[{"data":1,"prerenderedAt":1380},["ShallowReactive",2],{"content-/ai-bubble-gavin-baker":3,"all-pages-for-dir":1358,"related-/ai-bubble-gavin-baker":1359,"og-image-/ai-bubble-gavin-baker":1379},{"id":4,"title":5,"body":6,"category":1338,"concepts":1338,"description":1339,"extension":1340,"meta":1341,"navigation":352,"ogImage":1338,"path":1342,"project_name":1338,"published":1343,"publishedAt":1344,"seo":1345,"source":1338,"stem":1346,"tags":1347,"todo":1338,"unpublished":1343,"updatedAt":1338,"__hash__":1357},"pages/2025-10/2025-10-31/ai-bubble-gavin-baker.md","AIバブル論争に終止符？ Atrades CIOギャビン・ベイカー氏が語るAI投資の真実",{"type":7,"value":8,"toc":1319},"minimark",[9,13,17,21,44,47,50,54,56,59,63,66,70,75,90,95,100,105,109,114,128,133,152,157,245,249,254,268,273,288,293,400,404,409,424,429,446,451,462,467,657,661,666,674,679,694,698,703,717,722,735,740,751,756,767,772,868,873,877,882,898,903,967,970,975,993,998,1006,1010,1015,1020,1025,1167,1172,1186,1191,1203,1208,1219,1224,1235,1239,1244,1255,1260,1279,1284,1292,1294,1297,1300,1303,1306,1315],[10,11,5],"h1",{"id":12},"aiバブル論争に終止符-atrades-cioギャビンベイカー氏が語るai投資の真実",[14,15,16],"h2",{"id":16},"サマリー",[18,19,20],"h3",{"id":20},"今回のポイント3つ",[22,23,24,32,38],"ol",{},[25,26,27,31],"li",{},[28,29,30],"strong",{},"AIバブルは存在しない"," - 2000年のインターネットバブルでは97%の光ファイバーが未使用（ダークファイバー）だったのに対し、現在「ダークGPU」は存在せず、むしろGPUが溶けるほど稼働している。GPU投資のROICは約10ポイント上昇しており、投資リターンは極めて良好。",[25,33,34,37],{},[28,35,36],{},"推論（Reasoning）機能がゲームチェンジャー"," - 推論機能の登場により、フロンティアモデルの経済性が根本的に変化。ユーザーデータを活用した強化学習のフライホイールが機能し始め、OpenAI、Anthropic、XAIなどのビジネスモデルが持続可能に。",[25,39,40,43],{},[28,41,42],{},"粗利益率の低下は成功の証"," - AIビジネスはSaaS企業の80-90%と比べて構造的に粗利益率が低くなるが、これは失敗ではなく成功の証。MicrosoftやAdobeがクラウド移行で証明したように、マージン低下を恐れるべきではない。",[45,46],"hr",{},[14,48,49],{"id":49},"イベント概要",[51,52,53],"p",{},"A16Zのイベントにおいて、AtradesのマネージングパートナーでCIOのギャビン・ベイカー氏と、A16Zのジェネラルパートナーであるデイビッド・ジョージ氏が、現在のAI投資環境について対談を行った。",[45,55],{},[14,57,58],{"id":58},"議論の詳細",[18,60,62],{"id":61},"_0000-導入","00:00 - 導入",[51,64,65],{},"ベイカー氏は、AIに関する重要なニュースが出た際に「一体何が起こっているのか」を的確に説明する人物として知られており、今回のセッションでは「AIバブル」というタブーな質問から議論が始まった。",[18,67,69],{"id":68},"_0102-aiバブルは存在するのか","01:02 - AIバブルは存在するのか？",[51,71,72],{},[28,73,74],{},"インフラ投資の規模",[76,77,78,81,84,87],"ul",{},[25,79,80],{},"米国には現在約1兆ドルのデータセンターが存在",[25,82,83],{},"今後5年間で3〜4兆ドルを追加する計画",[25,85,86],{},"過去3年間で建設されたデータセンター容量は、インフレ調整後、40年かかった米国の州間高速道路システム全体を上回る",[25,88,89],{},"OpenAIだけで1兆ドル以上の契約を設定",[51,91,92],{},[28,93,94],{},"利用状況の爆発的成長",[76,96,97],{},[25,98,99],{},"Googleは過去17か月間で処理トークン量が150倍に増加",[51,101,102],{},[28,103,104],{},"ベイカー氏の結論：AIバブルには陥っていない",[18,106,108],{"id":107},"_0241-ダークファイバー-vs-ダークgpuは存在しない","02:41 - ダークファイバー vs. 「ダークGPUは存在しない」",[51,110,111],{},[28,112,113],{},"2000年のインターネットバブルとの比較",[76,115,116,119,125],{},[25,117,118],{},"2000年のバブルは「ダークファイバー」によって定義された",[25,120,121,122],{},"バブルのピーク時、米国に敷設されたファイバーの",[28,123,124],{},"97%が未使用",[25,126,127],{},"光学部品、スイッチ、ルーターがないため「点灯」されていなかった",[51,129,130],{},[28,131,132],{},"現在のAI投資の実態",[76,134,135,140,146],{},[25,136,137],{},[28,138,139],{},"ダークGPUは存在しない",[25,141,142,143],{},"技術論文によれば、トレーニング実行における最大の問題の1つは",[28,144,145],{},"GPUが溶けてしまうこと",[25,147,148,149],{},"GPU最大支出企業の投下資本利益率（ROIC）は、資本支出を増やし始めて以来",[28,150,151],{},"約10ポイント増加",[51,153,154],{},[28,155,156],{},"比較表：2000年バブル vs. 