Tuesday’s midterms gave Democrats both chambers of Congress, and plenty of them won by running against data centres. Bernie Sanders celebrated with “National data center moratorium now” on X. The bill he announced in March with Alexandria Ocasio-Cortez would halt construction or upgrades of any data centre drawing 20 megawatts or more until “strong national safeguards” exist. Investors took the hint, and by Thursday afternoon the Nasdaq was down 5.4% as chip and data-centre stocks were dumped. Further down the ballot, more than 30 local measures to restrict data centres went to voters in 13 states, and nearly all passed.

A national moratorium isn’t happening while Trump holds the veto, and he has waved off calls for mandatory AI rules all along. Governors are another matter. Texas holds 27% of planned US data-centre capacity by Wells Fargo’s count, and Gina Hinojosa beat Greg Abbott there after promising “a moratorium, a ban, on all data center approvals” until the state legislature passes protections. That needs nothing from Washington. Cleanview reckons the new governors could pause 41% of planned US data centres between them. In Congress the likelier weapon is cost, with Chuck Schumer naming household power bills as a priority and talk of making developers pay for their own grid upgrades.

The election didn’t create the problem. It landed on a build-out that had already run out of its own cash, and a frozen permit doesn’t cancel a loan, it just stops the building that was meant to repay it. Meta, Alphabet, Microsoft and Amazon spent close to $207 billion on capex last quarter, nearly double a year earlier, and over the same quarter Alphabet, Meta and Amazon spent $45.2 billion more than they generated in free cash flow. A year ago the same three were $36.4 billion in the black. Only Microsoft is still positive, by $18.6 billion.

More and more of the gap is borrowed, and lenders are getting nervous. Goldman counts $296 billion of hyperscaler debt issued this year. S&P cut Meta’s credit outlook to negative. CoreWeave has a 2032 bond with a 9.625% coupon that now yields above 16%, which is a market pricing the chance of not getting paid back. I wrote in February about money going in circles between chipmakers, labs and clouds, and the circle only turns while someone outside it keeps lending.

Gartner’s arithmetic is the real bubble case: data-centre owners need $8.5 trillion of cumulative revenue by the end of the decade to justify what they’re building, against no more than $3.6 trillion of AI revenue those data centres are expected to bring in. Nothing I’ve read closes that gap, and spending keeps climbing regardless, towards about $740 billion a year for the four big spenders. So I don’t expect a pop so much as a squeeze, weakest borrowers first, and squeezes still end in falls. Capital Economics has the S&P 21% lower by the end of 2027, and I’m with them.

A Texas freeze would hurt Alphabet, which now borrows too, but Alphabet still has search to fall back on. A neocloud with a stalled campus and bonds yielding 16% has nothing behind it, which is why I’m watching the CoreWeave yield rather than the Nasdaq. If it keeps climbing, or the paper stops selling at any price, the lenders will have answered the question before the stock market does.

Sources: