AI Data Centers vs. Residential Electricity: Who Gets the Next Megawatt?

Photo: Novoklimov, Wikimedia Commons, CC0

01 The fight now underway in every state capital

The dispute is no longer abstract. In Virginia — home to more data centers than any other jurisdiction on earth — the State Corporation Commission approved a Dominion Energy increase that adds about 16 dollars to the typical residential bill each month, pushing it to roughly 165 dollars, while creating a new GS-5 rate class for data centers starting January 1, 2027. Under GS-5, data center customers must sign 14-year contracts and pay for at least 85 percent of the transmission and distribution costs and 60 percent of the generation costs of what they request — whether or not they ever draw that power.

The consumer groups who won those provisions point out the leak in the win: the Piedmont Environmental Council's analysis found the decision still leaves 61 percent of the costs of those grid upgrades on individual ratepayers once the 14-year contracts end. “The SCC is asking their constituents to continue to subsidize the energy needs of the richest companies in the world,” the group's president said when the order landed. Both halves of that sentence are now the framing for every rate case in the country.

Analysis — not prediction. N43 and Hermes AI grounds every scenario in the documented record and verified reporting as of September 19, 2026; where evidence is incomplete we say so.

02 The new state playbook

What happened in Virginia is one entry in a rapidly filled ledger. Oregon's POWER Act produced a nearly 30 percent rate increase for data centers — approved by the Public Utility Commission with a simultaneous 1.3 percent decrease for residential customers. Florida's Senate Bill 484, which took effect in August, requires utilities to “reasonably ensure that each large load customer bears its own full cost of service” — and Duke Energy's first proposal under the law is now the test case before Florida's PSC, with consumer advocates arguing it “doesn't attempt to comply with the most basic provisions.”

Ohio went first and furthest: the Public Utilities Commission approved AEP Ohio's landmark tariff requiring large new data centers — those above 25 megawatts — to pay for at least 85 percent of their subscribed capacity for up to 12 years, with exit fees, collateral and a four-year ramp. Pennsylvania landed a settlement with PPL Electric that raises residential bills 4.9 percent to about 184 dollars a month while forcing large loads into 10-year minimum agreements with security equal to the cost of their grid upgrades. Texas, Georgia, Oklahoma and New Jersey have all passed or are weighing variants.

03 How rate design shifts who pays

The mechanics matter more than the headlines, because the same words — “data centers pay their share” — produce wildly different outcomes depending on five design choices: contract length (10, 14 or 20 years), minimum-take floors (paying for 85 percent of subscribed demand regardless of use), ramp schedules (a customer signing for 100 MW but using 20 while building pays for what it signed), collateral and exit fees (protection if the AI bubble pops), and — most technical and most decisive — the cost-allocation methodology itself.

That last one is where consumer advocates say the Virginia decision fell short: the current methodology assigns costs by when peak demand occurs, and data centers' flat 24/7 profile avoids the spikes that trigger cost assignment. The SCC has scheduled a future hearing on the methodology — the fight over the formula, not the formula's outputs, is where the next megawatt's cost is really decided.

The wholesale layer adds a harder problem. PJM's independent market monitor attributed roughly 40 percent of the record 16.4-billion-dollar 2027/28 capacity auction to phantom data center load — projects with transmission service agreements that have not begun operating — and 63 percent of the prior year's price increase to data center demand growth. Capacity costs flow back to every bill in the region.

04 The counterargument: growth can pay its own way

The strongest argument against the subsidy framing comes from the utilities winning the load. American Electric Power has signed agreements supporting 69 gigawatts of incremental demand through 2030 — mostly hyperscalers and data centers — and told investors those contracts could produce up to 16 billion dollars in cost offsets for residential customers, because take-or-pay contracts spread the same fixed system costs over far more kilowatt-hours.

One AEP subsidiary is trying to turn theory into bills: Indiana Michigan Power has proposed 59 million dollars of rate reductions in 2027 plus a three-year residential rate freeze, explicitly financed by revenue from new large loads — roughly 100 dollars a year back to a typical household, if regulators approve it in mid-2027. The condition attached is the same one Ohio imposed: the growth only protects ratepayers if the contracts are long, minimum-payment and collateralized. If a data center can leave before the assets built for it are paid off, the neighbors inherit the bill.

05 Commissions as the referee of the AI era

Strip away the technology and this is a classic regulatory question with an unfamiliar scale: a new customer class arrives, requests infrastructure sized in city-scale multiples, and the regulator must decide who guarantees the financing. What makes 2026 different is that the federal layer has joined in. On June 18, 2026, FERC issued show-cause orders to all six regional grid operators under its jurisdiction — PJM, MISO, SPP, CAISO, NYISO and ISO-NE — directing them to justify or reform the rules governing how data centers and other large loads connect, with reforms due by August 17, 2026, and five categories of reform teed up: faster interconnection study processes, preventing cost shifting, co-location and behind-the-meter rules, flexible-load service offerings, and study processes for co-located generation.

That converts every state commission into an implementer of a national policy direction, at least where the grid is FERC-jurisdictional. The state playbook — minimum takes, long contracts, collateral — is converging on the same answer from both parties: growth is welcome; socialization of its cost is not.

06 What to watch next

The Duke decision in Florida is the first full test of a statute that flatly bans cost-shifting; a written order is expected after post-hearing briefs. The Indiana rate-cut case decides by roughly June 2027 whether large-load revenue actually flows back to households or stays a talking point. The Virginia cost-allocation hearing will determine whether the 61 percent leak gets closed. PJM's next capacity auction tests whether phantom load keeps inflating prices as projects reach commercial operation or stall. And the Pennsylvania Ratemaking Working Group — opened in September to scrutinize how utility profit rides on rate increases — may produce the first state-level rewrite of how growth is monetized at all.

The deeper question is simpler than any tariff: electricity is becoming the scarce input of the AI economy, and the price of that input is set, mostly, by public commissions in windowless hearing rooms. Who gets the next megawatt is a question about chips. What households pay for it is a question about these dockets.