The Cloud Has to Land Somewhere
Whose water, whose power, whose backyard? Inside the real cost of the $650 billion AI buildout—and the fight against it.
by Lawrence Winnerman
The word we were given was cloud.
It is one of the most effective pieces of marketing in the history of technology, because it tells you the thing is weightless, floating, somewhere up there and nowhere in particular—a place your photos live, made of nothing, costing nothing, sitting on no one.
Ask Beverly and Jeff Morris what the cloud weighs.
They live in Newton County, Georgia, in the kind of rural country people from the cities drive through without seeing. In 2018, Meta broke ground on a data center about a thousand feet from their home. Within months, the Morrises’ well—the private well their household actually drinks from—began to fail. Sediment in the water. The dishwasher, the ice maker, the washing machine, the toilet, all faltering. They have spent roughly $5,000 trying to fix it and can’t afford the $25,000 it would take to replace the well. Meta commissioned a study and concluded its data center was “unlikely” to have affected their groundwater. Three of the Morrises’ neighbors have reported well trouble since the data center went in.
The cloud, it turns out, has to land somewhere. It landed on them.
This is an essay about where the cloud actually touches the ground—and about the fact that the people it’s landing on hardest are not the people you’d expect, not the laid-off coders in the blue cities, but the farmers and retirees and well-water families in rural America who never signed up to host the most resource-hungry machine ever built.
It is also an essay about the most interesting political possibility I’ve seen in years, which is what happens when those two groups of people figure out they’re fighting the same thing.
The Thing the Recession Is Hiding Inside
I wrote, in an earlier piece in this series, that the AI buildout is the only thing currently propping up the American economy—that strip out the estimated $650 billion in data-center capital spending and the non-AI economy is flat on its back. That’s the macro picture: the buildout is the cushion hiding the recession.
Here’s what that abstraction looks like when it lands in a county.
The “buildout” is not a number on a balance sheet to the people who live next to it. It is a building the size of a small town, running twenty-four hours a day, sold to their county commission as jobs and tax base and the future. What arrives instead is a windowless industrial complex that employs a few dozen people once construction ends, draws the electricity of a city, drinks from the same water table the farms and the wells draw from, and never sleeps.
The recession is hiding inside the buildout. And the buildout is hiding inside their backyard.
Sponsored by Ground News
A quick word on how to read a story this big.
You are going to see this data-center fight told in completely different ways depending on where you get your news. The O’Leary project (below) alone gets framed as a patriotic national-security win on one outlet, an ecological catastrophe on the next, and a “Chinese influence operation” on a third—same project, three different realities, depending on which feed you happen to live in.
The tool we use to see all of it at once is Ground News. They pull a single story together from across 50,000+ outlets and lay the bias breakdown out on a bar—left, center, right—so you can see who’s covering it, who’s spinning it, and who’s quietly ignoring it. Their factuality ratings flag which sources actually check their work. On a story this contested, reading it through one outlet is how you get played.
See every angle and get 40% off unlimited access: ground.news/blue.
Water
Start with water, because water is where this gets visceral, and because it’s where Erin Brockovich—yes, that Erin Brockovich—has planted her flag.
Data centers run hot. The machines inside them throw off enormous heat, and a great deal of that heat is managed with water—evaporative cooling that consumes the water rather than returning it. A large facility can draw up to 5 million gallons a day, the daily water use of a town of tens of thousands of people, and as much as 85 percent of it evaporates and never comes back to the local supply.
In 2023, U.S. data centers consumed an estimated 17.4 billion gallons of water; the EPA projects that figure could reach 38 to 73 billion gallons by 2028. In a wet year, in a place with water to spare, you might never notice.
But these facilities are not only being built in wet places, and the years are not reliably wet anymore.
Then there is the version of the story where the rules simply don’t apply to you. Twenty miles south of Atlanta, in Fayette County, a data-center campus developed by Quality Technology Services—owned by the private-equity giant Blackstone—drew roughly 29 million gallons of water through two connections the county utility didn’t even know existed. One pipe had been installed without the utility’s knowledge; the other was never tied into the billing system. The unmetered draw ran for about 15 months, and came to light only when residents started complaining about low water pressure at their own taps—some of them having been told, at the time, to stop watering their lawns.
It became public when a local attorney, James Clifton, obtained the utility’s letter to the company through a public-records request. The company owed about $147,000 in retroactive charges. The county declined to levy any fine. The explanation from the county water-system director, Vanessa Tigert, when asked why not, is the single most clarifying sentence in this entire essay: “They’re our largest customer, and we have to be partners. It’s called customer service.”
There it is, in one official’s words: the public’s water, taken outside the rules, at a scale no household could ever match, and the regulator’s instinct is not to enforce the law but to apologize to the violator for the inconvenience of being noticed. The pattern repeats wherever this lands. A facility consumes a public resource at industrial scale, the rules that are supposed to govern that consumption turn out to be soft or unenforced, and the cost—the dropped water table, the fouled well, the dry season made drier—is socialized onto the people who live there while the benefit is captured a thousand miles away.
This is the shape of story Erin Brockovich built a career on: a community, a stolen or poisoned water supply, a powerful entity that assumed no one in a small place would fight back.
And she has now turned directly to data centers.
In the spring of 2026 she launched a public reporting map, asking residents to self-report data-center problems in their own communities—and drew nearly 4,000 submissions in the first month. The single most common complaint, she reported, wasn’t noise or water or the utility bill. It was one word that kept appearing in submission after submission: transparency.
Nobody asked these communities first.
Heat
Here is the cost almost no one names, because it is the most counterintuitive of all: the heat itself, and the water that is gone for good.
Follow the water one step further than we did a moment ago. When a data center cools itself by evaporation, that water is not borrowed and returned a little dirtier, the way it is when you run a tap. It is consumed.
Up to 85 percent of it rises into the sky as vapor and does not come back to that aquifer, that river, that town—not this season, maybe not ever. The water it does eventually discharge is often so loaded with the biocides and anti-corrosion chemicals used in the cooling loop that it’s unfit to drink or to farm with. In a wet place you can pretend that’s a wash. In a desert, it is a one-way withdrawal from an account that was already overdrawn.
Then there is the heat, and the heat is pure physics. A data center does not make energy vanish; it converts it—essentially all of it—into heat. This is just the first law of thermodynamics: the electricity that pours into the building has to go somewhere, and where it goes, almost in its entirety, is back out as waste heat. A facility drawing a gigawatt of power is, functionally, a one-gigawatt furnace running in open country, around the clock, forever.
Now look at what Kevin O’Leary is building. O’Leary—“Mr. Wonderful” from Shark Tank—is the public face of the Stratos Project, nicknamed “Wonder Valley,” a data-center campus planned for some 40,000 acres of high desert in Box Elder County, Utah, just north of the already-shrinking Great Salt Lake.
The buildout could run past $100 billion and is designed for around 9 gigawatts of power—more than double what the entire state of Utah uses in a year.
A physics professor at Utah State University, Robert Davies, estimates that once you account for the inefficiency of the on-site natural-gas plant meant to power it, the real thermal load approaches 16 gigawatts—and that dumping that much heat into one valley could raise local nighttime temperatures by 8 to 12 degrees.
Hold that number against the most infamous yardstick we have.
By Davies’s estimate, the heat this single campus would shed amounts to something on the order of 23 Hiroshima bombs’ worth of energy—every day.
I want to be precise, because the comparison is easy to abuse: this is total energy, released slowly as heat over twenty-four hours, not a blast in a single second over a city. Nobody is detonating anything. But the energy bookkeeping is real, and it is poured, hour after hour, into one fragile, water-starved place that never agreed to absorb it.
This is the ecological cost that water and electricity figures alone miss. It isn’t only that the machine drinks the water and bids up the power. It’s that it takes the water permanently and bleeds a city’s worth of heat into a small, dry, living place—near a dying lake and a migratory-bird sanctuary—on a timeline measured in decades. And the people who can see that most clearly are the ones who live there. In May 2026, hundreds of Box Elder County residents packed the fairgrounds chanting “Shame! Shame! Shame!” as commissioners voted the project through anyway; one of them, Brenna Williams, who is helping lead a referendum effort to overturn the approval, put it plainly: “We’re already being told to ration our water.”
When O’Leary responded to the opposition by suggesting on Fox that his critics were “proxies for the Chinese government”—singling out two Utah political consultants, Gabi Finlayson and Jackie Morgan—Finlayson’s reply wrote itself: “The only foreign operative here is a Canadian.”
Electricity
Then there’s the power, and this is the one that reaches even the households that never see a data center.
These facilities consume staggering amounts of electricity, and that demand doesn’t stay politely contained. It bids against you. A single large AI data center can require more than a gigawatt of power—as much as a major city, or a large nuclear plant—and when that demand hits the grid, the cost of supplying it lands on everyone’s bill. The clearest place to see it is PJM, the grid operator that serves 13 states and the District of Columbia. In its 2025–2026 capacity auction, prices didn’t tick up—they detonated, an 833 percent jump from the year before, from $28.92 to $269.92 per megawatt-day, the sharpest single-year increase in the market’s history.
