Big Tech’s big AI buildout will not end until someone stops it

Four hyperscalers lined up CapEx of at least $720 billion this year, with another trillion-plus in multi-year commitments already teed up. Most of that money is going into AI data centers, not the cash-cow businesses that still fund the spend.

Wall Street finally flinched — Alphabet’s stock took its worst one-day drop on record after earnings — then bounced as midterm results raised the chance of local moratoriums and restrictions on new data centers. The market is pricing a forced slowdown as a feature, not a bug.

That is the tell. Investors are no longer sure the buildout pays for itself on schedule, and they are starting to hope politics does what the boards will not.

We read this as the clearest earnings-week confession yet that Big Tech will not voluntarily tap the brakes. Amazon’s Andy Jassy called the spend a once-in-a-lifetime shot at a trillion-dollar company and waved away the idea of spreading CapEx over 25 years. Alphabet’s CFO said 2027 spend will rise “significantly.” Meta is expected by banks to climb further from this year’s $135–145 billion band. Microsoft was the only one that spent under street expectations — because third-party capacity slipped, not because it chose austerity — and it still opened a $25 billion bond sale the next morning.

The revenue story is real and still not enough. Amazon posted its fastest growth in five years. Meta’s ads are surging. Google Cloud more than doubled. And yet Google’s own execs admitted AI Mode queries are growing while monetizing worse than classic Search, which slowed from 17% growth a year ago to 12% this quarter. The cash cow that financed the AI bet is being chewed by the product that was supposed to replace it.

Anthropic’s updated S-1 numbers make the same paradox smaller and sharper: roughly $17 billion in quarterly revenue against $17.5 billion in compute costs, plus nearly $900 billion in future compute and infrastructure commitments. Hyperscalers are selling capacity they say they do not have enough of, while fronting the hardware spend today for cash that arrives later.

We’re watching who blinks first — boards, debt markets, or voters. Midterms barely came up on any earnings call, which is its own kind of tell. Useful AI at this scale needs infra. Unbounded CapEx without a clear path to profit is not strategy; it is a race that only ends when someone outside the room forces a stop.

Ellen Thomas’s Verge piece stitches together the late-October earnings calls from Amazon, Microsoft, Alphabet, and Meta with Anthropic’s fresher S-1 compute figures. The selloff that followed Alphabet’s day and the rebound after midterm results are the market framing around the CapEx numbers.