Maryland’s ratepayer advocate filed a complaint this week against a bill its residents never agreed to incur: roughly two billion dollars to upgrade the electrical grid for AI data centers, most of them located in other states. The regional transmission operator, PJM, has proposed twenty-two billion dollars in upgrades to serve the buildout, and its cost-allocation formula spreads that bill across sixty-five million people in thirteen states — including Maryland households who will pay an estimated three hundred forty-five dollars each for capacity that powers server farms in Virginia and Ohio. In the same days, sixty-nine jurisdictions across the country moved to block new data center construction, four of them permanently; one facility was found to have quietly drawn twenty-nine million gallons of water; communities reported a low-frequency hum beneath their homes. The cost of the buildout has begun to arrive, in dollars and water and sound, on the doorsteps of people who do not use it and were never asked.
The Bill Arrives Elsewhere
The mechanism is the oldest one in the book, and its age is the point. The data centers are private. The profit they generate is private. The electricity they consume, however, requires a grid, and a grid is a shared public asset whose upgrades are paid for collectively, through everyone’s bills, according to formulas written long before any single customer demanded the load of a small city. PJM’s formula socializes the cost of the transmission lines across the entire region it serves, which means a household in Maryland subsidizes the infrastructure feeding a data center it will never enter, owned by a company whose product it may never use, generating returns it will never share in. The gain is concentrated. The cost is distributed. This is not a flaw in the arrangement. It is the arrangement.
The detail that gives the case its edge is the broken pledge. There exists a “ratepayer protection pledge” — a public commitment that the large technology firms driving this demand would shoulder the infrastructure costs rather than shifting them onto households and small businesses. Maryland’s complaint is, in essence, that the pledge has already failed, that the costs are flowing exactly where the pledge promised they would not, and that the formula was never altered to honor the promise. The pledge was the kind of commitment that costs nothing to make and binds no one, offered to soften the politics of the buildout, and it is dissolving on contact with the first real bill, precisely as such commitments do.
Three hundred forty-five dollars is a small number to the companies and a real one to the household, and the asymmetry is the engine of the entire model. No individual ratepayer’s burden is large enough to justify the cost of fighting it; the data center’s benefit is concentrated enough to justify any amount of lobbying to preserve it. This is the structure that allows a diffuse public to be charged, reliably, for a concentrated private good: the victims are too many and too lightly taxed to organize, and the beneficiaries are too few and too heavily rewarded not to. Maryland is fighting only because a public advocate exists whose entire function is to internalize the diffuse cost and contest it on the ratepayers’ behalf. Most jurisdictions have no such office, and most bills are simply paid.
Water, Sound, and the Veto
Electricity is only the most legible of the costs. A data center is a physical building, and physical buildings have physical appetites that the word “cloud” was specifically designed to make you forget. One facility drew twenty-nine million gallons of water — for cooling, drawn quietly, noticed only when residents reported the pressure in their own taps dropping — and the officials declined to impose a fine. Others emit infrasound, a low-frequency vibration below the threshold of conscious hearing that does not register on a standard meter but is felt in the body, and the communities living beside them report the kind of diffuse, dismissible symptoms that institutions are practiced at not measuring. The cloud is a metaphor. The water table is not. The hum under the bedroom is not.
So sixty-nine jurisdictions reached for the only instrument a community possesses when the cost is local and the benefit is elsewhere: the veto. They blocked the construction, four of them permanently, because zoning is the one form of power that has not yet been socialized away from the people who hold it. This is the same physical wall that drove Meta to reserve a gigawatt of sunlight from orbit — the buildout’s collision with the stubborn limits of the planet — observed now from the other side, from the towns rather than the firms. The companies experience the limit as a supply problem to be engineered around. The communities experience it as a thing arriving in their water and their walls, and they are voting, where they still can, to keep it out.
The veto is real and it is also losing, slowly, for a reason the export-control story already taught. Capital routes around a blocked jurisdiction to a willing one, the way restricted compute routed around a chokepoint to domestic silicon. A town that refuses a data center does not prevent the data center; it relocates it to the next town with a weaker advocate, a more desperate tax base, a less attentive council. The aggregate buildout is indifferent to which specific community hosts it, and therefore indifferent to any specific community’s refusal. The veto protects the town that exercises it and exports the cost to the town that cannot, and the buildout proceeds at the sum, undeterred, finding the path of least resistance through the map of who is too weak to say no.
What This Means
The economics of the buildout depend, structurally, on externalizing its physical costs — on the grid being shared, the water being cheap, the noise being unmeasured, the bill being spread thin enough across enough people that no one of them is moved to fight. For two years this worked invisibly, because the costs had not yet grown large enough to surface as line items and symptoms. They have now. The three-hundred-forty-five-dollar charge, the twenty-nine million gallons, the hum: these are the moment the abstraction broke, the moment the cloud acquired a location and a weight and a population that can feel it.
What surfaces with the costs is a question the buildout has been structured to never quite ask aloud: who is this for, and who pays for it. The intelligence being constructed in these buildings is sold as a universal good, a tide that lifts everyone, and the household paying three hundred forty-five dollars to power it is told it is a beneficiary. But the benefit is abstract and deferred, and the cost is concrete and monthly, and the gap between them is exactly the gap between the people who own the data centers and the people who live beside them. A universal good whose costs fall on the particular and whose gains accrue to the few is not universal. It is a transfer, wearing the language of a tide.
The resistance will not stop it; it will only redistribute it, toward the jurisdictions with the least capacity to refuse, which is the direction every externalized cost has ever flowed. But the resistance changes one thing, and the change is worth marking. It ends the period in which the buildout could be discussed purely as software — as models and benchmarks and capabilities, weightless and placeless. The data center has a return address now. It is in someone’s county, drinking someone’s water, charged to someone’s bill, humming under someone’s floor, and the someone has begun to notice. The cost was always physical. The only question this technology ever posed about it was who would be made to carry it, and the answer is arriving, jurisdiction by jurisdiction, on the people with the least power to set it down.