},"200",[76,2512,2513],{},"DPU",[76,2515,2448],{},[76,2517,2236],{},[61,2519,2520,2523,2525,2527,2529,2531,2534,2536],{},[76,2521,2522],{},"SoftBank Group",[76,2524,2066],{},[76,2526,2223],{},[76,2528,1685],{},[76,2530,1688],{},[76,2532,2533],{},"AIチップ",[76,2535,1685],{},[76,2537,2538],{},"英国",[61,2540,2541,2543,2545,2547,2549,2552,2555,2557],{},[76,2542,2522],{},[76,2544,2077],{},[76,2546,924],{},[76,2548,1685],{},[76,2550,2551],{},"6,500",[76,2553,2554],{},"Arm CPU",[76,2556,1685],{},[76,2558,2236],{},[48,2560],{},[20,2562,2564],{"id":2563},"zyl0の視点","ZYL0の視点",[702,2566,2567,2572,2575,2578],{},[29,2568,2569],{},[36,2570,2571],{},"「大型ラウンド＝確実性が高い」は誤解だ",[29,2573,2574],{},"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」のだ。",[29,2576,2577],{},"技術観点で個人的に最も興味深いのはCPOだ。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。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。",[29,2579,2580],{},"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する。",[11,2582,2583,2587,2591,2595,2598,2601,2608,2610,2614,2634,2636,2640,2840,2842,2846,2850,2853,2857,2860,2867,2871,2874,2877,2882,2936,2938,2942,2946,2949,2955,2959,2962,2965,2971,2975,2978,2981,2985,2988,2995,2999,3002,3005,3010,3061,3063,3067,3071,3074,3088,3095,3099,3102,3105,3125,3132,3138,3142,3187,3194,3196,3200,3204,3210,3216,3220,3223,3227,3230,3235,3278,3280,3284,3578,3580,3584],{"lang":717},[15,2584,2586],{"id":2585},"why-design-data-movement-packaging-attract-vc-capital","Why Design, Data Movement & Packaging Attract VC Capital",[20,2588,2590],{"id":2589},"series-technical-deep-dive-and-case-studies","— Series②: Technical Deep Dive and Case Studies",[20,2592,2594],{"id":2593},"introduction","Introduction",[29,2596,2597],{},"Part 2 goes deeper. As established in Part 1, top-tier VCs are not betting on compute itself — they're betting on the constraints that emerge around compute.",[29,2599,2600],{},"The logic is straightforward: as GPU performance scales exponentially, the surrounding layers become the limiting factor. Connecting 10,000 GPUs requires switches and interconnects. Exceeding the limits of individual chips requires packaging (CPO/Chiplet). Growing design complexity requires EDA and verification automation.",[29,2602,2603,2604,2607],{},"Why is large capital flowing into each of these areas ",[712,2605,2606],{},"right now","? We read the representative deals as case studies.",[48,2609],{},[20,2611,2613],{"id":2612},"executive-summary","Executive Summary",[303,2615,2616,2622,2628],{},[306,2617,2618,2621],{},[36,2619,2620],{},"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.",[306,2623,2624,2627],{},[36,2625,2626],{},"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.",[306,2629,2630,2633],{},[36,2631,2632],{},"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.",[48,2635],{},[20,2637,2639],{"id":2638},"technology-stack-investment-mapping","Technology Stack × Investment Mapping",[55,2641,2642,2661],{},[58,2643,2644],{},[61,2645,2646,2649,2652,2655,2658],{},[64,2647,2648],{},"Tech Domain",[64,2650,2651],{},"Representative