Those costs flow straight through to consumers.
The Natural Resources Defense Council found that data-center growth accounted for $9.3 billion of a $14 billion regional capacity bill, translating to roughly $70 a month for a typical household, and projected $100 billion to $163 billion in cumulative ratepayer costs through 2033 unless regulators intervene. In Washington, D.C., Pepco residential customers saw bills rise about $21 a month starting in mid-2025. Across the PJM region, supply rates have climbed between 5 and 44 percent since June of that year.
So the family that has never heard the word “hyperscaler” still pays for it, in the line item on the utility bill that keeps climbing for reasons no one at the kitchen table can see.
And the demand is so enormous, so fast, that it is bending the entire energy trajectory of the country backward. In the PJM region, utilities have delayed or canceled the retirement of roughly 60 percent of the fossil-fuel capacity that had been scheduled to shut down; at least 15 coal plants have had their closures pushed back since the start of 2025 alone, to feed rising demand. The cleaner energy future we’d been promised is being quietly deferred to run the machine.
And in the single most vivid emblem of the moment, the Three Mile Island nuclear plant—site of the worst commercial nuclear accident in American history—is being brought back from the dead. No, really.
Its Unit 1 reactor, retired in 2019 and now rebranded the Crane Clean Energy Center, is being restarted by Constellation Energy under a 20-year deal to sell its entire output to Microsoft, to power AI. The U.S. Department of Energy closed a $1 billion federal loan to help make it happen. It is on track to restart around 2027.
O, brave new world! We are restarting Three Mile Island to run chatbots.
If you wanted a single image for deploying a technology faster than we can think it through, you could not invent a better one than reopening the most infamous nuclear site in the country—with a billion dollars of public credit—to feed a machine whose business case is still, charitably, unproven.
Where It Lands, and on Whom
Now the part that I think matters most, the part the national conversation keeps missing.
We have been telling the story of AI and jobs as a blue-state, white-collar story: the laid-off marketing coordinator in Portland, the legal researcher in Cleveland, the coder in Austin. That story is real—I’ve written it myself. But it is only half the map.
The other half is rural, much of it red, and for those communities the impact isn’t an abstraction arriving in a quarterly earnings call. It is physical. Roughly two-thirds of new data centers are being built in rural areas, much of it on former farmland—developers have offered farmers as much as $120,000 an acre or more, several times the going rate, which sounds like a windfall until you realize it multiplies the price of the land around it and puts expansion permanently out of reach for the family farm next door. It is the constant low roar of cooling systems where there used to be quiet. It is the well, and the water table, and the electricity bill. It is the transformation of an exquisitely beautiful place into infrastructure, decided by a county commission dazzled by a tax-abatement deal, on behalf of a company headquartered in a city those residents will never visit.
This is the rubber-meets-the-road moment of the AI economy, and it is happening in exactly the communities the coastal AI conversation has trained itself to ignore. The blue-city knowledge worker loses a job and feels the future as a closing door. The red-county farmer watches the future bulldoze the back forty and drain the well. Same machine. Same handful of companies. Same direction—pointed at extraction, with the gains captured far away and the costs left on the land. Different victims, who have been carefully taught to believe they are on opposite sides.
They are not on opposite sides. That is the most important sentence in this essay.
The Coalition Hiding in Plain Sight
Here is the political possibility, and it is a big one.
Across the country, rural communities are already organizing against data centers—forming groups, packing county-commission meetings, challenging zoning variances and tax abatements, demanding environmental review, and in a striking number of cases, winning.
In Tucson, Arizona, residents raised enough alarm about water and energy that a proposed Amazon project was abandoned.
In Saline Township, Michigan, residents voted down a $16 billion OpenAI-and-Oracle project.
In Mason County, Kentucky, more than 1,500 residents have been turning out to plan opposition and pack hearings, and farmers there rejected a multimillion-dollar offer for their land.
This is not happening because those communities read the same Substacks as the laid-off coder in Portland. It’s happening because the cloud landed on them and they could see it.
What makes this the most interesting development in American politics right now is that it is genuinely, structurally cross-partisan—and it’s already being noticed. A March 2026 Gallup survey found that 71 percent of Americans oppose the construction of an AI data center in their own community, with nearly half strongly opposed—the first time Gallup had asked the question, and it arrived higher than opposition to a nuclear power plant has ever registered in the quarter-century Gallup has tracked it.
