OpenAI and Oracle are dumping their AI bill on local communities.
Two months after the companies broke ground on a massive data-centre campus in Saline Township, Michigan, Abdul El-Sayed held a campaign rally outside the site. He had narrowly won his primary earlier that week, and the project was useful as a backdrop: a giant, windowless building filled with servers, backed by enormous demands for electricity, land, water, and public accommodation.
The trouble is that “AI infrastructure” makes those demands sound like weather. It is not weather. It is a construction decision, a power-allocation decision, a land-use decision, a tax decision, and eventually a household-bill decision. Somebody chooses where the facility goes, who pays for its connections, which communities absorb the noise and pollution, and whether the promised economic benefit is specific enough to be checked.
To be fair, data centres are infrastructure. The servers have to be housed somewhere, and the current generation of large language models requires substantial computing capacity. The trouble is that the companies receive the concentrated upside while communities are asked to accept the diffuse cost as the price of being modern.
That is why a Gallup poll earlier this year found that seven in ten Americans oppose construction of an AI data centre in their area. The number does not prove that every proposed facility is unjustified. It does prove that the industry has a credibility problem too large to be dismissed as a few rural complaints or an internet-era misunderstanding of technology.
The political response is arriving before the policy response. El-Sayed has called for federal guardrails and argued that corrupting money in politics is what allows projects to be imposed on local communities. In Ohio, Republican gubernatorial nominee Vivek Ramaswamy released an “Ohioans-first data center pledge” calling for legislation and, until it passes, a moratorium. Democratic candidate Amy Acton had already proposed a “conditional moratorium.” In Wisconsin, Francesca Hong has made a full moratorium central to her campaign for the Democratic nomination for governor.
This is not a tidy partisan story. It is a power story.
Amanda Litman of Run For Something says the concern appears in urban, rural, suburban, red, blue, and purple communities. That makes sense. The data centre is the most physically legible part of an otherwise abstract AI economy. The model is invisible. The server building is not. Neither is the electric bill.
The candidates are responding to a material contradiction the industry has preferred to leave abstract. The companies describe artificial intelligence as a national economic transformation, but the facility arrives in a particular township, beside particular homes, connected to a particular grid, using public roads and public permitting systems. The benefit is described at national scale. The cost is delivered locally.
This is the old infrastructure trick with a newer noun attached. The public is invited to admire the aggregate investment while being told to negotiate individually with the noise, the power demand, the property pressure, and the uncertain employment promise. The relevant question is not whether the project creates economic activity. A construction project necessarily creates economic activity. The question is who captures the durable value after the ribbon-cutting, and who remains responsible for the durable costs.
A job is not the same thing as a durable local benefit. A tax payment is not necessarily a fair payment when public authorities must also provide roads, transmission capacity, emergency services, and administrative machinery to support the project.
The extractive dynamic follows a pattern. When the forces that once restrained corporate behaviour — competition, regulation, user exit, and labour power — are weakened or removed, the company does not need to become morally worse. It only needs to discover that the friction has gone.
The same logic applies to the data-centre buildout. A company that must pay the full cost of power, connection, land, water, cleanup, and local disruption has a reason to build carefully. A company that can secure subsidies, negotiate special arrangements, shift grid costs, and treat public resistance as an obstacle to be managed has a reason to build quickly and explain later.
The explanation later is already familiar. The industry will point to jobs, investment, competitiveness, and the future. Those benefits may be real. They are not self-proving, and they do not answer the distributional question.
The Ohio debate shows the political mechanism plainly. Ramaswamy says residents are concerned about rising electric bills, noise, pollution, and the absence of clearly defined economic benefits for Ohio families. He has proposed that data-centre companies cover the power bills of nearby residents. The proposal is blunt, but the underlying question is not: why should households that did not request the facility bear costs created by a private project?
The answer cannot be that the project is large. Scale is not a public purpose. Nor can it be that the technology is important. Importance is not an exemption from accounting.
The Main Street Independent has already documented how power companies are using eminent domain to seize land for data-centre transmission lines. That is the architecture beneath the campaign rhetoric: private demand translated into public compulsion. Once the power system is rearranged to serve the facility, the affected landowner is no longer negotiating with a company alone. The company has acquired the state’s power to take land.
The farming story makes the same point from another direction. Farmers and ranchers are warning that data-centre construction threatens farmland, which is not a sentimental objection to technology. It is an argument about what land is for, who gets to change its use, and whether food-producing communities should be expected to surrender productive ground for a speculative computing boom whose returns flow elsewhere.
The technical framing matters because “AI” is doing too much rhetorical work. A data centre is not an intelligence. It is a physical facility containing computing equipment, connected to power and communications infrastructure, operated through contracts and management decisions. The building does not decide to consume more electricity. People decide to build it, finance it, connect it, subsidize it, and approve it. Calling the demand an inevitable consequence of progress is simply a way of removing the decision-maker from the sentence.
That is why the AI economy cannot be discussed as software alone. Its physical substrate is land, electricity, construction, equipment, and labour. The data-centre campaign is therefore not a side dispute about local aesthetics. It is a dispute over who controls the physical infrastructure of the AI economy and who bears its costs.
There is a genuine concession for the industry here: communities cannot demand every facility be built somewhere else while also demanding uninterrupted access to the computing services those facilities provide. Someone will bear the material footprint. But that does not identify the someone, and it does not establish that the burden should fall on the communities with the weakest bargaining position.
The remedy is not a slogan about stopping technology. It is a public rule that makes the beneficiary carry the cost: no construction without transparent power accounting, enforceable limits on noise and pollution, direct compensation for affected households, and a public record of the jobs and revenue actually delivered. Moratoriums are reasonable while those rules do not exist. Open standards for the information, contracts, and grid obligations would make the claims checkable rather than leaving each town to bargain in the dark.
The political candidates have found the pressure point because voters have found it first. The data centre is where the AI promise becomes a transformer, a property assessment, a transmission line, or a monthly bill. It is difficult to sell a frictionless future to people standing beside the substation.
The companies can build their machines wherever they like. They cannot make the cost disappear. They can only send it to somebody else.