Suspected Chinese Bot farm of approx 200,000 accounts, involved in influence operations idenfied by X

On 28 August 2026, X’s Global Government Affairs account announced that its Safety team had identified a network of approximately 200,000 inauthentic accounts assessed as Chinese in origin. Within that farm, 200 accounts posted coordinated content designed to shape public debate over American AI infrastructure and energy policy. The posts claimed that AI data centers were driving household electricity prices higher and straining regional grids; they also circulated AI-generated cartoons portraying data-center operators as cigar-smoking profiteers enriching themselves at ratepayers’ expense. X stated it had suspended accounts that violated its authenticity policy while affirming that legitimate policy discussion should continue.

The campaign did not invent the underlying controversy. PJM Interconnection, the grid operator serving 67 million people across 13 states and the District of Columbia, has recorded successive capacity-auction shortfalls and price spikes directly linked to data-center load growth. The 2027/2028 auction cleared at the FERC-approved cap of $333.44 per megawatt-day and fell 6,623 MW short of the reliability target; subsequent auctions have continued to hit caps near $325/MW-day while remaining thousands of megawatts short. PJM’s independent market monitor has calculated that existing and forecast data-center demand accounted for tens of billions of dollars in added capacity costs across recent auctions, roughly 38–63 percent of certain wholesale capacity revenues depending on the year. Local outlets such as the Southern Maryland Chronicle accurately reported these results, and the inauthentic accounts frequently linked to or paraphrased those real articles while adding inflammatory visuals and hashtags such as #capacityauction.

Pattern matches with Open AI Report

This pattern matches an earlier cluster documented by OpenAI in June 2026. That “Data Center Bandwagon” operation involved accounts likely run by a private Chinese technology firm serving provincial-level government clients. Operators used VPNs to access ChatGPT, prompted the model in Simplified Chinese, and generated English-language talking points and comic strips that framed rising PJM capacity prices as a direct transfer of wealth from ordinary households to AI companies. The same visual tropes, greedy executives versus shocked families holding high electric bills, appeared in the examples X later published. OpenAI assessed the activity as small-scale and low-engagement at the time, rating it Category One on its breakout scale. X’s August disclosure indicates the network persisted or was reconstituted at larger scale.

Attributions

Attribution remains at the “suspected Chinese” and “likely PRC-origin” level rather than a publicly confirmed Ministry of State Security or United Front Work Department directive. Open-source indicators, Simplified-Chinese prompting, VPN use, commercial-contractor structure, and narrative alignment with official Chinese media commentary on U.S. AI energy costs, point to the commercial “public-opinion guidance” ecosystem that supports Party-state priorities. Similar activity has been observed from Russian and Iranian state media amplifying the same U.S. data-center grievances, suggesting a broader authoritarian interest in treating American infrastructure debates as a “domestic fracture point.”

AI Race - US vs China

The strategic logic is straightforward. The United States and China are in a high-stakes race to scale large language models and supporting compute. Data-center construction in Northern Virginia, Maryland, and Ohio is a binding constraint on U.S. AI progress; energy availability, interconnection queues, and ratepayer backlash already delay projects. Amplifying localized cost and reliability concerns can generate political pressure for moratoria, stricter interconnection rules, or cost-shifting mandates that slow American build-out without requiring Beijing to match U.S. capital expenditure dollar-for-dollar. Simultaneously, the operation tests whether generative AI can produce culturally fluent, visually compelling content that evades platform detection better than earlier “Spamouflage” slop. The use of American models against American infrastructure debates is itself a form of technological jujitsu. Effectiveness remains limited.

X and OpenAI both noted that the identified clusters generated little authentic engagement. Capacity-price increases and grid-stress reports predate the bot activity and would exist regardless of foreign amplification. Yet the operation’s value to an adversary is not necessarily immediate persuasion; it lies in volume, persistence, and the ability to launder talking points into local-looking accounts that real citizens then argue with. Over time such activity can raise the noise floor around a legitimate policy problem, complicate consensus-building on cost allocation, and consume the attention of U.S. policymakers and platform trust-and-safety teams.For international threat-intelligence consumers the episode underscores several enduring features of contemporary Chinese information operations: commercial outsourcing rather than purely state-run farms, opportunistic piggybacking on pre-existing Western controversies, dual-use of Western AI tools, and a focus on infrastructure and technology-competition narratives rather than crude election interference. Monitoring should prioritize detection of coordinated hashtag campaigns around regional transmission organizations, tracking of AI-generated political cartoons that reuse specific visual motifs, and mapping of contractor networks that serve both domestic Chinese public-opinion work and overseas influence tasks. Platforms, grid operators, and energy regulators will need closer information-sharing if they are to distinguish organic ratepayer concern from engineered amplification without suppressing the underlying policy debate.

The energy constraints on AI are real; the question is whether foreign actors will be allowed to set the terms of that debate.