AI万亿美元赌注能否如期兑现成疑

全球超大规模科技公司正在AI基础设施上进行前所未有的大规模投资，但这场赌注的成败存在重大风险。谷歌、微soft、亚马逊、Meta和甲骨文等巨头需要在2030年前将生产率提高2.7倍，才能收支平衡并实现15%的回报率1。若无法实现这一目标，将面临"历史上最大的资本配置失误"1。

这些公司目前面临投入与产出的严重不匹配。今年这些超大规模公司将花费约750亿美元建设数据中心1，到2027年AI支出将达到近1.1万亿美元1，未来四年总投资可能超过5万亿美元1。然而，今年总AI收入仅为1500亿至2000亿美元1，远低于支出规模。Alphabet最新季度收入近1200亿美元，但因AI基础设施支出吞没利润，自由现金流出现赤字约59亿美元，这是Google自2004年上市以来首次出现1。

现实中的生产率改善迹象有限。调查显示约90%的高管在过去三年未曾看到生产率提升，虽然他们预期未来三年提升1.45%（美国高管预期2.25%）1。与此同时，这些投资已演变成复杂的金融工程。摩根士丹利计算显示，2025至2028年超大规模公司2.9万亿美元支出中超过一半将通过外部融资完成1。Meta将其路易斯安那州Hyperion数据中心项目投资从100亿美元升至500亿美元，并向Blue Owl Capital转让该项目80%的股权1。

技术层面也存在不确定性。GPU芯片性能约每两年翻倍，数据中心面临芯片贬值的"计时炸弹"问题1。电力需求急剧上升，Entergy公司计划建设七座燃气电厂来满足AI数据中心的能源需求，总容量约7.5吉瓦——相当于新奥尔良用电量的六倍1。这些投资最终需要依靠全经济生产率增长才能被证明具有合理性。

Major technology companies including Alphabet, Microsoft, Amazon, Meta, and Oracle are betting hundreds of billions of dollars on artificial intelligence infrastructure, but the mathematics of their investment strategy presents a stark reality: they must nearly triple productivity by 2030 just to break even and achieve a 15% return on capital 1. Without such productivity gains, researchers warn this would represent "the largest capital allocation mistake in history" 1.

The scale of this wager is staggering. This year alone, hyperscale companies plan to spend approximately $750 billion constructing data centers 1, with AI expenditure projected to reach nearly $1.1 trillion by 2027 1. Over the next four years, total AI capital investment could exceed $5 trillion 1. Yet current revenues tell a different story: total AI revenue stands at $1.5 billion to $2 billion annually, creating a severe mismatch with spending levels 1. Alphabet illustrates this tension acutely—its latest quarterly revenue approached $120 billion, but AI infrastructure expenses consumed these gains entirely, leaving the company with a free cash flow deficit of approximately $59 billion, marking its first such shortfall since going public in 2004 1.

The financing of these massive commitments increasingly relies on external capital. Morgan Stanley calculates that more than half of the $2.9 trillion in hyperscale company spending projected for 2025-2028 will be funded through outside financing 1. Meta's approach exemplifies this trend: the company raised its Hyperion data center investment in Louisiana from $10 billion to $500 billion and transferred 80% equity to Blue Owl Capital 1. A survey found that approximately 90% of executives have not observed productivity improvements over the past three years, though they expect a 1.45% increase in the coming three years 1.

Infrastructure challenges compound the financial pressure. GPU chip performance doubles roughly every two years, creating a "time bomb" of depreciation for data centers 1. Energy demands are also escalating dramatically: Entergy plans to build seven natural gas power plants with combined capacity of approximately 7.5 gigawatts—six times the electricity consumption of New Orleans—to support these operations 1.