TL;DR
In March, staff at Ohio's utility regulator agreed with something the state's own manufacturers had been saying: the forecast used to justify building more grid had counted the same expected data center demand twice. That matters more than it sounds, because grids are sized on requests to connect, not on customers who show up, and everyone on the bill pays for the difference. Ninety years ago the same argument was settled in the opposite direction, and it is worth remembering who was being asked to pay then.
The complaint came from the factories, not the activists
Ohio has roughly a hundred data centers running and about a hundred and fifty more announced. AEP Ohio has 17,8 gigawatts of demand under signed contract stretching to 2035, and told regulators the state's power demand would nearly triple in a decade.
The Ohio Manufacturers' Association read the filing and said the forecast counted some of that demand twice. In March, the staff of the Public Utilities Commission of Ohio agreed with them.
Sit with who is making that argument. Not an environmental group. The people who run factories, who need the same grid, and who will be standing in the same queue behind whoever actually builds.
Requests to connect are not customers
A grid is planned against interconnection requests. A developer can file in several territories at once for the same project, because filing is cheap and optionality is worth more than accuracy. The utility adds them up. The regulator approves capital against the total. Ratepayers fund the capital over thirty years.
Jonathan Koomey has spent two years arguing that the data center numbers are inflated. He is the most quoted skeptic of the energy alarm, and in March he and GridLab published a guide for regulators saying so plainly: data centers were 1,5 percent of global electricity in 2024 and 4,4 percent of US electricity in 2023.
People cite him to calm everyone down. They are reading him backwards. If the forecasts really are padded with duplicate and speculative requests, then the grid is being oversized against demand that will never arrive, and the bill for the empty capacity still lands. His argument does not weaken the case that ratepayers are being charged for someone else's optionality. It is the strongest version of that case.
The most expensive data center is the one that was announced and never built.
What the market is already saying
PJM's capacity price went from 28,92 dollars per megawatt-day for 2024/25 to 269,92 for 2025/26 to 329,17 for 2026/27. In July the auction for 2028/29 hit the price cap for the third time in a row at 325 dollars, a total of 16,4 miliardi di dollari, with a reserve shortfall of 6,8 gigawatt below target. Without the cap it would have cleared near 555 dollars, and around 777 in northern Illinois.
A price cap that binds three times running is not protection. It is a deferral with a queue attached.
The politics moved faster than usual. On 16 September the US House passed the Ratepayer Protection Act 417 votes to 3, asking states to consider making large loads pay the full incremental cost of the generation and wires they need, with financial guarantees up front. Thirteen states currently have no large-load tariff at all. In an Economist/YouGov poll taken at the end of August, 63 percent of Americans said they would oppose a data center in their community, including 52 percent of Republicans.
When the advocacy campaign funded by the industry's own investors starts arguing that the build-out "must benefit communities first", the argument about whether there is a cost shift is finished. Only the invoice is still being drafted.
Why this matters for your business
In 1935 the problem ran the other way. A distribution line cost 500 to 800 dollars a mile, and there were one to three customers at the end of it, so private utilities did not build it. Rural residents who wanted power were told to front the capital themselves. Roosevelt signed the executive order creating the Rural Electrification Administration on 11 May 1935, and the loans that followed moved that cost onto the public in the name of universal access.
Same infrastructure, same public lever, direction reversed. In 1935 the small user at the edge was asked to pay in advance so they could be connected. In 2026 the large user at the center gets the public to pay in advance, and the public is now asking for the money back.
If you are buying compute, this is about to show up in your contracts as large-load tariffs, capacity guarantees and exit penalties. Read the interconnection terms of whoever hosts you before the state you are in writes them for you.