X Says China Linked Bots Manufactured Part of the Data Center Backlash
Per Engadget, X says its safety team uncovered a network of 200,000 inauthentic accounts linked to China, roughly 200 of which were posting comic strips, images and text attacking AI data centers. The company said the content could manipulate a legitimate debate about American AI and energy policy. The posts claimed data centers raise household electricity prices and strain local grids, illustrated with AI-generated cartoons casting operators as profiteers while ordinary families pay.
This is the second time the same campaign has surfaced. OpenAI published a report in June describing China-linked accounts using ChatGPT to generate anti-data-center material, and Engadget notes the illustrations in X's examples match the ones in that report. Two separate platforms, working from different data, identified the same operation.
The finding should change how the visible backlash is read. Opposition to data center siting has felt like an organic local movement, and some of it certainly is. But at least part of the online volume was produced by accounts that do not exist, with imagery generated on demand and talking points written by a chatbot. Neither X nor OpenAI has sized what share of the conversation that represents, and neither can, which is the actual problem. Once manufactured opposition is confirmed to be in the mix, no one can tell you how much of what remains is genuine.
The economic claim underneath deserves the same scrutiny. Goldman Sachs put US electricity prices up 6.9% year over year as of February and expects further increases as AI facilities add load, so power costs are rising. Attributing that increase to data centers specifically is a separate and contested question involving generation mix, transmission investment, fuel costs and rate design. The campaign asserted the causal link as settled, which is what an operation designed to shape policy would do.
The caveats run in both directions. This is X's own account of its own investigation, published on its Global Government Affairs page, with sample posts but no account list and no external verification. X also did not say what the other 199,800 accounts in the network were posting about, so whether data centers were a priority target or a small experiment inside a much larger operation is unclear from the disclosure.
What to watch is how much of the anti-data-center sentiment survives scrutiny. Siting decisions, capacity auctions and grid cost allocation are live in dozens of states, and they should be settled on load forecasts and rate filings rather than on engagement volume. The lesson from both reports is that engagement volume is now cheap to fabricate, and the side of an argument that looks loudest online is no longer evidence of anything.