現在のAI投資",[158,159,160,176],"table",{},[161,162,163],"thead",{},[164,165,166,170,173],"tr",{},[167,168,169],"th",{},"項目",[167,171,172],{},"2000年インターネットバブル",[167,174,175],{},"2025年AI投資",[177,178,179,193,206,219,232],"tbody",{},[164,180,181,187,190],{},[182,183,184],"td",{},[28,185,186],{},"インフラ利用率",[182,188,189],{},"97%が未使用（ダークファイバー）",[182,191,192],{},"ほぼ100%稼働（GPUが溶けるレベル）",[164,194,195,200,203],{},[182,196,197],{},[28,198,199],{},"ROI",[182,201,202],{},"マイナス（資産価値急落）",[182,204,205],{},"+10ポイントROIC増加",[164,207,208,213,216],{},[182,209,210],{},[28,211,212],{},"投資主体",[182,214,215],{},"新興企業、投機的資本",[182,217,218],{},"Mag 7など世界最高の企業群",[164,220,221,226,229],{},[182,222,223],{},[28,224,225],{},"キャッシュフロー",[182,227,228],{},"脆弱",[182,230,231],{},"年間3,000億ドル、現金5,000億ドル保有",[164,233,234,239,242],{},[182,235,236],{},[28,237,238],{},"需要",[182,240,241],{},"見込みベース",[182,243,244],{},"実需（Googleのトークン処理150倍増）",[18,246,248],{"id":247},"_0507-aiのroi今回が異なる理由","05:07 - AIのROI：今回が異なる理由",[51,250,251],{},[28,252,253],{},"経済的な健全性",[76,255,256,259,262,265],{},[25,257,258],{},"GPU投資のROIはこれまで非常にポジティブ",[25,260,261],{},"投資主体は「世界史上最高の企業」群",[25,263,264],{},"合計で年間約3,000億ドルのフリーキャッシュフローを生成",[25,266,267],{},"バランスシートに5,000億ドルの現金を保有",[51,269,270],{},[28,271,272],{},"採用の容易さ",[76,274,275,278,285],{},[25,276,277],{},"インターネット時代：ウェブサイトとユーザーという二面的なネットワーク構築が必要",[25,279,280,281,284],{},"AI時代：APIやChatGPTなどのウェブサイトを通じて",[28,282,283],{},"即座に10億人規模への配布","が可能",[25,286,287],{},"クラウドコンピューティングとインターネットの上に構築されているため、導入障壁が低い",[51,289,290],{},[28,291,292],{},"技術採用の比較：インターネット時代 vs. AI時代",[294,295,300],"pre",{"className":296,"code":297,"language":298,"meta":299,"style":299},"language-mermaid shiki shiki-themes vitesse-light vitesse-light","graph TB\n    subgraph \"インターネット時代（1990-2000年代）\"\n        I1[ウェブサイト構築] --> I2[インフラ整備]\n        I2 --> I3[ユーザー獲得]\n        I3 --> I4[ネットワーク効果]\n        I4 -.->|数年かかる| I5[大規模展開]\n    end\n\n    subgraph \"AI時代（2020年代～）\"\n        A1[API提供] --> A2[既存インフラ活用\u003Cbr/>クラウド+インターネット]\n        A2 --> A3[即座に配布可能]\n        A3 -.->|数ヶ月で達成| A4[10億人規模]\n    end\n\n    style I5 fill:#FFC107\n    style A4 fill:#4CAF50\n","mermaid","",[301,302,303,311,317,323,329,335,341,347,354,360,366,372,378,383,388,394],"code",{"__ignoreMap":299},[304,305,308],"span",{"class":306,"line":307},"line",1,[304,309,310],{},"graph TB\n",[304,312,314],{"class":306,"line":313},2,[304,315,316],{},"    subgraph \"インターネット時代（1990-2000年代）\"\n",[304,318,320],{"class":306,"line":319},3,[304,321,322],{},"        I1[ウェブサイト構築] --> I2[インフラ整備]\n",[304,324,326],{"class":306,"line":325},4,[304,327,328],{},"        I2 --> I3[ユーザー獲得]\n",[304,330,332],{"class":306,"line":331},5,[304,333,334],{},"        I3 --> I4[ネットワーク効果]\n",[304,336,338],{"class":306,"line":337},6,[304,339,340],{},"        I4 -.->|数年かかる| I5[大規模展開]\n",[304,342,344],{"class":306,"line":343},7,[304,345,346],{},"    end\n",[304,348,350],{"class":306,"line":349},8,[304,351,353],{"emptyLinePlaceholder":352},true,"\n",[304,355,357],{"class":306,"line":356},9,[304,358,359],{},"    subgraph \"AI時代（2020年代～）\"\n",[304,361,363],{"class":306,"line":362},10,[304,364,365],{},"        A1[API提供] --> A2[既存インフラ活用\u003Cbr/>クラウド+インターネット]\n",[304,367,369],{"class":306,"line":368},11,[304,370,371],{},"        A2 --> A3[即座に配布可能]\n",[304,373,375],{"class":306,"line":374},12,[304,376,377],{},"        A3 -.