The Ratepayer Subsidy Powering AI

Two billion dollars. That's the number Maryland's regulators just attached to a complaint they filed with the Federal Energy Regulatory Commission, describing what residential ratepayers in the state are about to be charged so AI data centers can run in Virginia.

Tom's Hardware has the reporting. Maryland's argument is that the cost allocation breaks an explicit ratepayer protection pledge, and that the upgrades being charged to Maryland customers are driven almost entirely by load growth from out-of-state hyperscaler facilities. The state wants FERC to intervene.

I read this as the most important AI infrastructure story of the year, and almost nobody in the AI conversation is paying attention to it.

What the filing is actually arguing

Strip the political framing and the math is simple. PJM, the regional grid operator, identified roughly two billion dollars of transmission upgrades needed to serve load growth concentrated in northern Virginia's data center corridor. Under current cost-allocation rules, a meaningful share of that bill flows back to all ratepayers across the PJM footprint, including Maryland households who will not run a single AI workload off the new capacity.

Maryland's complaint isn't really about Maryland. It's about whether the cost-causation principle that's supposed to govern transmission rate design actually does. If a discrete set of large loads causes the upgrade, the rules say those loads should pay. In practice, the allocation has been spread.

The framing the headline pushes, and where it's wrong

The Tom's Hardware framing ("Aren't AI hyperscalers supposed to pay for these upgrades?") is rhetorical, and it lands. But it also flattens the actual fight. Hyperscalers do pay for some grid costs. They sign interconnection agreements. They fund local upgrades. What they don't fully pay for is the broader transmission expansion their concentrated load patterns trigger across a multi-state grid.

That's not theft. It's a rate-design choice that was made when the dominant load growth was diffuse residential and commercial demand, and it never got revisited when a handful of buyers started building gigawatt-scale campuses. The accountability gap is regulatory, not moral. Which is exactly why FERC matters more here than any state-level outrage.

The ratepayer subsidy is about to be the default

But here's the part most people miss. If FERC doesn't issue clear cost-causation guidance that pushes more of the upgrade burden onto the loads driving it, the Maryland pattern becomes the template. Every state hosting downstream ratepayers for an out-of-state data center cluster ends up filing some version of this complaint. The patchwork response is slow, uneven, and politically explosive.

For anyone scoping enterprise AI commitments, the implication is concrete. Your model vendor's compute price reflects today's cost allocation. It does not reflect the version where FERC reallocates and hyperscaler infrastructure costs go up by some non-trivial percentage. If you're signing multi-year capacity deals at current rates, you're betting that the regulatory status quo holds. That's a bet with bad odds.

What to watch

The falsifiable version: by the end of 2027, expect at least three more states to file FERC complaints with the same structural argument as Maryland's. If FERC issues a cost-causation order that reallocates meaningful transmission costs to large loads, hyperscaler compute pricing has to rise. If FERC punts, the ratepayer subsidy becomes the permanent funding model for the AI build-out, and it gets litigated state by state for the next decade.

The AI infrastructure story everyone's tracking is chips and capex. The one that actually decides whether the build-out is sustainable is who pays for the wires.