Investors finally flinched on Friday at the scale of the AI build-out. Over two weeks Alphabet, Amazon, Meta and Microsoft outlined plans to spend more than 740 billion dollars on capital expenditures this year, most of it read by Wall Street as AI infrastructure: data centers and chips. The Nasdaq fell 2.9 percent on Friday, Nvidia dropped 4 percent, Broadcom more than 5 percent and Qualcomm 11 percent. Alphabet, which fell 12 percent on Thursday in its worst day on record and lost 500 billion dollars in market value, fell again.
Alphabet is the clearest case. It reported negative free cash flow of 9.1 billion dollars in the third quarter, its second quarter in the red, as quarterly capital spending reached 51.4 billion dollars. It expects up to 205 billion dollars this year, and more in 2027. Growth in Search advertising, roughly half its revenue, slowed to 12 percent, its weakest pace in five quarters, even as its AI Mode chatbot succeeds inside Search. Amazon is similar: negative free cash flow of 42.4 billion dollars over 12 months, versus positive 14.8 billion dollars the year before, expected spending of 220 billion dollars, and a 7 percent stock fall after earnings despite AWS revenue growing 39 percent.
Meta reported negative free cash flow of 2.1 billion dollars in the third quarter and guides to as much as 145 billion dollars in spending this year, while calling its Muse personal agent very compute-intensive to run. Microsoft was the exception, with positive free cash flow of 18.6 billion dollars on 45.2 billion dollars of spending, below its 50 billion dollar projection, which it blamed on third-party capacity delays, not a pullback. The shortfall is increasingly debt-funded: Microsoft, Alphabet, Amazon, Meta and Oracle raised a combined 326 billion dollars in debt in 2026, triple last year, per S and P Global Market Intelligence.
That climate now hangs over Anthropic's public offering. The Financial Times reported Friday, via TechCrunch, that Blackstone, an anchor investor in a recent 63 billion dollar deal to finance Anthropic's chips, told investors it expects a high junk rating after the IPO, in the double-B range, the top tier of speculative grade. Blackstone had previously suggested the opposite, an investment-grade rating before 2027. Both firms declined to comment. A junk rating would not stop Anthropic, which filed earlier this week, but it would make its enormous compute deals costlier to fund.
In a quieter coordination move the same day, OpenAI, Google, Microsoft and Anthropic said they will honor a single robots.txt token, __usage_ai-training, so a publisher can opt content out of AI training across all four crawlers in one line. Anthropic says it has already begun implementing it, OpenAI aims to follow in the coming months, Microsoft targets February 1, 2027, and Google targets 2027. The signal comes from an individual Internet Engineering Task Force draft dated October 30 by Anthropic's Kevin Kelley. It controls training, not fetching, and it does not remove material already in training sets.
For builders, demand is not the problem: the cloud businesses at Google, Amazon and Microsoft all grew. The problem is who pays, and on what terms, while returns are still uneven. Expect harder questions in every enterprise renewal about cost per task, model routing, and whether a cheaper open-weight option can do the job. The era of spending first and justifying later is not over, but on Friday the market started pricing the justification.