->|数ヶ月で達成| A4[10億人規模]\n",[304,379,381],{"class":306,"line":380},13,[304,382,346],{},[304,384,386],{"class":306,"line":385},14,[304,387,353],{"emptyLinePlaceholder":352},[304,389,391],{"class":306,"line":390},15,[304,392,393],{},"    style I5 fill:#FFC107\n",[304,395,397],{"class":306,"line":396},16,[304,398,399],{},"    style A4 fill:#4CAF50\n",[18,401,403],{"id":402},"_0836-ラウンドトリッピング環流取引の実態","08:36 - ラウンドトリッピング（環流取引）の実態",[51,405,406],{},[28,407,408],{},"懸念は誇張されすぎている",[76,410,411,414,417],{},[25,412,413],{},"Nvidiaが顧客に出資し、その資金でNvidiaのチップを購入する取引は客観的に発生",[25,415,416],{},"しかし規模は非常に小さい",[25,418,419,420,423],{},"資金調達の必要性ではなく、",[28,421,422],{},"競争上の力学","が背景",[51,425,426],{},[28,427,428],{},"競合分析",[76,430,431,437,440,443],{},[25,432,433,434],{},"Nvidiaの最大の競合はAMDやIntelではなく",[28,435,436],{},"Google",[25,438,439],{},"GoogleのTPUチップはトレーニングにおいてNvidiaに対する唯一の代替手段",[25,441,442],{},"GoogleはDeepMindとGeminiを所有し、トラフィックベースでOpenAIやAnthropicより大きい可能性",[25,444,445],{},"AnthropicはAmazonとGoogleのキャプティブで、TPUとTraniumで実行",[51,447,448],{},[28,449,450],{},"Nvidiaの戦略的合理性",[76,452,453,456,459],{},[25,454,455],{},"GoogleがAnthropicを支援している以上、Nvidiaが競争上の理由から対応しないのは困難",[25,457,458],{},"ベイカー氏：「Nvidiaがとっているすべての行動は戦略的に完全に合理的」",[25,460,461],{},"Jensen Huang氏は「良い投資になる」と発言",[51,463,464],{},[28,465,466],{},"GPU市場の競合構造",[294,468,470],{"className":296,"code":469,"language":298,"meta":299,"style":299},"graph TB\n    subgraph \"AI企業\"\n        OpenAI[OpenAI]\n        Anthropic[Anthropic]\n        DeepMind[DeepMind/Gemini]\n    end\n\n    subgraph \"チップ提供\"\n        Nvidia[Nvidia GPU]\n        TPU[Google TPU]\n        Tranium[Amazon Tranium]\n    end\n\n    subgraph \"支援企業\"\n        Google[Google]\n        Amazon[Amazon]\n        Microsoft[Microsoft]\n    end\n\n    Google -->|支援| Anthropic\n    Google -->|所有| DeepMind\n    Google -->|開発| TPU\n    Amazon -->|支援| Anthropic\n    Amazon -->|開発| Tranium\n    Microsoft -->|支援| OpenAI\n\n    Nvidia -->|出資| OpenAI\n    Anthropic -->|使用| TPU\n    Anthropic -->|使用| Tranium\n    OpenAI -->|使用| Nvidia\n    DeepMind -->|使用| TPU\n\n    style Google fill:#ea4335\n    style Nvidia fill:#76b900\n    style Anthropic fill:#9b6dff\n",[301,471,472,476,481,486,491,496,500,504,509,514,519,524,528,532,537,542,547,553,558,563,569,575,581,587,593,599,604,610,616,622,628,634,639,645,651],{"__ignoreMap":299},[304,473,474],{"class":306,"line":307},[304,475,310],{},[304,477,478],{"class":306,"line":313},[304,479,480],{},"    subgraph \"AI企業\"\n",[304,482,483],{"class":306,"line":319},[304,484,485],{},"        OpenAI[OpenAI]\n",[304,487,488],{"class":306,"line":325},[304,489,490],{},"        Anthropic[Anthropic]\n",[304,492,493],{"class":306,"line":331},[304,494,495],{},"        DeepMind[DeepMind/Gemini]\n",[304,497,498],{"class":306,"line":337},[304,499,346],{},[304,501,502],{"class":306,"line":343},[304,503,353],{"emptyLinePlaceholder":352},[304,505,506],{"class":306,"line":349},[304,507,508],{},"    subgraph \"チップ提供\"\n",[304,510,511],{"class":306,"line":356},[304,512,513],{},"        Nvidia[Nvidia GPU]\n",[304,515,516],{"class":306,"line":362},[304,517,518],{},"        TPU[Google TPU]\n",[304,520,521],{"class":306,"line":368},[304,522,523],{},"        Tranium[Amazon Tranium]\n",[304,525,526],{"class":306,"line":374},[304,527,346],{},[304,529,530],{"class":306,"line":380},[304,531,353],{"emptyLinePlaceholder":352},[304,533,534],{"class":306,"line":385},[304,535,536],{},"    subgraph \"支援企業\"\n",[304,538,539],{"class":306,"line":390},[304,540,541],{},"        Google[Google]\n",[304,543,544],{"class":306,"line":396},[304,545,546],{},"        Amazon[Amazon]\n",[304,548,550],{"class":306,"line":549},17,[304,551,552],{},"        Microsoft[Microsoft]\n",[304,554,556],{"class":306,"line":555},18,[304,557,346],{},[304,559,561],{"class":306,"line":560},19,[304,562,353],{"emptyLinePlaceholder":352},[304,564,566],{"class":306,"line":565},20,[304,567,568],{},"    Google -->|支援| Anthropic\n",[304,570,572],{"class":306,"line":571},21,[304,573,574],{},"    Google -->|所有| DeepMind\n",[304,576,578],{"class":306,"line":577},22,[304,579,580],{},"    Google -->|開発| TPU\n",[304,582,584],{"class":306,"line":583},23,[304,585,586],{},"    Amazon -->|支援| Anthropic\n",[304,588,590],{"class":306,"line":589},24,[304,591,592],{},"    Amazon -->|開発| Tranium\n",[304,594,596],{"class":306,"line":595},25,[304,597,598],{},"    Microsoft -->|支援| OpenAI\n",[304,600,602],{"class":306,"line":601},26,[304,603,353],{"emptyLinePlaceholder":352},[304,605,607],{"class":306,"line":606},27,[304,608,609],{},"    Nvidia -->|出資| OpenAI\n",[304,611,613],{"class":306,"line":612},28,[304,614,615],{},"    Anthropic -->|使用| TPU\n",[304,617,619],{"class":306,"line":618},29,[304,620,621],{},"    Anthropic -->|使用| Tranium\n",[304,623,625],{"class":306,"line":624},30,[304,626,627],{},"    OpenAI -->|使用| Nvidia\n",[304,629,631],{"class":306,"line":630},31,[304,632,633],{},"    DeepMind -->|使用| TPU\n",[304,635,637],{"class":306,"line":636},32,[304,638,353],{"emptyLinePlaceholder":352},[304,640,642],{"class":306,"line":641},33,[304,643,644],{},"    style Google fill:#ea4335\n",[304,646,648],{"class":306,"line":647},34,[304,649,650],{},"    style Nvidia fill:#76b900\n",[304,652,654],{"class":306,"line":653},35,[304,655,656],{},"    style Anthropic fill:#9b6dff\n",[18,658,660],{"id":659},"_1222-アプリケーション層まだ非常に初期段階","12:22 - アプリケーション層：まだ非常に初期段階",[51,662,663],{},[28,664,665],{},"謙虚さの必要性",[76,667,668,671],{},[25,669,670],{},"ChatGPTがAIにとってのNetscape Navigatorなら、Googleはまだ設立されておらず、マーク・ザッカーバーグは中学生だった時期",[25,672,673],{},"新しいテクノロジーの波の初期には、インフラストラクチャ層が安全な場所となることが多い",[51,675,676],{},[28,677,678],{},"持続的イノベーション vs. 破壊的イノベーション",[76,680,681,688,691],{},[25,682,683,684,687],{},"AIは破壊的ではなく",[28,685,686],{},"持続的なイノベーション","になる可能性",[25,689,690],{},"最大のテック企業（Mag 7）がデータ、資本、流通という「原材料」を豊富に保有",[25,692,693],{},"ただし実行を怠れば、IBMのような運命を辿る可能性もあり、危機的状況でもある",[18,695,697],{"id":696},"_1524-マージンとスケーリング法則","15:24 - マージンとスケーリング法則",[51,699,700],{},[28,701,702],{},"構造的な粗利益率の低下",[76,704,705,708,711,714],{},[25,706,707],{},"スケーリング法則（Richard Suttonの「Bitter