Co. (VC)",[64,2653,2654],{},"Round",[64,2656,2657],{},"Amount",[64,2659,2660],{},"Role",[71,2662,2663,2678,2692,2707,2721,2736,2751,2766,2781,2796,2811,2826],{},[61,2664,2665,2668,2671,2673,2675],{},[76,2666,2667],{},"Design AI/EDA",[76,2669,2670],{},"Ricursive (Sequoia)",[76,2672,1552],{},[76,2674,1555],{},[76,2676,2677],{},"Accelerate chip design cycles",[61,2679,2680,2682,2685,2687,2689],{},[76,2681,2667],{},[76,2683,2684],{},"ChipAgents (Bessemer)",[76,2686,1568],{},[76,2688,1571],{},[76,2690,2691],{},"Agentic AI for design/verification",[61,2693,2694,2696,2699,2702,2704],{},[76,2695,1579],{},[76,2697,2698],{},"Akeana (Kleiner Perkins)",[76,2700,2701],{},"Cumulative",[76,2703,1588],{},[76,2705,2706],{},"Open CPU instruction set",[61,2708,2709,2711,2714,2716,2718],{},[76,2710,1596],{},[76,2712,2713],{},"Kandou (Bessemer)",[76,2715,1602],{},[76,2717,1605],{},[76,2719,2720],{},"High-speed chip-to-chip PHY",[61,2722,2723,2726,2729,2731,2733],{},[76,2724,2725],{},"Coherent DSP",[76,2727,2728],{},"Retym (Kleiner Perkins)",[76,2730,1619],{},[76,2732,1622],{},[76,2734,2735],{},"Signal processing for optical links",[61,2737,2738,2741,2744,2746,2748],{},[76,2739,2740],{},"CPO/Optical I/O",[76,2742,2743],{},"Ayar Labs (Sequoia related)",[76,2745,1636],{},[76,2747,1639],{},[76,2749,2750],{},"Silicon photonics to bypass copper limits",[61,2752,2753,2756,2759,2761,2763],{},[76,2754,2755],{},"AI Switch",[76,2757,2758],{},"Nexthop AI (a16z)",[76,2760,1653],{},[76,2762,1639],{},[76,2764,2765],{},"Networking for large-scale AI clusters",[61,2767,2768,2771,2774,2776,2778],{},[76,2769,2770],{},"Analog compute",[76,2772,2773],{},"Unconventional AI (a16z)",[76,2775,1552],{},[76,2777,1671],{},[76,2779,2780],{},"Probabilistic computing for power efficiency",[61,2782,2783,2786,2789,2791,2793],{},[76,2784,2785],{},"AI chip (M&A)",[76,2787,2788],{},"Graphcore (SoftBank)",[76,2790,1685],{},[76,2792,1140],{},[76,2794,2795],{},"UK graph processor",[61,2797,2798,2801,2804,2806,2808],{},[76,2799,2800],{},"CPU/Arm (M&A)",[76,2802,2803],{},"Ampere (SoftBank)",[76,2805,1685],{},[76,2807,1704],{},[76,2809,2810],{},"Arm-based cloud server CPU",[61,2812,2813,2816,2819,2821,2823],{},[76,2814,2815],{},"Quantum",[76,2817,2818],{},"Rigetti (Bessemer)",[76,2820,1602],{},[76,2822,1720],{},[76,2824,2825],{},"Quantum cloud + hardware",[61,2827,2828,2830,2833,2835,2837],{},[76,2829,2815],{},[76,2831,2832],{},"Quantum Circuits (Sequoia)",[76,2834,1653],{},[76,2836,1735],{},[76,2838,2839],{},"Fault-tolerant quantum",[48,2841],{},[20,2843,2845],{"id":2844},"domain-designipeda-betting-on-development-lead-time-reduction","Domain①: Design/IP/EDA — Betting on Development Lead Time Reduction",[123,2847,2849],{"id":2848},"why-vcs-are-entering-design-automation-now","Why VCs Are Entering Design Automation Now",[29,2851,2852],{},"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.",[123,2854,2856],{"id":2855},"case-study-ricursive-intelligence-35m-2025-seed","Case Study: Ricursive Intelligence ($35M, 2025 Seed)",[29,2858,2859],{},"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.",[29,2861,2862,2863,2866],{},"The key advantage: ",[36,2864,2865],{},"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).