What happens to the AI boom when local communities simply stop wanting data centers? This week delivered a surprisingly direct answer. In the US midterm elections, voters decided on forty-one ballot measures related to data centers. In all but one, the majority chose stricter rules.
Markets took a day to process this, then reacted all the more forcefully. Roughly three trillion dollars in market value disappeared from Wall Street in a single day, mostly in tech stocks. The logic is simple: valuations of the large AI companies rest on the assumption that computing capacity can grow almost without limit. Once communities turn land, power and local burdens into political questions, that assumption becomes negotiable.
What stands out is that the doubts predate the election result. Spending data from Ramp shows that companies keep using AI more intensively, but are no longer paying correspondingly more for it. Price cuts and tighter cost control among the biggest customers are slowing revenue growth. For companies using AI, that is good news. For investors who bet on hypergrowth, less so, and the timing is awkward, with Anthropic nearing its IPO.
On the infrastructure side, the calculation looks just as fragile. Microsoft has reportedly ended an AI contract with Oracle because older Nvidia chips no longer met its requirements, and is said to be reviewing other cloud partnerships worth billions. Hardware that was recently considered scarce becomes a depreciation item within a few years. Anyone signing long-term compute contracts is buying not only capacity but also ageing risk.
OpenAI sits right in the middle of this tension. The company has reportedly raised an additional eleven billion dollars at a valuation of one point two five trillion dollars. Because of the sell-off and concerns about high costs, however, its targets were lowered. Two days after the election, OpenAI also announced new rules for more neighbour-friendly data centers. One could say the neighbourhood is now officially part of the supply chain.
At the same time, competition is moving to where AI actually acts and can cause damage. OpenAI introduced GPT-6 Cyber, a model meant to find more vulnerabilities while staying within defined limits. Meanwhile, Palo Alto Networks is in talks to buy Clutch, a startup that secures AI agents, for around five hundred million dollars. Its last valuation was eighty-five million. That premium says a lot about how urgently companies need control over software that acts on its own.
The legal framework is tightening as well. A court in Brazil has barred Meta from using media content for AI training without compensation and threatened penalty payments. And one employee obtained a religious exemption from having to use AI at work. Two very different cases, one common thread: who uses AI, and who feeds it data, under which conditions, is no longer decided by the providers alone.
For Europe, this opens a distinct perspective. Mistral is positioning its new ML4 model around openness as a contribution to sovereignty. Political scientist Nicole Lemke argues that writing Europe off is riskier than continuing to invest in its own large models. If infrastructure in the United States becomes politically more expensive, the relative chances of other locations shift too.
This week showed that computing power is made not only of chips and capital, but also of consent, and that resource cannot be raised in any funding round.