Lesson」）により、AIはよりコンピューティング集約的",[25,709,710],{},"2021-2022年のSaaS企業：80〜90%の粗利益率",[25,712,713],{},"フロンティアラボ：そのようなマージンを達成するには長い時間が必要",[25,715,716],{},"ただし、営業費用（Opex）は低くなる可能性",[51,718,719],{},[28,720,721],{},"アプリケーションSaaSの未来",[76,723,724,727,732],{},[25,725,726],{},"小売業者がAmazonとの取引で犯した過ち：低マージンを恐れること",[25,728,729],{},[28,730,731],{},"粗利益率の低下は不可避であり、成功の証",[25,733,734],{},"MicrosoftとAdobeの成功例：クラウド移行でマージン低下に対処し、株価も好調",[51,736,737],{},[28,738,739],{},"レガシーSaaS企業の優位性",[76,741,742,745,748],{},[25,743,744],{},"既存の収益性の高いビジネスを持つ",[25,746,747],{},"新しいAI製品を損益分岐点で運営可能",[25,749,750],{},"リーダーに追いつくことができる",[51,752,753],{},[28,754,755],{},"「名誉のしるし」としての低マージン",[76,757,758,761,764],{},[25,759,760],{},"低マージンは、実際にAIツールが利用されていることの証",[25,762,763],{},"粗利益率82%のAI企業：実際には使われていない可能性",[25,765,766],{},"10ドルの収益で90%マージン vs. 50ドルの収益で60%マージン：どちらを選ぶ？",[51,768,769],{},[28,770,771],{},"マージンの真実：収益性の比較",[158,773,774,793],{},[161,775,776],{},[164,777,778,781,784,787,790],{},[167,779,780],{},"シナリオ",[167,782,783],{},"売上",[167,785,786],{},"粗利益率",[167,788,789],{},"粗利益額",[167,791,792],{},"実態",[177,794,795,814,833,851],{},[164,796,797,802,805,808,811],{},[182,798,799],{},[28,800,801],{},"高マージン企業",[182,803,804],{},"$10",[182,806,807],{},"90%",[182,809,810],{},"$9",[182,812,813],{},"AI機能が使われていない可能性",[164,815,816,821,824,827,830],{},[182,817,818],{},[28,819,820],{},"低マージン企業",[182,822,823],{},"$50",[182,825,826],{},"60%",[182,828,829],{},"$30",[182,831,832],{},"AIが実際に活用されている証",[164,834,835,840,843,846,848],{},[182,836,837],{},[28,838,839],{},"レガシーSaaS",[182,841,842],{},"-",[182,844,845],{},"80-90%",[182,847,842],{},[182,849,850],{},"従来型ビジネスモデル",[164,852,853,858,860,863,865],{},[182,854,855],{},[28,856,857],{},"AI SaaS",[182,859,842],{},[182,861,862],{},"60-70%",[182,864,842],{},[182,866,867],{},"成功している新モデル",[51,869,870],{},[28,871,872],{},"粗利益額では3倍以上の差（$9 vs $30）- どちらがビジネスとして成功か？",[18,874,876],{"id":875},"_2326-推論モデルとフライホイール効果","23:26 - 推論モデルとフライホイール効果",[51,878,879],{},[28,880,881],{},"推論機能のインパクト",[76,883,884,887,890],{},[25,885,886],{},"フロンティアモデルの経済性を根本的に変化",[25,888,889],{},"推論登場前：ユニークなデータとインターネット規模の流通を持たないフロンティアモデルは「史上最も急速に減価償却される資産」",[25,891,892,893],{},"推論登場後：ユーザーが多いことで後処理の強化学習における「フライホイール」が機能\n",[76,894,895],{},[25,896,897],{},"良い製品 → 多くのユーザー → アルゴリズム改善 → 製品向上 → 好循環",[51,899,900],{},[28,901,902],{},"AIフライホイール効果",[294,904,906],{"className":296,"code":905,"language":298,"meta":299,"style":299},"graph LR\n    A[良い製品\u003Cbr/>Quality Product] --> B[多くのユーザー\u003Cbr/>More Users]\n    B --> C[大量のデータ\u003Cbr/>Data Collection]\n    C --> D[強化学習\u003Cbr/>Reinforcement Learning]\n    D --> E[アルゴリズム改善\u003Cbr/>Algorithm Improvement]\n    E --> A\n\n    style A fill:#4CAF50\n    style B fill:#2196F3\n    style C fill:#FF9800\n    style D fill:#9C27B0\n    style E fill:#F44336\n",[301,907,908,913,918,923,928,933,938,942,947,952,957,962],{"__ignoreMap":299},[304,909,910],{"class":306,"line":307},[304,911,912],{},"graph LR\n",[304,914,915],{"class":306,"line":313},[304,916,917],{},"    A[良い製品\u003Cbr/>Quality Product] --> B[多くのユーザー\u003Cbr/>More Users]\n",[304,919,920],{"class":306,"line":319},[304,921,922],{},"    B --> C[大量のデータ\u003Cbr/>Data Collection]\n",[304,924,925],{"class":306,"line":325},[304,926,927],{},"    C --> D[強化学習\u003Cbr/>Reinforcement