全球科技巨头正在掀起历史上规模最大的资本支出浪潮。亚马逊、谷歌和微软等公司在2025年投入了4500亿美元用于数据中心和AI芯片等基础设施建设，2026年的资本支出将翻倍至9000亿美元，并预计在2027年进一步达到1.4万亿美元。然而，如此高昂的投入与回报之间存在着巨大鸿沟：粗略计算表明，要使这些资本支出产生合理的回报，全球AI行业每年需要产生约2.5万亿美元的收入，而目前整个科技行业的总收入甚至未达到这一水平。
虽然AI技术的扩散速度极快——美国人口普查局调查显示约20%的美国企业在业务中使用AI，英格兰银行数据显示英国约三分之一的企业使用AI——但企业的实际付费意愿和使用深度仍然非常有限。研究显示，约半数使用AI的企业仅依靠免费版或开源模型；在美英加三大国的中小企业中，仅有约十分之一购买了付费AI工具，且企业员工平均每周使用AI仅1.7小时。粗略统计显示，目前全球生成式AI和AI服务的年化总收入仅在1500亿至2200亿美元之间（其中Anthropic年化收入约750亿美元，OpenAI为数百亿美元），远远无法填补万亿美元级的支出漏洞。
国际清算银行（BIS）学者富里查·隆查伦基库的最新研究指出，当前AI算力消费的快速增长中，至少有三分之一的资本支出属于零和竞争，即企业建成的算力并非用于扩大付费市场总量，而是用于在竞争对手间争夺既有客户。此外，高达90%的高管报告称过去三年AI并未对其企业的生产力产生显著提振，且企业在组织流程与人才重组等“无形资本”上的投入依然在下降。这表明AI要彻底变革实体经济并兑现万亿美元收入承诺，仍面临着极其漫长而艰巨的落地挑战。
Big technology firms are orchestrating the largest capital expenditure surge in history. After spending $450bn in 2025, tech giants are projecting $900bn in capex in 2026 and $1.4trn in 2027 to build AI compute infrastructure. However, financial modeling indicates that justifying this level of capex requires annual AI-generated revenues of approximately $2.5trn. Since peaking in June, major AI tech stocks have fallen 20%, while South Korea's chip-heavy benchmark index dropped nearly 40% over concerns regarding return on investment.
Despite rapid initial adoption—with Census Bureau data showing 20% of US businesses use AI and surveys showing 33% of workers employ AI—monetization remains constrained. Intuit reports that only 1 in 10 small businesses pay for dedicated AI software, while Bank of England research finds executives average just 1.7 hours of AI use per week. Current annualized global AI service revenues are estimated at between $150bn and $220bn (with Anthropic generating ~$75bn and OpenAI tens of billions), leaving a multi-trillion-dollar revenue gap.
Research by Phurichai Rungcharoenkitkul at the Bank for International Settlements suggests that at least one-third of current AI capital expenditure represents zero-sum market-share competition rather than net market expansion. Furthermore, 90% of executives report no measurable productivity gains from AI over the past three years, while US corporate investment in organizational capital continues to decline relative to GDP, raising concerns that massive infrastructure spending is outpacing actual enterprise value creation.
Source: AI revenues are growing fast, but not fast enough
Subtitle: The returns on trillions of dollars of spending are deeply uncertai
Dateline: 7月 30, 2026 05:43 上午