In the space of twelve months, one electric utility in Houston watched its data-center requests grow from one gigawatt to twenty-five — roughly the entire electricity demand of Denmark, arriving in a single city's inbox in a single year.

But most of that demand does not exist. It is the same handful of projects, asking five different utilities at once — because asking is almost free, and whoever says yes fastest wins. The grid gets planned, built, and priced on those requests. So you are already paying, in your monthly bill, for wires to buildings that will never be built. And this month, one American state reached for the only lever it had: it stopped approving new data centers altogether.

On July 14th, the governor of New York signed the first order of its kind in the United States: for up to a year, no environmental permits for new giant data centers — any facility drawing 50 megawatts or more, about the electricity of a town of 40,000. Not a tax, not a permanent ban; a pause, bought so the state can write the rulebook it doesn't yet have. The reason is money, not machines: New Yorkers' electricity has climbed 68% since 2019.

But the order says something most coverage skipped, and it is really the whole story. Data center demand, it says, is unpredictable, and the electricity those centers ask for does not always show up — and when it doesn't, the utility has already built the wires, and the people who pay that utility are left holding the bill. The official reason for stopping isn't that the data centers are coming. It's that they might not be — and we're building for them anyway. Fourteen states have bills to restrict data centers; 300+ were filed across 30-plus states in six weeks. The industry's trade group calls the freeze a mistake: the jobs will simply go somewhere else.

The easy version: AI is eating the power grid, and a state finally pushed back. That version is popular, and mostly wrong. The demand is real — the four biggest cloud companies have committed something like $650 billion to this build-out, and they have the money to do it. The question is stranger: how much is real? Nobody knows — not the utilities, not the regulators, not the state that just hit pause. And the difference between how much was asked for and how much actually gets built is the part that lands on your electricity bill.

Filing a request to connect to the grid is cheap — a form and a modest fee. What it buys is enormous: land with power attached is worth far more than land. So the rational move isn't to ask one utility; it's to ask five, in three states, for the same building, then build wherever says yes first and cheapest. The other four requests sit in line, counted on the books as real electricity someone will need — often nobody ever takes them out. Nobody is cheating; every request is legal. The problem is we made asking free and building expensive, then planned as if they were the same thing. A former Google engineer who now runs a grid software company estimates, conservatively, five to ten times more requests to connect than data centers actually built.

A utility can't legally guess: if a customer requests power, it must plan to serve it. And when the utility spends that money, it doesn't eat the cost — it's allowed to add it to everyone's bill. So the developer asks for free, the utility builds, and you pay, whether the building arrives or not. The forecast — simply the sum of the requests, same project counted four times — decides substations, power lines, electricity bought years ahead. In the market supplying 13 eastern states, the cost of guaranteeing future power rose roughly eightfold in three years, driven substantially by these forecasts; Virginia's electricity got about a fifth more expensive in a year. And the quietest loop: your utility is a regulated monopoly guaranteed a profit on what it builds — not on the electricity it sells, on the concrete and the copper. The one organization expert enough to challenge the forecast is the one that profits if nobody asks. Not a conspiracy — the business model we designed.

Make asking cost something: Ohio's new deal makes large data centers pay for at least 85% of the power they sign up for — used or not — for twelve years, with proof they can afford it and a fee if they walk. Virginia adds longer contracts and a deposit of tens of millions; Georgia makes big customers pay for the power plants and long lines behind their own hookup. Twenty-four states now have a version of this. The result that tells you everything: when Ohio asked for money up front, requests to connect fell by half. Half the "demand" was never demand — it was people reserving a place they hadn't decided to take, because reserving cost nothing. The trade-off: ~$10M in year one for a large project, which giants absorb and small operators can't.

Ireland got here first: data centers grew to 22% of the country's entire electricity — more than every home combined — so in 2022 Dublin stopped connecting them. In December, after four years, Ireland ended the pause — not because the problem was solved, but because a pause is not a policy. The condition that replaced it: build enough of your own power or batteries to cover everything you use, and send some back when the grid is short. The data center stops being something the grid carries and becomes something it can lean on. The cost: expensive, slower, often gas burned on site — and for four years the buildings simply went to other countries. The third answer questions the premise: grids are built for the rare peak, but training a model can wait an hour or run elsewhere. Connect fast, power down when the system is strained. The limit: when you ask an AI a question, that answer can't wait.

Come back to that utility in Houston, watching one gigawatt become twenty-five in a year. Nothing about that number was a lie — every request was real, filed by someone who genuinely might build. The number was just never a measurement of the future. It was a measurement of how cheap it is to ask. That is the whole flaw, and it's bigger than data centers: any time a system lets you claim something valuable for almost nothing, it will get flooded with claims — and if it plans as though every claim were a commitment, it will build for a world that doesn't arrive and send the bill to people who never asked for any of it. New York's pause isn't the answer, and the state has been honest that it isn't: it's a year, bought so they can find one. Ohio bought its answer with a deposit. Ireland bought four years, spent them, and came back with a condition. So the question underneath your power bill isn't whether the machines are coming. It's simpler, and older: who pays for the wire — the person who asked for it, or the person who happened to be standing there?