Learning]\n",[304,929,930],{"class":306,"line":331},[304,931,932],{},"    D --> E[アルゴリズム改善\u003Cbr/>Algorithm Improvement]\n",[304,934,935],{"class":306,"line":337},[304,936,937],{},"    E --> A\n",[304,939,940],{"class":306,"line":343},[304,941,353],{"emptyLinePlaceholder":352},[304,943,944],{"class":306,"line":349},[304,945,946],{},"    style A fill:#4CAF50\n",[304,948,949],{"class":306,"line":356},[304,950,951],{},"    style B fill:#2196F3\n",[304,953,954],{"class":306,"line":362},[304,955,956],{},"    style C fill:#FF9800\n",[304,958,959],{"class":306,"line":368},[304,960,961],{},"    style D fill:#9C27B0\n",[304,963,964],{"class":306,"line":374},[304,965,966],{},"    style E fill:#F44336\n",[51,968,969],{},"この循環により、OpenAI、Anthropic、XAIなどのフロンティアモデル企業のビジネスモデルが持続可能になる。",[51,971,972],{},[28,973,974],{},"GPT-5に関する誤解",[76,976,977,980,987,990],{},[25,978,979],{},"「GPT-5がスケーリング法則の終わり」という主張にベイカー氏は強く反発",[25,981,982,983,986],{},"GPT-5は",[28,984,985],{},"より小さなモデル","として設計",[25,988,989],{},"OpenAIとMicrosoftがより経済的に実行できることが目的",[25,991,992],{},"スケーリング法則の終焉ではない",[51,994,995],{},[28,996,997],{},"コンシューマー側の競争",[76,999,1000,1003],{},[25,1001,1002],{},"AI企業が独自のAIブラウザを立ち上げたことは、50億人のユーザーを持つChromeを所有するGoogleにとって好機",[25,1004,1005],{},"Googleは慎重に後を追う戦略（Google Buzzの教訓）",[18,1007,1009],{"id":1008},"_2718-ビジネスモデルの未来","27:18 - ビジネスモデルの未来",[51,1011,1012],{},[28,1013,1014],{},"プラットフォーム移行とビジネスモデル移行",[76,1016,1017],{},[25,1018,1019],{},"大きな市場機会は両者が同時に起こるときに生まれる",[51,1021,1022],{},[28,1023,1024],{},"ビジネスモデルの進化",[294,1026,1028],{"className":296,"code":1027,"language":298,"meta":299,"style":299},"graph TD\n    subgraph \"従来のSaaS\"\n        A1[月額固定課金\u003Cbr/>Seat-based Pricing]\n        A2[80-90%粗利益率]\n    end\n\n    subgraph \"AI時代のビジネスモデル\"\n        B1[成果ベース課金\u003Cbr/>Outcome-based]\n        B2[消費量ベース課金\u003Cbr/>Usage-based]\n        B3[アフィリエイト料\u003Cbr/>Affiliate Revenue]\n    end\n\n    subgraph \"具体例\"\n        C1[カスタマーサポート\u003Cbr/>タスク解決数で課金]\n        C2[コーディング\u003Cbr/>消費量で課金]\n        C3[AI検索・推奨\u003Cbr/>成果報酬]\n    end\n\n    A1 -.->|進化| B1\n    A1 -.->|進化| B2\n    A2 -.->|マージン低下\u003Cbr/>但し成功の証| B1\n\n    B1 --> C1\n    B2 --> C2\n    B3 --> C3\n\n    style B1 fill:#4CAF50\n    style B2 fill:#2196F3\n    style B3 fill:#FF9800\n",[301,1029,1030,1035,1040,1045,1050,1054,1058,1063,1068,1073,1078,1082,1086,1091,1096,1101,1106,1110,1114,1119,1124,1129,1133,1138,1143,1148,1152,1157,1162],{"__ignoreMap":299},[304,1031,1032],{"class":306,"line":307},[304,1033,1034],{},"graph TD\n",[304,1036,1037],{"class":306,"line":313},[304,1038,1039],{},"    subgraph \"従来のSaaS\"\n",[304,1041,1042],{"class":306,"line":319},[304,1043,1044],{},"        A1[月額固定課金\u003Cbr/>Seat-based Pricing]\n",[304,1046,1047],{"class":306,"line":325},[304,1048,1049],{},"        A2[80-90%粗利益率]\n",[304,1051,1052],{"class":306,"line":331},[304,1053,346],{},[304,1055,1056],{"class":306,"line":337},[304,1057,353],{"emptyLinePlaceholder":352},[304,1059,1060],{"class":306,"line":343},[304,1061,1062],{},"    subgraph \"AI時代のビジネスモデル\"\n",[304,1064,1065],{"class":306,"line":349},[304,1066,1067],{},"        B1[成果ベース課金\u003Cbr/>Outcome-based]\n",[304,1069,1070],{"class":306,"line":356},[304,1071,1072],{},"        