\"",[123,2868,2870],{"id":2869},"case-study-chipagents-21m-2025-series-a","Case Study: ChipAgents ($21M, 2025 Series A)",[29,2872,2873],{},"Bessemer-led, ChipAgents automates the verification and debug loop using agentic AI.",[29,2875,2876],{},"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.",[29,2878,2879],{},[36,2880,2881],{},"EDA Automation Investment Logic:",[55,2883,2884,2894],{},[58,2885,2886],{},[61,2887,2888,2891],{},[64,2889,2890],{},"Dimension",[64,2892,2893],{},"Detail",[71,2895,2896,2904,2912,2920,2928],{},[61,2897,2898,2901],{},[76,2899,2900],{},"Market size",[76,2902,2903],{},"Global chip designers × design cycle cost reduction",[61,2905,2906,2909],{},[76,2907,2908],{},"Manufacturing risk",[76,2910,2911],{},"Low (primarily software)",[61,2913,2914,2917],{},[76,2915,2916],{},"Competitive risk",[76,2918,2919],{},"M&A target for EDA incumbents (Synopsys/Cadence)",[61,2921,2922,2925],{},[76,2923,2924],{},"Exit paths",[76,2926,2927],{},"M&A (EDA giants or chip design companies) or IPO",[61,2929,2930,2933],{},[76,2931,2932],{},"Timeline",[76,2934,2935],{},"Relatively short (2–5 years)",[48,2937],{},[20,2939,2941],{"id":2940},"domain-interconnectdata-movement-connecting-gpus-is-the-main-arena","Domain②: Interconnect/Data Movement — \"Connecting GPUs\" is the Main Arena",[123,2943,2945],{"id":2944},"the-copper-bottleneck-and-the-rise-of-silicon-photonics","The Copper Bottleneck and the Rise of Silicon Photonics",[29,2947,2948],{},"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.",[146,2950,2953],{"className":2951,"code":2952,"language":151},[149],"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",[153,2954,2952],{"__ignoreMap":155},[123,2956,2958],{"id":2957},"case-study-ayar-labs-500m-2026-series-e","Case Study: Ayar Labs ($500M, 2026 Series E)",[29,2960,2961],{},"This round, valuing Ayar Labs at $3.75B, is one of the largest fundraises in the CPO space.",[29,2963,2964],{},"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.",[29,2966,2967,2970],{},[36,2968,2969],{},"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.",[123,2972,2974],{"id":2973},"case-study-retym-75m-2025-series-d","Case Study: Retym ($75M, 2025 Series D)",[29,2976,2977],{},"Kleiner Perkins-backed Retym specializes in coherent DSP optimized for cloud and AI workloads.",[29,2979,2980],{},"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.",[123,2982,2984],{"id":2983},"case-study-kandou-923m-2020-series-c","Case Study: Kandou ($92.3M, 2020 Series C)",[29,2986,2987],{},"Bessemer-led, Kandou's Matterhorn IP has demonstrated deployment in USB4 and Thunderbolt implementations.",[29,2989,2990,2991,2994],{},"The strategy rides the ",[36,2992,2993],{},"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.",[123,2996,2998],{"id":2997},"case-study-nexthop-ai-500m-2026-series-b","Case Study: Nexthop AI ($500M, 2026 Series B)",[29,3000,3001],{},"a16z's $500M investment in Nexthop AI frames AI-era Ethernet switching as the next infrastructure layer.",[29,3003,3004],{},"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.",[29,3006,3007],{},[36,3008,3009],{},"Interconnect Investment Common