読書リサーチレーダー投資フレームワーク
ログイン / 新規登録
← 読解に戻る
ニューストピック · AI资本支出挤出效应重大3 分で読めるDisconfirmAI 編集部

Economist Alex Tabarrok: Data Center Investment Is Not Crowding Out Other Investment—AI Is a Bright Spot Against Headwinds

Alex Tabarrok, co-author of the economics textbook Modern Principles, systematically refutes the claim by Paul Krugman and others that data center investment is crowding out other investment: in the standard loanable-funds model, crowding out is less than 1:1; Krugman's GDP-share graph suggests crowding out, but GDP itself is growing, so one share rising while another falls cannot be equated with crowding out; in real-dollar terms, other real investment still grew about 3% from Q2 2023 to Q2 2026; the rise in US 10-year TIPS real rates occurred largely before the AI investment boom; and real rates rose across all advanced economies, not just the US, undermining a US-specific crowding-out explanation. Tabarrok argues a more plausible narrative: tariffs, policy uncertainty, and the fading of Biden-era CHIPS and IRA subsidies weakened non-AI investment, while AI investment served as a bright spot amid overall investment weakness.

2026-10-19 19:16 GMT+8

Paul Krugman and several economists recently advanced the claim that massive capital spending on data centers is crowding out investment in all other sectors of the US economy. Alex Tabarrok, co-author of the economics textbook Modern Principles, systematically refuted this argument on his Marginal Revolution blog.

The first problem is arithmetic. Tabarrok pointed out that Rattner's chart shows data center construction spending rising by $51 billion, while the so-called 'crowded-out' other investment allegedly fell by $120 billion. In the standard loanable-funds model, the crowding-out effect is generically less than 1:1—$51 billion of additional borrowing will crowd out less than $51 billion of other investment spending, and cannot explain a $120 billion decline.

The second problem is how to read the graph. Krugman uses a comparison of 'tech investment as a share of GDP' versus 'other investment as a share of GDP' to suggest crowding out. Tabarrok noted that when GDP itself is growing, one share rising while another falls is a natural outcome and cannot be directly equated with crowding out. He proposed regraphing the same data in real dollars: from Q2 2023 to Q2 2026, real tech investment rose about 46%, while other real investment rose about 3%. Other investment has weakened since its 2024 peak but remains above its starting level.

ログイン / 新規登録
読書リサーチレーダー投資フレームワーク

The third problem is time sequence. If data center borrowing were truly pushing up interest rates, the rise should coincide with the AI investment boom. But the average US 10-year TIPS real yield rose from −1.00% in late 2021 to 1.49% in late 2022—most of that increase occurred before the AI investment boom. Over the periods covered by Rattner's and Krugman's charts, rates rose only 0.5 and 0.68 percentage points respectively, relatively small compared to the earlier jump. Current real rates are approximately 2%–3%, in line with historical averages and hardly prohibitive for investment.

The fourth problem is scope. Tabarrok observed that real rates are rising across all advanced economies, while the vast majority of AI spending is concentrated in the United States. If rate increases were caused by US data center borrowing, why are rates rising in other countries too? The global savings base is much larger than the US savings base, making it difficult for a single country's investment demand to explain a global rate movement.

Tabarrok's alternative explanation is that two things are happening simultaneously. On one hand, tariffs, policy uncertainty, and the fading of Biden-era CHIPS Act and IRA subsidies have weakened investment outside AI and tech. On the other hand, the AI boom has supported overall investment levels. In his framing, AI investment is not crowding out other investment but rather serving as a bright spot in what would otherwise be a more negative economy.

The author himself preserves the boundary: 'This does not rule out crowding out: other investment might have grown faster without the tech boom.' The counterfactual cannot be identified with these graphs. Tabarrok's argument weakens the strength of the crowding-out claim but does not entirely dismiss it.

ソース

  • marginalrevolution.com↗
トピック

AI资本支出挤出效应

围绕美国AI基础设施巨额投资是否“挤出”其他经济活动或推高国债收益率的争论正在升温。主流机构如PIMCO和高盛指出,尽管AI资本支出可能通过储蓄-投资渠道推高实际利率,但缺乏证据支持其直接挤出国债市场或导致非AI投资绝对下降;相反,数据显示…

4 ステップ2 件更新日 2026-10-23 14:05

経緯

  1. 2026-08-11高盛发布报告称,AI资本支出对GDP的贡献和对其他投资的挤出效应均被夸大,估计间接挤出效应约为500亿美元,对2026年GDP增速影响约0.1个百分点。businessinsider.com ↗
  2. 2026-08-29《财富》杂志报道,埃德·亚德尼等分析师提出“反向挤出”观点,认为AI公司大量发债吸走资金,导致国债收益率上升以吸引买家,而非传统意义上的私人投资被政府债务挤出。fortune.com ↗
  3. 2026-09-21PIMCO播客《The Credit Market Lens》指出,AI资本支出可能推高实际利率,但这是通过储蓄-投资平衡机制,而非AI债券发行直接挤出国债;实证研究显示意外AI债务交易对10年期收益率无显著统计足迹。pimco.com ↗

未解決の問題

  • 随着AI债券发行规模持续扩大,是否会最终在长期内显著改变国债市场的供需结构并推高风险溢价?
  • 非AI领域投资的相对疲软究竟多大程度上归因于AI的资源竞争,而非宏观经济政策(如关税、补贴退坡)的影响?
  • 未来几个季度的企业财报中,超大规模云服务商(Hyperscalers)的资本支出指引是否会调整,从而缓解市场对资源挤占的担忧?

このスレッド内の他の記事

  • PIMCO Podcast: AI Capex and the Limits of Crowding Out2026-10-14
トピック全文を読む →
リサーチを続ける →