B2[消費量ベース課金\u003Cbr/>Usage-based]\n",[304,1074,1075],{"class":306,"line":362},[304,1076,1077],{},"        B3[アフィリエイト料\u003Cbr/>Affiliate Revenue]\n",[304,1079,1080],{"class":306,"line":368},[304,1081,346],{},[304,1083,1084],{"class":306,"line":374},[304,1085,353],{"emptyLinePlaceholder":352},[304,1087,1088],{"class":306,"line":380},[304,1089,1090],{},"    subgraph \"具体例\"\n",[304,1092,1093],{"class":306,"line":385},[304,1094,1095],{},"        C1[カスタマーサポート\u003Cbr/>タスク解決数で課金]\n",[304,1097,1098],{"class":306,"line":390},[304,1099,1100],{},"        C2[コーディング\u003Cbr/>消費量で課金]\n",[304,1102,1103],{"class":306,"line":396},[304,1104,1105],{},"        C3[AI検索・推奨\u003Cbr/>成果報酬]\n",[304,1107,1108],{"class":306,"line":549},[304,1109,346],{},[304,1111,1112],{"class":306,"line":555},[304,1113,353],{"emptyLinePlaceholder":352},[304,1115,1116],{"class":306,"line":560},[304,1117,1118],{},"    A1 -.->|進化| B1\n",[304,1120,1121],{"class":306,"line":565},[304,1122,1123],{},"    A1 -.->|進化| B2\n",[304,1125,1126],{"class":306,"line":571},[304,1127,1128],{},"    A2 -.->|マージン低下\u003Cbr/>但し成功の証| B1\n",[304,1130,1131],{"class":306,"line":577},[304,1132,353],{"emptyLinePlaceholder":352},[304,1134,1135],{"class":306,"line":583},[304,1136,1137],{},"    B1 --> C1\n",[304,1139,1140],{"class":306,"line":589},[304,1141,1142],{},"    B2 --> C2\n",[304,1144,1145],{"class":306,"line":595},[304,1146,1147],{},"    B3 --> C3\n",[304,1149,1150],{"class":306,"line":601},[304,1151,353],{"emptyLinePlaceholder":352},[304,1153,1154],{"class":306,"line":606},[304,1155,1156],{},"    style B1 fill:#4CAF50\n",[304,1158,1159],{"class":306,"line":612},[304,1160,1161],{},"    style B2 fill:#2196F3\n",[304,1163,1164],{"class":306,"line":618},[304,1165,1166],{},"    style B3 fill:#FF9800\n",[51,1168,1169],{},[28,1170,1171],{},"具体例",[22,1173,1174,1180],{},[25,1175,1176,1179],{},[28,1177,1178],{},"カスタマーサポート","：測定可能なタスクの解決に基づく価格設定",[25,1181,1182,1185],{},[28,1183,1184],{},"コーディング","：消費量に基づく価格設定",[51,1187,1188],{},[28,1189,1190],{},"成果ベースの支払いモデル",[76,1192,1193,1196],{},[25,1194,1195],{},"AIは人間を拡張または一部代替",[25,1197,1198,1199,1202],{},"支払いモデルは人間と同様に",[28,1200,1201],{},"成果に基づいた支払い","に",[51,1204,1205],{},[28,1206,1207],{},"コンシューマービジネスモデル",[76,1209,1210,1213,1216],{},[25,1211,1212],{},"アフィリエイト料の重要性が増大",[25,1214,1215],{},"例：ユーザーを熟知したAI（Grockのバージョン）が最適なホテルを特定し、成果に対してアフィリエイト料を徴収",[25,1217,1218],{},"広告の非効率性（広告主の組織的な過払い）を絞り出す",[51,1220,1221],{},[28,1222,1223],{},"仕事の未来",[76,1225,1226,1229,1232],{},[25,1227,1228],{},"イーロン・マスク氏：「最終的に仕事がオプションになる可能性」",[25,1230,1231],{},"アンドレイ・カルパシー氏：「AGIが10年後」という発言",[25,1233,1234],{},"ベイカー氏の反応：「10年は長すぎる。もっと短いタイムラインを望む」",[18,1236,1238],{"id":1237},"_2941-ロボティクスとイーロンのoptimus-ビジョン","29:41 - ロボティクスとイーロンのOptimus ビジョン",[51,1240,1241],{},[28,1242,1243],{},"ロボティクスの現実性",[76,1245,1246,1249],{},[25,1247,1248],{},"「非常に現実的」",[25,1250,1251,1252],{},"競争構造：電気自動車市場と同様に",[28,1253,1254],{},"テスラ vs. 