Who Pays for the Server Farms? The Utility Dockets Quietly Deciding Your Power Bill

The numbers arrive quietly, in a filing most people will never open. In PJM Interconnection's capacity auction for the 2026/27 delivery year — the annual purchase of guaranteed future power across a grid stretching from Illinois to New Jersey — the clearing price hit $329.17 per megawatt-day, with several zones pinned at the cap. Two years earlier, the same auction cleared at $28.92. That is roughly a tenfold increase in the price of reliability, and under the rate structures most states still use, the bill lands on household meters.

What changed is demand. After fifteen years in which American electricity consumption was essentially flat, grid operators are now forecasting load growth not seen since the 1970s. The proximate cause is the hyperscale data center: warehouse-scale computing campuses drawing hundreds of megawatts each, built to train and run artificial intelligence models, clustered in Virginia's Data Center Alley, suburban Atlanta, central Ohio, the Texas prairie outside Abilene, and increasingly in rural counties where land is cheap and zoning boards are tired.

The socialization of risk, the privatization of return

The economics here are not mysterious. A data center developer signs a long-term contract, secures a decade of local tax abatements, and books the profit. The transmission lines, substations, and gas-fired peaking plants required to serve that load are planned, permitted, and paid for through a process in which the utility is guaranteed a rate of return on capital investment — a return collected from every customer in the service territory, whether or not they ever touch a chatbot.

This is the central equity question of the energy transition in 2026, and it is being settled in dockets that draw a handful of public commenters. Regulators in several states have begun to notice. Large-load tariffs approved in recent years — AEP Ohio's data center tariff, Georgia Power's rules for very large customers, Dominion Energy's contract terms in Virginia — attempt to make new mega-customs pay a share of the infrastructure they trigger. The details matter enormously, and they vary wildly: minimum demand thresholds, exit fees, collateral requirements, the difference between a fifteen-year commitment and a five-year one. A tariff that looks tough in a press release can leave ratepayers holding the stranded asset if the developer walks.

The strongest objection, and why it falls short

There is a serious case on the other side, and it deserves a fair hearing. Data centers generate real property tax revenue — in some rural counties, the single largest line in the budget. They are a hedge against the utility death spiral, adding customers who spread fixed costs across more meters. And the argument that AI capacity is a matter of national competitiveness is not frivolous; the compute buildout is a strategic asset, and if the United States makes it impossible to build here, the buildout goes elsewhere.

But each of those claims needs testing against the record. The property tax windfall is frequently negotiated away before it arrives: abatement packages in county after county have reduced assessments to a fraction of the headline capital figure, with the local school district — the largest consumer of property tax revenue — often the last to know what it gave up. The death-spiral argument assumes new load pays its way, which is precisely what is in dispute. And 'national competitiveness' is a claim on the public purse that comes with no public governance: the developer gets the strategic subsidy, the ratepayer gets the bill, and the profits accrue to shareholders who bear none of the risk.

Who actually absorbs the cost

Household energy burden is not distributed evenly. Low-income households spend a substantially larger share of their income on electricity than affluent ones — in some regions three times as much — and they have the least capacity to absorb a double-digit rate increase or to invest in efficiency upgrades. Fixed charges, which utilities have pushed upward in state after state, fall hardest on the smallest users: the senior on a fixed income in a one-bedroom apartment pays the same connection fee as the owner of a 4,000-square-foot house.

Then there is the geography. Transmission corridors and gas peakers get sited in places with the least political resistance: rural counties, often poor ones, often with Black and Latino populations historically overburdened by industrial siting. The data center sits on the county line taking water from the aquifer; the compressor station and the new high-voltage line go where the lawyers are cheaper. Meanwhile, the water footprint of evaporative cooling at scale is a live conflict in the Southwest and parts of the Southeast, and the emissions from new gas generation complicate every state climate plan that assumed coal retirements would proceed on schedule.

A docket is a political document

Public utility commissions are the least-covered organs of state government, and in most states their members are either appointed for staggered terms or elected in contests with turnout in the single digits. This is where the energy transition is actually being written — not in the climate statutes that get the speeches, but in rate design, integrated resource plans, and cost-allocation formulas that determine who pays for what.

The 2026 midterms will be fought over the cost of living, and electricity is now a cost-of-living issue. Candidates who want to speak credibly about affordability will need to show up at the commission, file comments, and demand that the companies profiting from the buildout carry the risk rather than exporting it to the kitchen table.

The server farms are coming either way; the only open question is whether the public gets a fair price for hosting them.