Can you imagine? Americans would now rather live next to a nuclear reactor than a data center.
In May 2026, even The New York Times ran a piece on liberals and conservatives finding common ground against the buildout. The fight does not care how you vote. The retiree in red Georgia whose well filled with sediment and the union organizer in a blue city worried about displacement are, whether either of them knows it yet, fighting the same concentration of power, in the same direction, for the same reason.
The thing that has kept them apart is not their interests. It’s the story they’ve each been told about the other.
Erin Brockovich is a useful patron saint for exactly this reason. Her whole career has been small communities—not left communities or right communities, just communities—discovering that a powerful entity has quietly decided their water doesn’t matter, and refusing to accept it.
That is a politics older and deeper than the red-blue divide, and the data-center buildout is generating it at scale, in real time, in places the political class isn’t watching.
Sponsored by Ground News
One more thing, because it’s the whole point of this piece.
I argued above that the rural-red water fight and the blue-city labor fight are the same fight — and that the thing keeping those people apart is the story each was told about the other. That’s not an accident. It’s what happens when your news comes from one side of the aisle and the other side’s coverage is simply invisible to you.
Ground News built a feature called Blindspot for exactly this: it surfaces the stories the left isn’t covering and the stories the right isn’t covering, side by side, so you can see what your own feed is hiding from you.
The data-center buildout is a textbook case—covered hard by local and progressive outlets, underplayed by the national press that should be all over it. If you want to break out of the bubble that keeps this coalition from forming, start here: ground.news/blue —40% off unlimited access.
What to Do
If you’ve read the other pieces in this series, you know I refuse to end on dread.
So here is the turn to action, and for this fight it is unusually concrete, because the data-center buildout is one of the few pieces of this whole AI story that can actually be fought—and beaten—at the local level.
You cannot personally slow the frontier labs. You can absolutely show up at a county-commission meeting.
Find out what’s proposed near you. Data-center projects move through local zoning and permitting before they’re built. That process is public, and it is where these fights are won or lost—on the tax abatement, the water permit, the environmental review, the zoning variance. The deal is usually struck before most residents have heard about it. Hearing about it early is half the battle.
Connect to the people already fighting. Erin Brockovich has built a hub specifically for this: an interactive map of data centers across the country—operational, under construction, and proposed—overlaid with the locations where residents have reported problems, plus a form to report what’s happening in your own community. If there’s a project near you, you can find it; if there’s a problem, you can put it on the map. Start there: brockovichdatacenter.com.
Build the coalition that scares them. The single most powerful thing about this fight is that it crosses the aisle. If you are a progressive reading this, the most useful thing you can do is show up for a rural community’s water fight without making them pass a political purity test first.
The concentration of power that drains their well is the same one coming for your livelihood.
Act like it.
The Cloud Has to Land Somewhere
There is no cloud.
There was never a cloud.
There is a machine, and the machine is made of concrete and copper and water and an appetite for electricity that is reshaping the physical landscape of the country, and it has to be built somewhere, and “somewhere” is turning out to be the quiet, beautiful, underdefended places where people assumed no one important was paying attention.
The companies building it are counting on the same thing every powerful entity in a Brockovich story has ever counted on: that the place is too small, too rural, too scattered, too far from the cameras to fight back. And they are counting on one more thing—that the laid-off worker in the blue city and the dried-out farmer in the red county will never look up from their separate griefs long enough to notice they are being ground down to dust by the same machine.
The cloud has to land somewhere.
It’s landing on us.
All of us.
Time to find each other.
Sources
The Morris family, Newton County, Georgia. Meta data center, construction begun 2018; well failure, sediment, roughly $5,000 spent and $25,000 to replace, Meta’s “unlikely” finding, neighbors’ reported well trouble. The New York Times, 2025, and corroborating coverage.
Data-center water use. Up to ~5 million gallons/day for a large facility; up to ~85 percent evaporative loss; ~17.4 billion gallons consumed by U.S. data centers in 2023; EPA projection of 38–73 billion gallons by 2028. Environmental and Energy Study Institute (EESI); U.S. EPA.
Fayette County, Georgia (the “customer service” case). Quality Technology Services (QTS), a Blackstone company; 615-acre Fayetteville campus; roughly 29 million gallons drawn through two connections over ~15 months; about $147,474 in retroactive charges; no fine levied. Brought to light by attorney James Clifton via public-records request; quote from county water-system director Vanessa Tigert. Tom’s Hardware, The Daily Caller, Cybernews, and POLITICO/E&E coverage, May 2026.
