OpenAI says China-linked accounts targeted US data centers

OpenAI said Wednesday that China-linked ChatGPT accounts drafted messages against US data centers and tariffs, putting the AI infrastructure boom into the same geopolitical fight already running through chips, power contracts and capital markets.

In a threat intelligence report, the company said it banned two clusters of accounts tied to China after they used its tools to write posts, comments and scripts for US political debates. One cluster worked on local resistance to data centers. The other leaned into anger over tariffs and prices. OpenAI said it found no evidence the activity spread widely, but the target was not random: the largest AI companies still need permits, grid access and political tolerance to keep infrastructure plans moving.

Ben Nimmo, OpenAI’s principal investigator on intelligence and investigations, told Axios the operation tried to ride an argument already playing out in US cities and statehouses.

“The debate existed already. This was an influence operation from China trying to interfere in it.”
Ben Nimmo, OpenAI principal investigator, to Axios

That distinction matters for investors. OpenAI is not saying foreign accounts invented opposition to data centers. It is saying a China-linked network tried to amplify a domestic pressure point just as investors are starting to underwrite a much larger AI capital cycle.

Polling cited by Axios showed 40 per cent of Americans supported data centers in their area, while 32 per cent opposed them. Axios also cited a Harris poll finding that 70 per cent of Americans said President Donald Trump’s tariffs had caused higher prices. The outlet said this was the first time OpenAI had reported a China-linked operation using its models to meddle in the AI data-center debate.

Why data centers matter

Compute sites are the hard constraint behind the AI trade. Model developers can release new systems quickly. Each new generation still needs more capacity, electricity and agreements with utilities, landowners and local governments. That gives opponents leverage in zoning hearings, power-rate fights and water-use reviews.

The local fights are already visible. The Verge reported last week that New York lawmakers passed a one-year ban on new data centers. A separate Verge report said Amazon employees had asked Seattle officials to slow new projects over electricity prices, water use and noise. Local issues can become financing issues when companies are promising investors faster model development and lower unit costs.

OpenAI, though, framed the accounts as low-impact rather than a successful mass campaign. Company researchers said the operation did not appear to break out beyond its own activity.

“we found no evidence of meaningful breakout beyond its own activity”
OpenAI threat intelligence report

That caveat limits the immediate market read-through. OpenAI’s report does not show that the alleged campaign changed policy outcomes, shifted polls or delayed a project. The route of attack is the point. If foreign influence operators see AI infrastructure debates as useful targets, local energy fights may become part of national-security risk assessments for the sector.

The market risk

The data-center fight lands as AI companies try to turn private-market enthusiasm into long-duration financing. Large model developers need cloud contracts, chips, power capacity and construction partners before revenue from enterprise AI products fully catches up. Opposition can raise costs even when it does not stop a project. Developers may need concessions, longer permitting timelines or more expensive power arrangements.

The episode is awkward for OpenAI as well. Its executives are selling artificial intelligence as productivity infrastructure while acknowledging that its own models can be used to draft influence material at scale. Publishing the details lets OpenAI argue it is policing abuse without slowing adoption.

A narrower policy question follows. US officials, utilities and investors are already deciding where data centers should be built, who pays for grid upgrades and how much of the AI buildout should be treated as strategic infrastructure. The geopolitical layer is now harder to separate from those decisions: the same projects that underpin model leadership may also become pressure points for rivals trying to slow them.

For now, the evidence points to a small campaign with limited reach. The larger risk is that the politics of AI infrastructure become investable risks. Power bills, local moratoriums and foreign influence operations now sit in the same conversation as chips and model benchmarks.