They sold the town’s light and called it freedom, and now the town is being asked to host the machine that took it; when the town finally answered, the money called its answer a market correction. In a column in The Guardian, Heather Stewart reports that AI and data-center expansion became ballot questions in the 2026 midterms: dozens of measures calling for limits on expansion passed, a technology sell-off unsettled the industry’s financial story, and voters delivered a warning to the men who wanted the public to regard the buildout as inevitable. At a Texas campaign event, Senator-elect James Talarico recalled a mother telling him that the lights from the neighboring data center were so bright her son thought it was daytime in the middle of the night. Stewart is right about almost all of it. Her comparison with Theodore Roosevelt points toward the wrong instrument, and the instrument is the whole argument.
In Lowell, Massachusetts, the Markley data center stands in a black slab among the houses, and some of the people next door have taken to calling it “the Dungeon.” The machine was never weightless. It wants land, substations, construction crews, cooling systems, contracts, and a place where a horizon can be reduced to an entry in a distant developer’s spreadsheet. It wants the grid, the water, the roads, the labor, and the law, while reserving for its owners the right to call every local cost an unfortunate side effect.
Stewart’s strongest point deserves to be stated in its strongest form before anything is said against it: “Ultimately, all businesses in a democracy – even those embraced by the US president – have to operate within the bounds set by society.” The buildout’s creed had been that private accumulation was the delivery mechanism for the public good. We were to set aside our qualms about the sheds, the power, the water, and the light because a greater good was coming; instead, we watched the men who run the servers grow rich and were asked to treat the arithmetic as a favor. It is possible that AI will make search easier, spare some people dreary work, and that large facilities can sometimes deliver a service more efficiently than many small ones. None of that entitles the industry to presume the consent of the settlement around it. That is not a grievance against a machine. It is the oldest story in the country: somebody owns the toll road and calls it freedom of passage.
The public face of the same buildout has included the sea of online slop and technology firms shoehorning AI into every app, platform, and device—the process Cory Doctorow calls “enshittification.” The user is made to carry the worsening while the owner keeps the revenue.
Then there is the sentence in Stewart’s column that ought to trouble its admirers more than it troubles me—the one about the people outside the ecosystem, “regular people” as she has to call them now, not being that interested in AI. Dario Perkins is right about the substance: the people who worked in tech and finance did not realize that the rest of us were never going to organize our lives around a chatbot, and the insiders were living in a bubble in at least two senses. But “regular people” is a way of describing a constituency, and a constituency is a thing you court; a community is a thing you answer to. That difference is the entire moral argument. One is satisfied when the vote goes the right way. The other is satisfied when the power that reorganizes a place is answerable to the people who live in it.
Her analogy to Theodore Roosevelt points toward an implicit turn toward federal trust-busting: when private power grows, reach for public power. That is the oldest instinct in the business of reform, and it deserves more respect than a postscript on an election column usually gets. A republic does have tools beyond local zoning boards and lawsuits for restraining firms that cannot be refused. But to place the burden of every remedy in one federal center is to mistake concentration for a remedy. The answer cannot be reduced to Washington or to Silicon Valley. It begins at the county board, the zoning hearing, the water district, and the electric co-op.
The mistake is the same on both coasts of this argument: when private power grows, reach for public power. A giant corporation and a giant regulator can both become private governments with no place left in the room. The honest conservative position—the one the whole tradition of subsidiarity was built to state—is that the smaller competent body is the right one. Pius XI called it “a grave evil” to take from lesser organizations the work they can do themselves. Michael Oakeshott preferred “the familiar to the unknown, the actual to the possible.” A county board that has lived with its own power line for a hundred years is not the unknown.
I traded agricultural futures on a Chicago desk, where the crop was a promise and the farmer was an abstraction. From inside that machine, the trick now being run is plain. It has a name—circular financing—and the name is the warning: capital, vendors, and purchasers can be linked in ways that make booked demand look firmer than it is until somebody outside the circle actually buys. When I came home to manage the farm cooperative in Friendship, the scale changed; no forecast could tell a member whether the co-op had enough cash to meet its bills and keep its machinery in repair. A market is useful when it bears part of the cost of its promises. It becomes predatory when somebody else owns the road, the warehouse, and the terms of every turn.