Erin Brockovich data-center campaign. Public reporting map at brockovichdatacenter.com, launched spring 2026; nearly 4,000 community submissions in the first month; “transparency” cited as the most common concern. The Brockovich Report; Newsweek; TechCrunch, 2026.
The Stratos Project / “Wonder Valley” (Kevin O’Leary), Box Elder County, Utah. ~9-gigawatt AI data center plus on-site natural-gas plant; ~40,000-acre campus near the Great Salt Lake; potential buildout cost exceeding $100 billion; “more than double” the electricity Utah uses in a year; approved by the Box Elder County Commission, May 4, 2026, amid protest; referendum effort underway. CNN (Clare Duffy, May 9, 2026); Salt Lake Tribune; Fortune; Common Dreams. Utah State University physicist Robert Davies: real thermal load approaching ~16 GW accounting for gas-plant inefficiency; projected local nighttime temperature rise of 8–12°F; the “~23 atomic bombs of heat per day” framing (reported via ZME Science and corroborating coverage). Resident Brenna Williams (“ration our water”) via Common Dreams. O’Leary’s “proxies for the Chinese government” remark and the Gabi Finlayson / Jackie Morgan “only foreign operative here is a Canadian” reply via Business Insider, HuffPost, and Yahoo.
Waste-heat math (for the “Hiroshima” comparison). A data center converts essentially all input electricity to heat (first law of thermodynamics). Using Davies’s effective ~16 GW figure: 16 GW × 86,400 seconds ≈ 1.38 × 10¹⁵ joules/day. Hiroshima (“Little Boy”) ≈ 15 kilotons TNT ≈ 6.28 × 10¹³ joules. 1.38 × 10¹⁵ ÷ 6.28 × 10¹³ ≈ 22 Hiroshima-equivalents of energy per day, consistent with the “~23 bombs/day” figure in circulation. Comparison is of total energy released slowly as heat, not destructive blast delivered instantaneously. At the nameplate 9 GW the figure would be ~12/day; the higher number reflects the physicist’s full-system estimate and is attributed to him in text.
Electricity / PJM. 2025–2026 capacity auction rose from $28.92 to $269.92 per megawatt-day (~833 percent), the largest single-year increase in PJM history. Utility Dive; DataCenterDynamics. NRDC: data-center growth responsible for $9.3 billion of a $14 billion PJM capacity bill, ~$70/month for a typical household, $100–163 billion cumulative ratepayer costs projected through 2033. Natural Resources Defense Council; Citizens Utility Board, 2025. Pepco residential bills up ~$21/month in Washington, D.C., from mid-2025; PJM-region supply rates up 5–44 percent since June 2025. E&E News/POLITICO.
Coal retirements / Three Mile Island. Roughly 60 percent of scheduled PJM fossil-fuel retirements delayed or canceled; 15+ coal plants pushed back since January 2025. OilPrice.com; EnergyNow. Three Mile Island Unit 1 (retired 2019), rebranded Crane Clean Energy Center; Constellation Energy 20-year PPA with Microsoft; $1 billion DOE loan closed November 2025; targeted restart ~2027. Constellation Energy; Orrick; NucNet.
Rural siting and farmland. Roughly two-thirds of new data centers built in rural areas, much on former farmland; offers up to ~$120,000/acre and higher. Brookings Institution; Farm Progress.
Local outcomes. Tucson, Arizona (Amazon project abandoned); Saline Township, Michigan ($16 billion OpenAI-Oracle project voted down); Mason County, Kentucky (1,500+ residents organizing; farmers rejected a multimillion-dollar offer). Brookings Institution; Civil Eats; The New York Times.
Public opinion. Gallup, “Americans Oppose AI Data Centers in Their Area,” released May 13, 2026 (telephone survey by Recon MR, March 2–18, 2026, n=1,000): 71 percent oppose a local AI data center (48 percent strongly); opposition exceeds the highest level Gallup has recorded for nuclear power plants since it began asking in 2001. The New York Times, “Liberals and Conservatives Are Finding Common Ground—Against Data Centers,” May 1, 2026.
THANK YOU for this immeasurably important article. More people need to be aware of the real cost of the AI bulldozer.
And there are people that actually believe that embracing all of this technology is a good thing. They haven’t felt the negative effects of it yet! Everything in moderation is still a good rule of thumb!