Trump and Sam Altman smear data-center opponents as Chinese dupes to protect the buildout.
It is true that a few stray social-media posts linked to China-generated accounts mentioned data-center electricity prices, and it is true that the United States and China are engaged in a genuine competition over AI capability — in the narrow sense in which a real geopolitical rivalry is always true. OpenAI did say it banned two clusters of ChatGPT accounts it believed likely originated in China, engaged in an “apparent covert influence operation” that produced, among other things, social-media content claiming data centers were raising electricity prices. X released its own findings; Senator Cynthia Lummis found it “not surprising” that the CCP was “working overtime to derail America’s lead in AI.” Fine. Take the disclosure at face value. Albert Zhang of OpenAI’s misuse team then clarified that nobody at OpenAI was claiming all the data-center debate came from foreign actors, and that some of the suspected posts had very few views — “like three views.” The trouble is that the campaign now deploying that rivalry as a weapon against domestic dissent has nothing to do with foreign influence and everything to do with the fact that majorities in state after state oppose the construction — well over sixty percent in Michigan’s polling and even larger margins where the polls have been run — and the firms whose valuations depend on compute expansion — OpenAI and Anthropic, alongside the hyperscalers whose capex story carries the same assumption (Google, Meta) — cannot let that opposition harden into policy before the midterm.
Now notice what the alleged operation actually said. The content is not that the data centers are poisoning the children. The content is the mundane observation that big new electric loads can raise electricity prices. That’s a claim out of a rate docket, not an op-ed. When a gigawatt-scale load — the kind that requires new transmission, new generation, new engineering, and somewhere to put the dispatch — asks to connect, it does not just appear at the outlet. Wholesale markets are set by what marginal units charge; the grid cost is passed somewhere. You are welcome to argue the size of the arithmetic — that is exactly what a rate-design process exists for — but the sentence “large data centers do something to power prices” is not disinformation. It is the premise of a very boring utility proceeding. The campaign has announced that boring proposition as a foreign operation and is now asking the most wakeful citizens in the United States to distrust the two-thirds of their neighbors who just quote it.
The engineering reality is straightforward. A hyperscale data-center campus draws power measured in tens to hundreds of megawatts — enough to supply a small city — and a modern AI training rack alone can consume well over a hundred kilowatts. The water consumption is comparable. Google’s environmental disclosures have placed its 2023 global data-center water use at roughly 5.6 billion gallons, with the company’s reporting identifying AI-training workloads as the principal driver of the year-over-year increase. The physical infrastructure — substations, transmission lines, cooling towers, diesel backup generators — is not metaphorical. It is a land-use decision that competes with agriculture, housing, and the existing grid. The political-economy question is not whether China is posting about it on TikTok; it is who benefits from the expansion and who pays the energy, water, and land-use costs.
The numbers deserve a place in the file. An EPIC-MRA poll had 62% of Michigan voters opposed to new data-center development in their state; a separate EPIC-MRA survey put unfavorable attitudes toward socialism at 51% and favorable at 22%. Put those two polls side by side and you have the entire argument in arithmetic already. The majority the administration calls “socialist dupes” is a majority that also tells pollsters it does not care for socialism by a margin of fifty-one to twenty-two. This is not a coalition whose persuasion lends itself to a single foreign-paymaster story. This is a coalition whose complaint begins with the bill and ends with the bill, and whose arithmetic the President is asking the country to replace with a foreign-flag emoji.
It is also, as the opposition has snowballed this summer, an unusually bipartisan one. Steve Bannon and Bernie Sanders and Alexandria Ocasio-Cortez are standing on the same side of a data-center moratorium — we covered that coalition last week — and Republican governors in Texas and Pennsylvania have been curbing the approvals in their own states. These are not the people whose conversion a CCP bot farm would buy. They are people who have read that the load of the new facilities is the size of a decent city and have asked, reasonably, who pays for the wire. The answer is not “trust us,” because the problem is not trust. The problem is the bills. The package to fix it is mostly not the architecture: honest interconnection queues; a national mapping of what FERC and the state commissions do with transmission-cost allocation; full rate-case treatment of contracts and grid costs and water withdrawals; and clarity on who gives the liabilities for all of it to ratepayers and who does not. In such a world, data-center tariffs look like the program of a regulated electric utility — open, contested, docketed — and the result is a slower process, but it is one regulators decide.
To counter concerns about electricity prices and water consumption, the pro-AI campaign has been pushing back with arguments apparently aimed at appealing to rural America. On Tuesday, OpenAI chief executive Sam Altman suggested in a podcast that new data centers were more efficient than almond farming. The next day, Commerce Secretary Howard Lutnick pointed to how much water cattle consume; in a CNBC interview, he laughed off the water issue, calling it “propaganda by our adversaries.” The engineering-substance discrimination here is simple: almond farming does not draw from the same transmission feed as the data center; cattle do not draw from the same aquifer in the same season at the same rate. Both lines are aimed at the same voter — the rural voter whose electricity, land, water, and grids are being appropriated by the buildout — and both lines substitute mockery of the citizen’s arithmetic for an answer to the citizen’s arithmetic. The water question is not a propaganda question; it is a hydrology question. The kWh question is not a CCP question; it is a tariff question.
