Three things happened within days of each other this week, and each one weakens a different support holding up the AI boom.
The first support was political goodwill, and the US midterms just knocked it sideways. In dozens of local ballot measures, voters rejected data center projects outright, and they elected politicians who ran on restricting them, like Texas governor-elect Gina Hinojosa. On TechCrunch's Equity podcast, the panel noted that Elon Musk seems to have gained little from roughly half a billion dollars in Republican donations, while his companies now sit in a state that, as one host put it, turned about as blue as it could possibly turn. Data centers were never the headline issue of the campaign, yet they became a grassroots one, because of how quietly hyperscalers went about winning approvals.
The second support was the assumption that AI spending only moves up. Ramp's card data says otherwise. Token spend grew 1.5 percent across August and September, compared with 42 percent growth in July. Usage keeps climbing while budgets hold flat, which means price competition and savvier buyers are capping what the labs can charge. Ramp's lead economist, Ara Kharazian, put the concentration risk plainly: almost the entire spend on AI comes from venture-backed, high-growth tech companies. The boom's customer base is much narrower than its headlines suggest.
The third support was insider confidence, and here the tells are subtle rather than loud. Goldman Sachs, a lead underwriter on Anthropic's IPO, has pulled Claude Code and GitHub Copilot away from its own traders and bankers, reserving the tools for engineers only. Google bought Crusoe's troubled Wyoming data center campus after declining to be its customer, committing $17 billion to finish a project the developer could not deliver on its own. Nobody is leaving the table. Everyone is hedging their seat.
None of this is a crash narrative. OpenAI is still reportedly raising fresh billions, and Anthropic's float remains the event the whole industry is pacing itself against. What changed is the terms of the conversation. A month ago the question was how fast AI revenue could grow. This week the questions are who pays, at what price, and with whose political cover.
For builders, the practical read is straightforward. Cheap tokens and flat vendor revenue mean pricing power sits with you for now, but plan for providers chasing enterprise value rather than loyalty. For anyone reading an AI prospectus this season, net-dollar retention and customer mix now matter more than benchmark charts.
And beneath the numbers sits the human story. Traders who built their daily workflow on a tool had it quietly taken away. Towns voted the future out of their backyards. The most valuable private companies in tech history are sprinting toward the public market at the exact moment the tide pauses. That tension is the real story of the week, and it is not finished yet.
Sources: TechCrunch on the midterms fallout Ramp's spending data Goldman's Claude Code pullback Crusoe's Wyoming sale