中国",[51,1256,1257],{},[28,1258,1259],{},"ヒューマノイド論争の終焉",[76,1261,1262,1265],{},[25,1263,1264],{},"人間型ロボット（ヒューマノイド）が勝利",[25,1266,1267,1268],{},"理由：\n",[76,1269,1270,1273,1276],{},[25,1271,1272],{},"YouTubeの動画から学習可能",[25,1274,1275],{},"人間がスーツを着てロボットにやり方を教えることが容易",[25,1277,1278],{},"例：食器洗い機にグラスを入れる動作",[51,1280,1281],{},[28,1282,1283],{},"Optimusの進捗",[76,1285,1286,1289],{},[25,1287,1288],{},"50体のロボットが50種類の異なるタスクを実行",[25,1290,1291],{},"すべてのロボット工学者が「非常に感銘を受けている」",[45,1293],{},[14,1295,1296],{"id":1296},"結論",[51,1298,1299],{},"ベイカー氏の分析は、現在のAI投資ブームが2000年のインターネットバブルとは根本的に異なることを明確に示している。実際の利用、健全なROI、そして世界最高の企業群による慎重な投資という三位一体が、今回のAI革命を持続可能なものにしている。",[51,1301,1302],{},"アプリケーション層はまだ初期段階だが、インフラストラクチャ層への投資は合理的であり、推論機能の登場によってビジネスモデルの持続可能性も見えてきた。従来のSaaS企業は粗利益率の低下を恐れるのではなく、それを成功の証として受け入れ、積極的にAI製品に投資すべきだという明確なメッセージが示された。",[51,1304,1305],{},"ロボティクスから新しいビジネスモデルまで、AI革命はまだ始まったばかりであり、その未来は明るいというのがベイカー氏の一貫したメッセージである。",[51,1307,1308,1309],{},"出所：",[1310,1311,1312],"a",{"href":1312,"rel":1313},"https://x.com/a16z/status/1983904346225213804",[1314],"nofollow",[1316,1317,1318],"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 .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":299,"searchDepth":313,"depth":313,"links":1320},[1321,1324,1325,1337],{"id":16,"depth":313,"text":16,"children":1322},[1323],{"id":20,"depth":319,"text":20},{"id":49,"depth":313,"text":49},{"id":58,"depth":313,"text":58,"children":1326},[1327,1328,1329,1330,1331,1332,1333,1334,1335,1336],{"id":61,"depth":319,"text":62},{"id":68,"depth":319,"text":69},{"id":107,"depth":319,"text":108},{"id":247,"depth":319,"text":248},{"id":402,"depth":319,"text":403},{"id":659,"depth":319,"text":660},{"id":696,"depth":319,"text":697},{"id":875,"depth":319,"text":876},{"id":1008,"depth":319,"text":1009},{"id":1237,"depth":319,"text":1238},{"id":1296,"depth":313,"text":1296},null,"Atrades CIOギャビン・ベイカー氏が、AIバブル論を否定し、GPU投資の健全性、推論機能のインパクト、AI時代のビジネスモデルについて包括的に解説","md",{},"/ai-bubble-gavin-baker",false,"2025-10-31T00:00:00.000Z",{"title":5,"description":1339},"2025-10/2025-10-31/ai-bubble-gavin-baker",[1348,1349,1350,1351,1352,1353,1354,1355,1356],"AI","投資","バブル","GPU","OpenAI","Anthropic","ビジネスモデル","推論","ロボティクス","3amGq7wlI1KlaRxtGJVe9CEjnPCgVr5OUAL2wM4fM74",[],[1360,1364,1368,1372,1375],{"title":1361,"path":1362,"publishedAt":1363},"AI業界の重要指標まとめ - NVIDIA決算・OpenAI/Anthropic収益・トークン処理量","/ai-industry-metrics-summary","2025-12-05T00:00:00.000Z",{"title":1365,"path":1366,"publishedAt":1367},"「世界でAIを使っているのは800万人」は本当か — 公式発表の確認と10年後までのフェルミ推定","/ai-user-count-fermi-estimate","2026-07-15T00:00:00.000Z",{"title":1369,"path":1370,"publishedAt":1371},"Claude Codeで他社モデルが動く非対称性からMCP解剖まで — AIベンダー戦略を対話で掘って記事5本を書いた日","/ai-vendor-strategy-writing","2026-07-14T00:00:00.000Z",{"title":1373,"path":1374,"publishedAt":1371},"OpenAIがClaude Codeで自社モデルを動かす手順を自ら公開した。逆方向は規約違反なのに","/gpt-56-sol-claude-code-harness",{"title":1376,"path":1377,"publishedAt":1378},"AI格差は、まず体験格差として現れる","/ai-experience-gap-frontier-models","2026-06-29T00:00:00.000Z","https://log.eurekapu.com/og/blog/ai-bubble-gavin-baker.png?v=2025-10-31T00%3A00%3A00.000Z&title=AI%E3%83%90%E3%83%96%E3%83%AB%E8%AB%96%E4%BA%89%E3%81%AB%E7%B5%82%E6%AD%A2%E7%AC%A6%EF%BC%9F%20Atrades%20CIO%E3%82%AE%E3%83%A3%E3%83%93%E3%83%B3%E3%83%BB%E3%83%99%E3%82%A4%E3%82%AB%E3%83%BC%E6%B0%8F%E3%81%8C%E8%AA%9E%E3%82%8BAI%E6%8A%95%E8%B3%87%E3%81%AE%E7%9C%9F%E5%AE%9F&author=Kei%20Komatsu&sig=45e49dbb5a68a653",1785654658264]