The tactic is almost vintage. What is being run here is the Old Red Scare playbook: if ordinary resistance to a corporate build-out is causing the problem, you don’t have to prove the reality of the foreign plot against it — you only have to hang him with “the foreign.” The expectation is simply the stain. The Wall Street Journal’s editorial complement to the amplification calls this season “Hot Luddite Summer.” Let me be more careful with the word — the Luddites of 1811 and 1812 were textile workers who broke the machines that were breaking up their trade, not primitives; Brian Merchant’s Blood in the Machine describes their movement as a response to an employment bomb primed by the people who held the property and locked the wages. They were the few who felt the effect of a mechanical change — the same people this government is now treating as a national-security threat.
The line to the Department of Justice has been drawn into this acquisition. When a federal department tells a federal judge that rules of law that make building an AI industry in the United States more difficult “threaten national security and give a competitive advantage to foreign adversaries” — as the Justice Department inserted itself into OpenAI’s copyright fight with the New York Times — that is a kind of saying that your rights and your objections are a national-security threat. That is a strange way of thinking of a rate complaint as treason. But the desperation is real. President Trump warned communities that opposing data centers would “end up being backwards and poor” and that “China could not be happier with this anti Data Center movement.” The language is precise: it does not say China is funding the opposition. It says China benefits from it, which is the same argument the tobacco industry used for decades — the public-health campaign helps our competitors — except here the competitor is a nation-state and the product is compute, not nicotine. The framing converts a legitimate engineering question — what happens to a regional grid when a single hyperscaler draws as much power as a mid-sized factory — into a loyalty test.
Trae Stephens, a venture capitalist and co-founder of Anduril, posted on X: “It’s almost like not banning TikTok has had real world consequences as the U.S. population continues being poisoned by CCP propaganda.” The sentence conflates two things — TikTok as a platform and the specific data-center opposition — and treats the second as evidence of the first. If CCP propaganda is the explanation for data-center opposition, then the engineering questions about grid capacity, water rights, and land use must not be real. The bad-faith technique is the blanket half-truth: one true thing (foreign influence operations exist on social media) paired with an invalidating suppression (the legitimate, bipartisan, documented domestic opposition has a different cause), and an extrapolation (therefore the opposition is not authentic). The step from “we found some bot activity” to “China is behind the anti-data-center movement” is the same step that the Iraq War intelligence took: a real datum, a real threat, and a political interpolation that serves the interests of the party making it. The interpolation is the move.
Not all observers in the AI world are convinced. Andy Masley, a high-profile AI researcher who has fought to debunk concerns about data-center water usage, questioned the suggestion that the CCP was behind the campaign: “This looks much more like a general bot farm trying to get attention.” The strongest inside-the-beltway skeptic of the Chinese-influence story is a man whose day job is defending AI against the very water critiques the campaign is calling foreign. When the people trying to debunk the concern are also the people calling the foreign-influence story thin, you have arrived at a campaign whose own natural allies are walking out.
It will probably not work the way the President intends, and that is close to the trading point. Marjorie Taylor Greene — a former congresswoman who has had a public falling out with the president — posted on social media: “Good luck winning the midterms now that MAGA means mass data centers, flock cameras and surveillance state.” Greene is a Trump ally and a China hawk by any honest measure, and the “socialist dupe” frame is the least plausible label that could be hung on her. The coalition the administration is trying to discredit is not the Berkeley left; it is the rural ratepayer and the suburban mayor and the Republican governor. The discrediting is failing in the wrong places first.
The defense of the AI boom — the stock valuations, the power purchase agreements, the whole arrangement — has been mounted on the pile. Key to the lofty valuations being assigned to the likes of Anthropic and OpenAI is the assumption of massive growth, growth that can only occur if more data centers are opened to build and fuel AI’s expanse. Galbraith called this kind of hidden gap between the books and the market the bezzle — the magic interval when the discrepancy is known to the insider and not yet priced by the public. If a claim that the market would collapse is the only pile holding up the model, the freedom is too precious; whatever the actual scale of CCP-linked accounts, the protection being mounted is for a stock price, not for the republic.
The Innis staples-economy frame applies with uncomfortable precision. Harold Innis argued that the resources a society extracts determine its political economy. Data is the newest staple. The hyperscalers are the extraction apparatus. The data center is the mill, the mine, the grain elevator. The question is not whether extraction will happen — it will — but who controls it and who pays. The CCP frame is the political economy’s way of saying: do not ask the question. The answer is in the filing, not in the post. The filing says national security. The post says China. The filing is the law. The post is the story they tell to make the law go through.
The paternalism is the point, not the China.