Who really pays for the AI data-center boom

The electric bill is becoming the place where the artificial intelligence boom first shows up for ordinary households, and the accounting behind it is upside down. The companies building the data centers that train and run large AI models are not, in most of the country, paying the full cost of the generation, transmission and grid upgrades their facilities require. Residential customers are. That asymmetry is not a law of physics; it is a policy choice made, state by state, inside utility-rate cases that almost nobody outside the industry watches, and it deserves far more scrutiny than it is getting.

The bill is already showing up

The scale is no longer speculative. The U.S. Energy Information Administration said in its latest outlook that national electricity consumption will grow 3.1 percent in 2027, with growth "primarily driven by the commercial sector," which the agency expects to overtake residential consumption for the first time on record. Average residential electricity prices are forecast to rise roughly 5 percent in 2026 alone, and Utility Dive's reading of that same EIA data ties the acceleration explicitly to demand "particularly from the commercial sector, which includes data centers," with the steepest increases concentrated on the East Coast.

Economists at the Federal Reserve Bank of Dallas have tried to isolate exactly how much of that increase belongs to data centers specifically. In a research note published in March, Owen Kay, Lutz Kilian and Reid Taylor modeled scenarios in which data-center demand alone adds between 0.05 and 0.13 percentage points to headline personal-consumption inflation by 2030 under plausible utilization assumptions, and as much as 1.02 points in an extreme, full-capacity case the authors themselves call implausible. Even the modest end of that range is a nationwide inflation effect caused by one category of commercial customer, which is an unusual thing for a central-bank research shop to need to quantify at all.

Wholesale capacity markets show the mechanism at work in sharper relief. In the PJM Interconnection, which covers roughly 65 million people from Illinois to New Jersey, the clearing price for the 2025-26 delivery year jumped to $269.92 per megawatt-day from $28.92 the year before, a roughly ninefold increase that reporting compiled from PJM's own market monitor attributes nearly half of, across the four most recent auctions, to data-center load growth. That is not a rounding error passed along quietly; it is a multibillion-dollar bill that gets divided among every utility and, eventually, every customer in the region.

The dispute over who actually pays

Georgia makes the underlying argument concrete. Georgia Power has frozen base rates through 2028 while regulators approved roughly 9,900 megawatts of new generation, mostly gas-fired, to serve a surge of large-load customers that is overwhelmingly data centers. The company's defenders, including its own executives, argue that large-load revenue is what is keeping that freeze possible and will eventually save a typical residential customer more than $100 a year. Critics on the Georgia Public Service Commission itself disagree sharply about where the money is actually going. The commission voted unanimously to investigate whether a pricing structure used almost exclusively by its largest industrial customers lets them avoid paying their share of fuel costs that other ratepayers still cover.

Commissioner Peter Hubbard, citing an independent analysis, said regulators had "identified nearly $1 billion in costs that data centers pay nothing towards," and staff separately estimated that unreallocated costs could push fuel bills up as much as 11 percent a month for other customers by 2028.

Georgia Power's answer, that large customers "keep rates lower" in aggregate, may well be true on net and still miss the point. The argument for data centers as a public good rests on averages; the argument against the current system rests on who inside that average absorbs the risk. A fixed-price, decades-long industrial contract and a monthly residential bill are not interchangeable just because a regulator's spreadsheet nets them against each other.

A handful of states are getting the incentives right

The honest test of whether a state takes this seriously is whether it has rewritten its tariffs so that large new loads pay for the infrastructure they require, rather than relying on the diffuse protection of a statewide average. A few have started. Ohio's AEP territory now requires data centers above 25 megawatts to commit to paying for at least 85 percent of their subscribed capacity for up to twelve years, which prevents a facility from reserving enormous grid capacity and then leaving ratepayers to cover the fixed cost if it never materializes or relocates. Virginia's Dominion Energy and Oregon's Portland General Electric have each created separate rate classes for the largest loads, and Oregon's version raised data-center rates by an average of 29 percent while residential rates actually fell.

New York took the blunter route. Governor Kathy Hochul's Executive Order No. 62, signed on July 14, paused state permitting for any data center drawing 50 megawatts or more, citing nearly 12 gigawatts of new load requests already sitting in the state's interconnection queue, and directed regulators to consider designing a "Grid Acceleration Fund" so that data centers, not the broader ratepayer base, cover the cost of the grid upgrades their arrival requires. A yearlong freeze is a blunt instrument, and it will slow genuinely useful investment along with the costly kind. But it buys the one thing most other states skipped: time to design a tariff before, rather than after, tens of gigawatts of new demand are already under contract.

That is the standard every state regulator ought to be held to now, while the buildout is still accelerating rather than after it has locked in another decade of cost allocation nobody chose on purpose. Cost-causation pricing, not a blanket freeze and not a blank check, is the actual fix: large loads should pay for the specific capacity, transmission and fuel-hedging risk they create, with that obligation enforceable through the kind of minimum-take contracts Ohio and Virginia have adopted. States still debating whether data centers are an unambiguous economic blessing are asking the wrong question. The right one, which Georgia's commissioners and New York's governor have at least started asking out loud, is who is holding the bill when the next capacity auction clears nine times higher than the one before it. Right now, in most of the country, the answer is still: whoever already pays the electric bill every month, whether they asked for any of this or not.