OpenAI banned two clusters of ChatGPT accounts that it says were likely operated from China and used to generate posts, comments, and images aimed at U.S. technology debates, including public opposition to AI data centers.
The campaigns, detailed in a June 10 threat report, did not appear to break through into a meaningful public audience. That is the important caveat. But the choice of target is still notable: AI infrastructure has moved from a back-end engineering issue into a political, energy, and national-security fight, and influence operators appear to be testing how easily that fight can be pushed.
What OpenAI Found
OpenAI described one cluster as the “Data Center Bandwagon” campaign. According to the report, the accounts used ChatGPT to create social media-style comments and imagery arguing that AI data center buildouts were raising electricity prices for ordinary families. The material tried to sound local and populist, leaning on existing concerns about power bills and the visible footprint of large data center projects.
A second cluster, which OpenAI called “Tech and Tariffs,” generated comments and political-cartoon concepts criticizing U.S. tariff policy and broader technology competition. OpenAI wrote that prompts in that cluster instructed the model to avoid depicting Chinese leader Xi Jinping while focusing criticism on President Trump. The same network was also tied to false claims that ChatGPT user data had been compromised.
The report characterizes both clusters as likely originating from China and says OpenAI banned the accounts. Outside coverage from Axios, Al Jazeera, and CyberScoop has emphasized the same restraint: the activity is significant because of what it targeted, not because there is evidence that it shifted the debate.
Why Data Centers Are a Useful Target
Data centers are an unusually good issue for influence campaigns because the underlying debate is real. Communities are already arguing over land use, water, power-grid strain, local tax incentives, noise, jobs, and whether big technology companies are shifting infrastructure costs onto residents. An influence campaign does not need to invent that tension. It only needs to amplify the most polarizing version of it.
The scale of the buildout gives those arguments weight. The International Energy Agency has projected that global data center electricity consumption could roughly double from 485 terawatt-hours in 2025 to about 950 terawatt-hours in 2030, with AI-focused data centers growing faster than the sector as a whole. In the United States, the Energy Information Administration’s 2026 outlook projects server electricity consumption rising sharply through 2050, especially in standalone data centers.
Local opposition is no longer a niche planning-board issue. Data Center Watch, a research project tracking U.S. opposition to large data center developments, has estimated that tens of billions of dollars in projects have been blocked or delayed by community, regulatory, and political resistance. The exact number will vary with methodology, but the direction is clear: AI infrastructure has become visible enough to generate organized pushback.
That makes the topic attractive to foreign influence operators. Electricity prices are personal. Utility infrastructure is technical and difficult to explain. Data centers are often owned or leased by large companies that many residents already distrust. Even accurate concerns can be repackaged into misleading claims if audiences cannot tell the difference between a local resident, an advocacy group, a political campaign, and a coordinated account network.
The Platform Problem Is Bigger Than One Report
OpenAI’s report lands at a difficult moment for AI companies. The same companies selling models as productivity tools now have to show that those models cannot be quietly turned into influence-work accelerators. Text generation lowers the cost of writing posts in different tones. Image generation can make cartoons, memes, and pseudo-local graphics faster. Translation and style transfer can help operators test messages across audiences.
The mechanics matter. A small campaign can use an AI model to generate many versions of the same argument, vary the apparent author’s voice, draft replies to critics, and localize claims around power bills or construction fights. Even if most of that activity is clumsy or low-impact, it gives operators a cheap way to test which narratives deserve more investment.
That does not mean every complaint about an AI data center is suspect. Treating legitimate local opposition as foreign manipulation would be both unfair and politically reckless. Residents can have good reasons to question a project’s water use, tax deal, utility plan, environmental review, or job claims. The sharper lesson is that real grievances become more vulnerable to manipulation when companies and local governments leave basic questions unanswered.
What To Watch Next
The next test is whether platforms and public officials can separate disclosure from spin. AI companies will keep publishing threat reports because they have the logs and account-level evidence that outsiders lack. Those reports are useful, but they are also self-interested documents from companies defending their role in the infrastructure fight. They should be read with both seriousness and skepticism.
For local governments, the practical response is not to dismiss opposition. It is to demand better project information early: expected electricity load, who pays for grid upgrades, water use, backup generation plans, noise controls, tax incentives, emergency response requirements, and enforceable community benefits. Transparent data leaves less room for anonymous accounts to fill the gap with exaggeration.
For platforms, the hard part is detection across formats. Influence operations rarely stay inside one product. A campaign may use a chatbot to draft posts, an image tool to produce visuals, social platforms to distribute them, and messaging apps to coordinate. Effective disruption depends on linking behavior patterns across account creation, prompting, repeated themes, off-platform personas, and coordinated posting.
The OpenAI case is best understood as an early signal. The campaigns were apparently small. Their reach was limited. The issue they targeted is not small at all. AI data centers are becoming one of the physical pressure points of the AI boom, and that makes them a natural place for influence operations to probe for weakness.
