PJM Hit Its Price Cap a Third Time. The Cap Is Doing the Work.

The most important number in PJM's latest capacity auction is not the one the grid operator published. It is the one the price cap suppressed.

On July 15, PJM Interconnection — the largest grid operator in the United States, serving more than 67 million people across 13 states and the District of Columbia — released results for its 2028/2029 Base Residual Auction. The auction procured 138,318 MW of unforced capacity and demand response, at a total cost to consumers of $16.4 billion. Prices cleared at $325 per MW-day: precisely the administrative ceiling, for the third consecutive auction.

Without that ceiling, the auction would have cleared at $555/MW-day across the PJM footprint, and $777/MW-day in northern Illinois. The bill would have been $29.7 billion.

That gap — roughly $13 billion of demand that the market wanted to price and was not permitted to — is the clearest available measure of how far the AI buildout has outrun the grid that has to power it.

Three caps in a row is not a price signal

Capacity markets exist to do one job. They pay generators to promise availability years in advance, and the price is supposed to be a summons: when it rises, new supply is meant to answer. That is the entire theory. A high clearing price is not a failure of the market; it is the market working, right up until it isn't.

Three consecutive auctions at the ceiling means the signal is saturated. It cannot get louder. And the supply response has been close to nonexistent.

This auction brought in roughly 525 MW of new resources — and 208 MW of that figure was not new construction at all, but uprates to generation that already existed. Demand response, the other lever, moved backward: 7,365 MW of unforced capacity cleared, down 277 MW from the prior auction.

The reliability math got worse in the same motion. PJM ended up 6.8 GW below its 20% installed reserve margin target, a wider shortfall than the 6.5 GW gap recorded in the December 2025 auction. The region is paying record prices for a thinner cushion than it had last time.

"The current system doesn't work to bring online new capacity or stimulate demand response," Aurora Energy Research analyst Julia Hoos said of the result — naming the two resources the auction most needed and least produced.

Where the demand came from

PJM does not have to speculate about the source of the pressure. Of the $16.4 billion in capacity charges, roughly $6.3 billion traces to data-center-driven demand — a bit under 40 cents of every dollar. PJM COO Stu Bresler attributed the auction's dynamics to a roughly 2 GW increase in forecast demand, "largely caused by data center development."

The trajectory underneath is steeper than any single auction shows. Independent analysis puts PJM's annual customer capacity costs at $2.2 billion in 2024, rising to $14.7 billion the following year — a $12.5 billion annual increase absorbed by households and businesses across the territory.

This is the part of the AI capital cycle that does not appear in any hyperscaler's capex disclosure. A company announcing a multi-gigawatt training campus books the land, the silicon, and the construction. The capacity-market consequence of that announcement — the forecast revision, the tightened reserve margin, the higher clearing price paid by every ratepayer in 13 states — lands somewhere else entirely, on a bill that arrives monthly at an address with no relationship to the model being trained.

Critics have been direct about the asymmetry. The Sierra Club's Jessi Eidbo faulted PJM for failing to prioritize "responsible guardrails for data center development that protect households," arguing the market structure leaves ordinary customers carrying Big Tech's power bills. Eidbo also noted that despite favorable economics for battery storage, wind, and solar, participation from those resources in the auction was thin — a reminder that the shortfall is not purely a generation-cost problem but an interconnection and process problem.

The cap converts a price problem into a reliability problem

Price caps are politically legible. They keep a headline number from becoming a scandal, and they protect consumers from the sharpest edge of a supply crunch in the delivery year. What they cannot do is manufacture electrons.

When a capacity auction clears at its ceiling with a widening reserve shortfall, the constraint has quietly changed form. The market is no longer negotiating over how much capacity costs. It is discovering that the capacity does not exist to be bought, and the cap is the mechanism by which that discovery gets deferred rather than resolved. Suppressed prices are still prices; the difference is that the unmet portion shows up later, as a thinner margin on a hot August evening rather than as a line item in 2028.

Washington has noticed the political shape of this. The White House has been assembling utilities and data-center developers around a voluntary pledge to keep AI-driven demand growth from landing on household bills — an initiative whose existence concedes the premise that, absent intervention, it will. Voluntary is doing considerable work in that sentence. PJM's auction is a market outcome produced by a tariff, a forecast, and an interconnection queue, none of which respond to a pledge.

What this actually forecloses

The AI industry has spent two years treating power as a procurement problem: a matter of signing the right PPA, siting near the right substation, or funding the right nuclear restart. Those deals are real and some of them are large. But the capacity market is the aggregate, and the aggregate has now returned the same answer three times in a row.

The binding constraint on AI infrastructure in the eastern United States is not chips, capital, or even generation cost. It is the rate at which new firm capacity can clear an interconnection queue and reach commercial operation — a process measured in years, against a demand forecast being revised in gigawatt increments every auction cycle.

Until that changes, every additional campus announced inside PJM is a claim on a pool that is already short, priced at a ceiling that is already binding. The $13 billion the cap held back is not a savings. It is an invoice with the date left blank.