Logic:",[55,3011,3012,3020],{},[58,3013,3014],{},[61,3015,3016,3018],{},[64,3017,2890],{},[64,3019,2893],{},[71,3021,3022,3030,3038,3046,3054],{},[61,3023,3024,3027],{},[76,3025,3026],{},"Technical driver",[76,3028,3029],{},"GPU/accelerator scaling makes I/O the bottleneck",[61,3031,3032,3035],{},[76,3033,3034],{},"Customer base",[76,3036,3037],{},"Hyperscalers (Google/AWS/Microsoft/Meta)",[61,3039,3040,3043],{},[76,3041,3042],{},"Capital intensity",[76,3044,3045],{},"Medium to high (manufacturing qualification required)",[61,3047,3048,3051],{},[76,3049,3050],{},"Risks",[76,3052,3053],{},"Production yield, supplier qualification, geopolitics",[61,3055,3056,3058],{},[76,3057,2924],{},[76,3059,3060],{},"M&A (major chip companies) or IPO",[48,3062],{},[20,3064,3066],{"id":3065},"domain-ai-accelerators-new-compute-betting-on-computing-paradigms","Domain③: AI Accelerators / New Compute — Betting on Computing Paradigms",[123,3068,3070],{"id":3069},"two-approaches-mega-seed-vs-vertical-integration","Two Approaches: Mega-Seed vs. Vertical Integration",[29,3072,3073],{},"Two distinct approaches exist for the \"change the computing paradigm\" thesis:",[648,3075,3076,3082],{},[306,3077,3078,3081],{},[36,3079,3080],{},"Mega-seed in a startup"," (a16z × Unconventional AI): High risk, but success redefines the market.",[306,3083,3084,3087],{},[36,3085,3086],{},"Acquisition-based vertical integration"," (SoftBank × Graphcore/Ampere): Lower technology realization risk, but integration cost and market timing risk remain.",[29,3089,3090,3091,3094],{},"Both are ",[36,3092,3093],{},"ecosystem positioning plays"," — not simply chip investments.",[123,3096,3098],{"id":3097},"case-study-unconventional-ai-475m-2025-seed","Case Study: Unconventional AI ($475M, 2025 Seed)",[29,3100,3101],{},"One of the largest seed rounds in semiconductor investment history, co-led by a16z.",[29,3103,3104],{},"The analog/mixed-signal approach for probabilistic computing differs fundamentally from current digital CMOS architectures. Key challenges:",[303,3106,3107,3113,3119],{},[306,3108,3109,3112],{},[36,3110,3111],{},"Manufacturing reproducibility",": Analog circuits are sensitive to process variation",[306,3114,3115,3118],{},[36,3116,3117],{},"Toolchain",": Design and verification tools are immature",[306,3120,3121,3124],{},[36,3122,3123],{},"Ecosystem",": Software/framework adaptation required",[29,3126,3127,3128,3131],{},"But if successful, the potential for ",[36,3129,3130],{},"orders-of-magnitude power efficiency gains"," represents an overwhelming competitive advantage over current GPU architectures.",[29,3133,3134,3137],{},[36,3135,3136],{},"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.",[123,3139,3141],{"id":3140},"softbanks-vertical-integration-arm-graphcore-ampere","SoftBank's Vertical Integration: Arm × Graphcore × Ampere",[55,3143,3144,3155],{},[58,3145,3146],{},[61,3147,3148,3150,3152],{},[64,3149,1042],{},[64,3151,2660],{},[64,3153,3154],{},"SoftBank Acquisition",[71,3156,3157,3167,3177],{},[61,3158,3159,3161,3164],{},[76,3160,2055],{},[76,3162,3163],{},"CPU IP, ISA standard, ecosystem core",[76,3165,3166],{},"Acquired 2016 ($32B)",[61,3168,3169,3171,3174],{},[76,3170,2066],{},[76,3172,3173],{},"Graph-optimized AI processing",[76,3175,3176],{},"Acquired 2024",[61,3178,3179,3181,3184],{},[76,3180,2077],{},[76,3182,3183],{},"Arm-based cloud server optimization",[76,3185,3186],{},"$6.5B acquisition announced 2025",[29,3188,3189,3190,3193],{},"This is a ",[36,3191,3192],{},"business strategy",", not a fund strategy. Evaluated on group synergy IRR, not VC fund returns.",[48,3195],{},[20,3197,3199],{"id":3198},"domain-quantum-long-term-but-capital-is-returning","Domain④: Quantum — Long-Term, But Capital Is Returning",[123,3201,3203],{"id":3202},"the-shift-to-commercialization-phase-quantum-investment","The Shift to \"Commercialization Phase\" Quantum Investment",[29,3205,3206,3207,825],{},"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 ",[36,3208,3209],{},"platform competition + cloud delivery",[29,3211,3212,3213],{},"The evaluation shift: from \"can quantum achieve quantum advantage?\" to ",[36,3214,3215],{},"\"can the company commercialize error correction, cloud delivery, and software stacks as an integrated system?\"",[123,3217,3219],{"id":3218},"case-study-quantum-circuits-series-b-60m-2024","Case Study: Quantum Circuits (Series B >$60M, 2024)",[29,3221,3222],{},"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.",[123,3224,3226],{"id":3225},"case-study-rigetti-computing-79m-2020-series-c","Case Study: Rigetti Computing ($79M, 2020 Series C)",[29,3228,3229],{},"Bessemer-led, with Quantum Cloud Services as the vertical integration model. Subsequently NASDAQ-listed via SPAC (2022). Having a cloud API entry point maintains enterprise access even before hardware reaches quantum advantage — a critical commercialization design choice.",[29,3231,3232],{},[36,3233,3234],{},"Quantum Investment Evaluation Criteria (Current):",[55,3236,3237,3246],{},[58,3238,3239],{},[61,3240,3241,3244],{},[64,3242,3243],{},"Axis",[64,3245,2893],{},[71,3247,3248,3255,3263,3270],{},[61,3249,3250,3252],{},[76,3251,1064],{},[76,3253,3254],{},"Error rates, qubit count, coherence time",[61,3256,3257,3260],{},[76,3258,3259],{},"Commercialization",[76,3261,3262],{},"Cloud API, SDK, customer deployments",[61,3264,3265,3267],{},[76,3266,3123],{},[76,3268,3269],{},"Algorithm and software partners",[61,3271,3272,3275],{},[76,3273,3274],{},"Exit",[76,3276,3277],{},"Acquisition by major cloud (AWS/Azure/GCP) or IPO",[48,3279],{},[20,3281,3283],{"id":3282},"complete-investment-reference-table","Complete Investment Reference Table",[55,3285,3286,3308],{},[58,3287,3288],{},[61,3289,3290,3292,3295,3298,3300,3303,3305],{},[64,3291,2190],{},[64,3293,3294],{},"Portfolio",[64,3296,3297],{},"Year",[64,3299,2654],{},[64,3301,3302],{},"Amount ($M)",[64,3304,2648],{},[64,3306,3307],{},"Geography",[71,3309,3310,3327,3344,3360,3376,3393,3410,3427,3443,3459,3476,3493,3509,3528,3544,3562],{},[61,3311,3312,3314,3316,3318,3320,3322,3324],{},[76,3313,2217],{},[76,3315,2220],{},[76,3317,2223],{},[76,3319,1653],{},[76,3321,2228],{},[76,3323,2815],{},[76,3325,3326],{},"US",[61,3328,3329,3331,3333,3335,3337,3339,3341],{},[76,3330,2217],{},[76,3332,2243],{},[76,3334,2246],{},[76,3336,1552],{},[76,3338,2251],{},[76,3340,1579],{},[76,3342,3343],{},"India",[61,3345,3346,3348,3350,3352,3354,3356,3358],{},[76,3347,2217],{},[76,3349,2265],{},[76,3351,2246],{},[76,3353,1552],{},[76,3355,2272],{},[76,3357,2275],{},[76,3359,3343],{},[61,3361,3362,3364,3366,3368,3370,3372,3374],{},[76,3363,2217],{},[76,3365,2286],{},[76,3367,924],{},[76,3369,1552],{},[76,3371,2293],{},[76,3373,2667],{},[76,3375,3326],{},[61,3377,3378,3381,3383,3385,3387,3389,3391],{},[76,3379,3380],{},"Sequoia (related)",[76,3382,2307],{},[76,3384,2310],{},[76,3386,1636],{},[76,3388,2315],{},[76,3390,2740],{},[76,3392,3326],{},[61,3394,3395,3397,3399,3401,3403,3405,3408],{},[76,3396,2327],{},[76,3398,2330],{},[76,3400,924],{},[76,3402,1552],{},[76,3404,2337],{},[76,3406,3407],{},"Analog new compute",[76,3409,3326],{},[61,3411,3412,3414,3416,3418,3420,3422,3425],{},[76,3413,2327],{},[76,3415,2350],{},[76,3417,2310],{},[76,3419,1653],{},[76,3421,2315],{},[76,3423,3424],{},"AI switch",[76,3426,3326],{},[61,3428,3429,3431,3433,3435,3437,3439,3441],{},[76,3430,2367],{},[76,3432,2370],{},[76,3434,2223],{},[76,3436,2701],{},[76,3438,2377],{},[76,3440,1579],{},[76,3442,3326],{},[61,3444,3445,3447,3449,3451,3453,3455,3457],{},[76,3446,2367],{},[76,3448,2391],{},[76,3450,924],{},[76,3452,1619],{},[76,3454,2398],{},[76,3456,2725],{},[76,3458,3326],{},[61,3460,3461,3463,3465,3467,3469,3471,3474],{},[76,3462,2410],{},[76,3464,2413],{},[76,3466,924],{},[76,3468,1568],{},[76,3470,2420],{},[76,3472,3473],{},"EDA automation",[76,3475,3326],{},[61,3477,3478,3480,3482,3484,3486,3488,3490],{},[76,3479,2410],{},[76,3481,2435],{},[76,3483,2438],{},[76,3485,1602],{},[76,3487,2443],{},[76,3489,1596],{},[76,3491,3492],{},"Switzerland",[61,3494,3495,3497,3499,3501,3503,3505,3507],{},[76,3496,2410],{},[76,3498,2458],{},[76,3500,2438],{},[76,3502,1602],{},[76,3504,2465],{},[76,3506,2815],{},[76,3508,3326],{},[61,3510,3511,3513,3515,3517,3520,3522,3525],{},[76,3512,2410],{},[76,3514,2478],{},[76,3516,2481],{},[76,3518,3519],{},"Series B (participant)",[76,3521,2398],{},[76,3523,3524],{},"AI accelerator",[76,3526,3527],{},"Israel",[61,3529,3530,3532,3534,3536,3538,3540,3542],{},[76,3531,2499],{},[76,3533,2502],{},[76,3535,2505],{},[76,3537,1602],{},[76,3539,2510],{},[76,3541,2513],{},[76,3543,3326],{},[61,3545,3546,3548,3550,3552,3554,3556,3559],{},[76,3547,2522],{},[76,3549,2066],{},[76,3551,2223],{},[76,3553,1685],{},[76,3555,1140],{},[76,3557,3558],{},"AI chip",[76,3560,3561],{},"UK",[61,3563,3564,3566,3568,3570,3572,3574,3576],{},[76,3565,2522],{},[76,3567,2077],{},[76,3569,924],{},[76,3571,1685],{},[76,3573,2551],{},[76,3575,2554],{},[76,3577,3326],{},[48,3579],{},[20,3581,3583],{"id":3582},"zyl0s-perspective","ZYL0's Perspective",[702,3585,3586,3591,3594,3597],{},[29,3587,3588],{},[36,3589,3590],{},"A large round does not mean high probability — it means high expected value",[29,3592,3593],{},"The $475M seed into Unconventional AI, and the $500M rounds into Nexthop and Ayar Labs, are not bets on certainty. They are bets where \"if wrong, we lose the check; if right, we create a new category worth hundreds of billions.