The International Monetary Fund said last month that there was “growing evidence” the successful rollout of AI “may not require investment on the massive scale being undertaken by ‘hyperscaler’ firms,” that “potential economic gains may not be as large as expected,” and that circular financing had become rife. Dario Perkins put the arithmetic more starkly: roughly $200 billion in current revenues supporting more than $1 trillion in future commitments from Anthropic and OpenAI. The tech-heavy Nasdaq had peaked a month earlier. Paper can move in an instant. Concrete has to set.
The market has begun to read the fine print. AI shares have rebounded from a rout that erased more than $2 trillion as data-center votes unsettled investors, while AI companies are reassessing their data-center plans. Corporate users are already hedging, limiting staff use of costly products such as Claude or turning to cheaper models hosted on their own servers. Capital can change its mind in an afternoon. A transmission line, a transformer, and a neighborhood bearing their consequences remain.
And the bill for the machine keeps arriving at the door of people who never chose it. In Britain, MPs’ offices, local authorities, and schools are receiving floods of chatbot-generated grievances, requests, and legal challenges that nobody budgeted for; the convenience of the machine is paid for in the hours of the clerk who has to read it. That is what this technology does to a community. It arrives as a free convenience with the bill addressed to whoever is nearest. The question is not whether the machine works. The question is which institution gets to answer for it.
Here is where my own side fails, and I will not pretend otherwise. For two decades, the conservative movement I once believed in made a virtue of deregulation, called the data center “infrastructure,” and sold the country the idea that a shed full of servers was the same kind of thing as a grain elevator. It is not the same kind of thing. One of them is where your food is; the other is where somebody’s money is. The fusionism of those years was not incidental. A movement that promised to bring power and authority back to the towns sold the towns’ permission cheap, because cheap permission was what the shed-builders were buying.
Conserve what, exactly?
Louis Brandeis wrote a chapter called “The Curse of Bigness” in 1914, and he was thinking about railroads. The curse of bigness is not a complaint about efficiency. It is a fact about power: a firm too large to be refused has already won, whatever the merits of its case. A business institution severed from its place is not merely big. It is a private settlement presuming the consent of a public world.
But the midterms do not tell you the thing that matters. The counties that will be offered these sheds are not the counties best placed to refuse them. They are the ones that have already lost the mill.
My own Adams County is a railroad town. The Chicago & North Western laid its line through the sand in 1911 and built a city out of the yard, housing the workers in boxcars; the last Twin Cities 400 ran on July 23, 1963; Union Pacific absorbed the line in 1995. Across the region, the paper mills went next—Port Edwards in 2008, Whiting in 2010, and the Verso mill at Wisconsin Rapids idled on June 9, 2020, taking roughly nine hundred to a thousand jobs with it. By the 2020 census, nearly half of Adams County’s housing units stood empty, most of them seasonal, and the USDA classified the county as a high-recreation retirement destination. In 2017 its labor-force participation rate was 48.6 percent, the lowest in Wisconsin. The sand barrens that would not grow a decent crop have been sold, at last, as a view: Sand Valley, the golf resort on the barrens, reportedly employing about 500 people, with its lake named Lake Leopold.
Mellencamp put the answer to that question in a song forty years ago, and I have never had a better use for him: “No I cannot forget where I come from.” A data-center developer has figured out what my neighbors have understood for a hundred years—the answer to that question is worth an enormous amount to somebody who is not from here.
So do not tell me the public said no. The public says no where it has a city council, a newspaper, a union hall, a wage worth defending, and a neighbor who will get up for the hearing. The public says yes where the alternative is another empty storefront. A family that votes for a data center in a county like Adams is not surrendering to the machine; it is voting for a payroll. The market for permission will be won by the places with the least capacity to refuse.
We were not raised to shut the door on the stranger; the parish taught us to set a place at the table. The county has lost its work, not its manners. What it will be offered is a check, while the machinery of the new economy comes and goes across the land. A host can invite the stranger to the table. It cannot surrender the family’s name, its fields, or its future to whoever buys the meal.
