The Data-Center Power Panic Has a Control Group - and It's in Ohio

Two numbers, filed weeks apart, describe the same industry.

The first: PJM Interconnection's independent market monitor reported in late July that data centers accounted for $6.3 billion - 38% - of the $16.4 billion in charges from the grid operator's latest capacity auction, and $29.4 billion, or 46%, across the last four. Capacity prices in the 13-state PJM footprint have gone from $28.92 per megawatt-day for the 2024/25 delivery year to $329.17 for 2026/27. Roughly tenfold.

The second: AEP Ohio, which has signed up 17,861 megawatts of contracted data-center load - more than most countries operate in total - has filed to *decrease* average residential distribution rates by about $1.22 a month.

Both things are true. Reconciling them is the story, and almost nobody is doing it.

The tempting version

The intuitive account is nearly irresistible: AI needs electricity, electricity is finite, therefore your bill goes up. Goldman Sachs gave it a headline number - U.S. electricity prices rose 6.9% in 2025 against 2.9% headline inflation, with data centers driving about 40% of demand growth. The politics followed fast. Mikie Sherrill and Abigail Spanberger won governorships in November 2025 partly on utility-bill pledges. By March, tech executives were at the White House signing a nonbinding "ratepayer protection pledge" promising to "build, bring, or buy" their own power.

Then EPRI published research finding that data-center growth was associated with retail electricity prices *falling* roughly 3.5% for every doubling of capacity between 2015 and 2024. Coverage largely filed this under "dueling studies" and moved on.

It isn't a duel. Read EPRI's mechanism rather than its conclusion: large new load lowers average rates when it spreads fixed network costs across more kilowatt-hours - provided generation and transmission expand alongside it. Prices rise only when supply fails to keep pace.

PJM has already conceded the supply half

This is where the grid operator's own paperwork becomes the most useful document in the file. PJM faces roughly 70 gigawatts of forecast new large-load demand by 2038 while about 15 GW of generation has retired since 2022. It has acknowledged missing its 20% reserve margin target, and its near-term backstop procurement targets a shortfall of roughly 6.8 GW, with tens of gigawatts more implied over the longer horizon.

A capacity auction is not a bill for electricity consumed. It is an auction for the *promise* of availability - and its price is exquisitely sensitive to scarcity at the margin. When a market clears near shortfall, small increments of forecast demand produce enormous price moves. The market monitor's attribution figures are best understood as answering "whose forecast load pushed the auction into scarcity pricing?" rather than "who used the power?" Those are different questions with very different policy answers.

And the forecast itself is soft. Utilities, regulators and at least one FERC complaint have described speculative and duplicated interconnection requests - the same project shopping itself across multiple territories, or hedging on sites it will never build - inflating load projections well beyond likely buildout. Part of what ratepayers are being charged for, in other words, is a queue.

The control group

Which brings us back to Ohio, where regulators did the unglamorous thing. AEP Ohio's data-center tariff requires large loads to commit to paying for at least 85% of their requested capacity for 12 years, whether they use it or not. Take-or-pay kills phantom demand at the source: a speculative request becomes expensive rather than free, and the developer who does show up funds the infrastructure built for it.

The result is a jurisdiction with enormous contracted AI load and a filing to lower residential distribution rates. Two honest caveats: distribution charges are one line on a bill, not the whole thing, and Ohio customers are still exposed to PJM-wide capacity costs like everyone else. A $1.22 monthly reduction does not offset a tenfold capacity price. But it demolishes the assumption that data-center growth mechanically raises what households pay. The variable that changed in Ohio was not demand. It was who signed for the risk.

PJM is now copying the homework

On July 27, PJM filed to do at regional scale roughly what Ohio did at utility scale: define "Large Loads" at 50 MW and above, run a Reliability Backstop Procurement auction with a $555/MW-day capacity ceiling, and expose large loads that don't self-supply capacity to curtailment. Most consumer coverage stopped at "PJM readies a backstop filing." The mechanics matter more than the headline: this is a cost- and obligation-shift already in motion, filed while the political argument about whether one should happen was still underway.

Virginia, meanwhile, imposed a data-center electricity consumption tax of $0.011/kWh from July 1, capped at $600 million a year with a refund mechanism if collections overshoot. Dollar-denominated, not a pledge.

The ledger problem

None of this makes the backlash irrational - but it does reveal that the public and the utility bill are keeping separate books. Gallup finds 71% of Americans oppose a data center in their community, 48% strongly, making it less welcome than a nuclear plant. Nearly three-quarters of Virginia voters blame data centers for rising costs. Yet in Loudoun County, the densest data-center geography on earth, a 2026 NVTC analysis found data centers supply roughly half of county property tax revenue and return about $26 per $1 of county services they consume; without them, residential property taxes would rise by something like $5,800 a year.

So a Loudoun homeowner may be paying more for electricity and dramatically less in property tax, while a New Jersey homeowner with no data center nearby pays the PJM capacity charge and collects none of the tax base. That asymmetry - not the raw kilowatt-hours - is the genuine grievance, and it is a jurisdictional design flaw, not a law of physics.

The honest conclusion is duller than the viral one and more actionable. Data centers did not raise your bill by drinking electricity. A grid that retired 15 GW while forecasting 70 GW raised your bill, and a market that prices scarcity brutally sent the invoice to whoever hadn't signed a contract. Ohio made them sign. Everyone else is now, expensively, catching up.