\" The asymmetry justifies the size.",[29,3595,3596],{},"The 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 may be one of the few \"cutting-edge intersections\" where Japanese industry can participate at the frontier.",[29,3598,3599],{},"Part 3 (the final installment) synthesizes the strategic implications: how to construct a portfolio across these themes, and how to think about the risk-reward across short, medium, and long time horizons.",{"title":155,"searchDepth":1390,"depth":1390,"links":3601},[3602,3603,3604,3605,3606,3611,3618,3623,3628,3629,3630,3631,3632,3633,3634,3639,3646,3651,3656,3657],{"id":1474,"depth":1390,"text":1475},{"id":1478,"depth":1390,"text":1478},{"id":1492,"depth":1390,"text":1492},{"id":1517,"depth":1390,"text":1518},{"id":1743,"depth":1390,"text":1744,"children":3607},[3608,3609,3610],{"id":1747,"depth":1397,"text":1748},{"id":1754,"depth":1397,"text":1755},{"id":1768,"depth":1397,"text":1769},{"id":1838,"depth":1390,"text":1839,"children":3612},[3613,3614,3615,3616,3617],{"id":1842,"depth":1397,"text":1842},{"id":1854,"depth":1397,"text":1855},{"id":1870,"depth":1397,"text":1871},{"id":1880,"depth":1397,"text":1881},{"id":1890,"depth":1397,"text":1891},{"id":1958,"depth":1390,"text":1959,"children":3619},[3620,3621,3622],{"id":1962,"depth":1397,"text":1963},{"id":1990,"depth":1397,"text":1991},{"id":2029,"depth":1397,"text":2030},{"id":2091,"depth":1390,"text":2092,"children":3624},[3625,3626,3627],{"id":2095,"depth":1397,"text":2096},{"id":2109,"depth":1397,"text":2110},{"id":2119,"depth":1397,"text":2120},{"id":2180,"depth":1390,"text":2181},{"id":2563,"depth":1390,"text":2564},{"id":2589,"depth":1390,"text":2590},{"id":2593,"depth":1390,"text":2594},{"id":2612,"depth":1390,"text":2613},{"id":2638,"depth":1390,"text":2639},{"id":2844,"depth":1390,"text":2845,"children":3635},[3636,3637,3638],{"id":2848,"depth":1397,"text":2849},{"id":2855,"depth":1397,"text":2856},{"id":2869,"depth":1397,"text":2870},{"id":2940,"depth":1390,"text":2941,"children":3640},[3641,3642,3643,3644,3645],{"id":2944,"depth":1397,"text":2945},{"id":2957,"depth":1397,"text":2958},{"id":2973,"depth":1397,"text":2974},{"id":2983,"depth":1397,"text":2984},{"id":2997,"depth":1397,"text":2998},{"id":3065,"depth":1390,"text":3066,"children":3647},[3648,3649,3650],{"id":3069,"depth":1397,"text":3070},{"id":3097,"depth":1397,"text":3098},{"id":3140,"depth":1397,"text":3141},{"id":3198,"depth":1390,"text":3199,"children":3652},[3653,3654,3655],{"id":3202,"depth":1397,"text":3203},{"id":3218,"depth":1397,"text":3219},{"id":3225,"depth":1397,"text":3226},{"id":3282,"depth":1390,"text":3283},{"id":3582,"depth":1390,"text":3583},"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・量子にわたる代表案件を技術視点で深掘り。$500M超の大型ラウンドが示す投資ロジックを解読する。",{"date":3660,"image":3661,"alt":3662,"tags":3663,"tagsEn":3666,"published":1456},"11th Apr 2026","/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",[3664,3665,1447,1446,1450],"半導体","VC投資",[3667,3668,1452,738,3669],"Semiconductor","VC Investment","Startup","/blogs/7-semiconductor-vc-series-2",{"title":1463,"description":3658},"blogs/7-semiconductor-vc-series-2","-ywFilRUg1zg9yFBaMw2NC0snzsGm14MEamxZbdax9E",1775926672821]