Here is the answer, and it is harder and better than a trust-bust.
Not the regulator; the co-op.
In my own town of Friendship there sits the Adams-Columbia Electric Cooperative, a member-owned utility with about 31,560 member-owners across twelve counties, the largest rural electric co-op in Wisconsin, governed by a seven-person board the members elect. That is not nostalgia; it is the receipt. When the federal government wanted rural America electrified in the 1930s, it did not hand the wires to a holding company. It wrote low-interest loans and routed them through member-owned cooperatives; by the end of 1938, some 350 cooperative projects in forty-five states were serving about 1.5 million farms. The money was public. The ownership was local. And the people who took the wire were the people who lived behind it.
Where the model has gone big, it has stayed honest in the way that matters. Mondragon, in the Basque Country, now has about 70,000 employees and more than €11 billion in annual sales, with a pay ratio near five to one. It is honest, too, to name the failure mode. Cooperatives drift toward management, and the degeneration is real: a 2022 review of 83 enterprises found that 63.5 percent resisted or regenerated and only 9.5 percent fully degenerated. The instrument is not perfect. It is the one instrument whose logic runs in the right direction: the more members, the more owners, the harder it is for anybody to own the town instead of living in it.
I come to the cooperative tradition the way I come to most of my localism, and I should say what I am not taking. The Southern agrarians of 1930 wrote a beautiful case for the land in the shadow of the Lost Cause; the rural order they defended bore the stain of segregation and white supremacy, and they could not bring themselves to name it—or the enslaved people on whom it rested. I do not take my localism from them. I take it from Wendell Berry, who wrote of solving for pattern, and from the encyclicals, which hold that the earth was given for all and that property carries a social mortgage. A member-owned utility says that in its ordinary business. A shed on leased ground under a long-term supply agreement says the opposite.
Apply that grammar to artificial intelligence, with the scale kept honest. Not every town should host a data center. But every affected town should have a binding local vote on siting and expansion, a member-owned stake in the local physical infrastructure, and clear rules assigning operating and repair liability. Regional computing cooperatives could own feeder lines and physical sheds; schools, hospitals, libraries, small employers, and households could buy capacity through member accounts; worker-owned firms could maintain and operate the systems; outside technology companies could lease power and space without owning the local decision or taking the whole surplus. A locally owned public option could serve communities too, provided it remains contestable rather than becoming the local monopoly it was meant to replace. Privacy, labor standards, competition, due process, safety, and interoperability would be floors, not favors; a floor is not a landlord. This is distributed ownership, not another sovereign computer built to replace the present one.
It is harder here. A data center is not a grain elevator. Capital, engineering, grid planning, and global software competition do not bend to a parish resolution. A co-op cannot conjure hyperscaler demand or live on good intentions. It would need patient public credit without public ownership, routed through member-owned lending institutions or locally authorized entities, with capped exposure, published terms, no operational control, and the first loss left with members or operators. It would need anchor members willing to sign long-term commitments, phased construction, independent engineering, published assumptions, member approval for each expansion, and clear rules for who bears the loss when the forecast is wrong.
Concretely, the member-owned electric co-op would negotiate and hold the power and capacity rights under a board-approved contract, with residents voting on the terms; the water district and a land trust would hold their respective pieces where the law allowed them to do so. The local members would need to see the books, the engineering assumptions, the contracts, and the liabilities. They would need the power to say no to the next phase. A member-owned utility has no balance sheet for a hundred-megawatt campus, no engineers at San Francisco prices, and no way to outbid a county desperate for a payroll. That does not mean it should surrender. It means the county should not sell the public’s future to buy its way back from vacancy.
If the only viable version requires one federal compute czar, we have changed the logo on the warehouse. If the only viable version socializes the gamble while privatizing the returns, we have changed nothing.
The November vote did not save any town. It only restored, for a moment, the town’s standing to say no before the concrete was poured. We should not mistake a pause for a settlement. But a settlement begins where the people who must live with the decision have a share in the power, the surplus, and the repair.
The machine may come. Let it come as a guest with a share in the door and the power bill, not as the landlord of the town. The light belongs to the town. Leave the town its life.