PJM's Capacity Price Stopped Being a Price — Three Auctions at the Cap, Two of Them Short

Three numbers, from the three most recent capacity auctions run by PJM, the grid operator for 67 million people across 13 states and the District of Columbia:

Read as a series, that is a market that spiked and then settled — three years of near-identical prices, the last one slightly down. It is the shape of a problem being absorbed.

It is not a series. All three are the same number: the administrative ceiling. And underneath them, the thing that was actually moving got worse every year.

What the cap replaced

Capacity auctions are how PJM pays generators to exist — to be available in three years' time, whether or not they run. For a decade the clearing price wandered between $28.92 and $164.77 per MW-day, and the wandering was the information: a low price meant spare capacity, a high one meant scarcity.

Then the 2025/2026 auction cleared at $269.92/MW-day, against $28.92 the year before. Total cost to load went from $2.2 billion to $14.7 billion. Two zones cleared higher still — Baltimore Gas & Electric at $466.35 and Dominion at $444.26.

In response, PJM, the governors of all 13 states, and FERC agreed a price collar — a cap and a floor — for four auctions, "to protect both consumers and investors from market volatility."

Every auction since has printed the cap.

The number that kept moving

1. 2026/27

   Cleared 139 MW above the reliability requirement

   PJM procured 134,311 MW, plus 11,933 MW under the Fixed Resource Requirement. The margin over what PJM's own standard required was 139 MW — on a system of roughly 146,000 MW.

2. 2027/28

   Fell 6,623 MW short

   The first auction in which the entire RTO, including the Fixed Resource Requirement areas, fell short of the reliability requirement. PJM's later release rounds this to "approximately 6,500 MW"; the auction's own release states 6,623.

3. 2028/29

   Fell 6,831 MW short

   Short again, by more. PJM plans to ask FERC for a special "Backstop Procurement" in September to address the near-term gap.

Those three prices — $329.17, $333.44, $325.00 — describe a system that went from a 139 MW cushion to a 6,831 MW hole. The price says almost nothing about that, because the price is not free to say it.

PJM is direct about the limit: "While the price cap and floor may reduce volatility, they do not solve the underlying supply-demand imbalance."

Why the prices are not comparable anyway

There is a second reason to distrust the flat line, and it has nothing to do with the collar.

PJM says its own year-over-year comparison does not hold. From the 2025/2026 report: "2025/26 to 2024/25 Delivery Year supply and demand comparisons are not straightforward because of the implementation of marginal Effective Load Carrying Capability accreditation for all resources and the associated reduction of the reliability requirement through the Forecast Pool Requirement (FPR) as well as the transition of load from FRR into RPM."

In plain terms: the rules for counting how much a generator contributes changed, and load moved between two different accounting regimes, in the same year the price jumped tenfold. Some of that 833% is scarcity. Some of it is arithmetic. PJM does not separate the two, and neither will I.

The cap adds a third break. It is "calculated using the accredited capacity of the PJM reference resource" — so the ceiling itself moves year to year for reasons unrelated to whether the lights stay on. The 2.5% fall from $333.44 to $325.00 is a change in the reference resource's accreditation. It is not good news, and it is not bad news. It is not news about scarcity at all.

Where the demand comes from

PJM names it. The 2028/2029 auction "saw the continued trend of the addition of large data center loads to the load forecast," and among the fixes PJM lists is developing "Connect and Manage" frameworks that let "large new loads, such as data centers, connect to the system but operate flexibly when needed to limit disruptions to other consumers."

PJM president and CEO David Mills, in the same release: "These auction results show that demand for electricity continues to grow faster than electricity supply."

There is a detail in PJM's remedy list worth sitting with, given this site's subject. Among the actions to clear the interconnection backlog is "a collaboration with Google's Tapestry to leverage AI to reduce study timelines." The queue is congested substantially by load that AI is driving, and the tool being brought to unclog it is AI, supplied by one of the companies driving the load.

What I am not claiming

I have not established that data centers caused these prices. PJM names them as a growing component of the load forecast; it does not quantify their share of the clearing price, and neither does this piece. Retirements, the ELCC accreditation change, the FRR-to-RPM transition, and a compressed auction schedule are all in the same window.

I have also not converted any of this into a household bill. The cost-to-load line in the chart stops at 2025/26 deliberately. PJM's July 2026 release gives $16.4 billion as cleared supply times clearing price for 2028/29 and says in the same paragraph that this "does not equate to the total cost to load because load that is hedged through self-supply or bilateral contracts is not exposed to the clearing prices in the auction." Those are two different quantities. Plotting them on one line would make a cleaner chart and a false one.

The takeaway

A market price that equals its own ceiling is not measuring scarcity. It is measuring where the ceiling was set. For three auctions running, PJM's capacity price has told you the cap and nothing else — while the quantity it used to express moved from a 139 MW surplus to a 6,831 MW deficit, twice in a row, for the first time in the market's history.

If you are watching PJM to understand what the AI build-out is doing to the American grid, the price is the wrong number to watch. Watch the shortfall.

How I compiled this

Every figure is transcribed from a PJM primary document — the 2025/2026 Base Residual Auction Report for the eleven-year series, and PJM's own news releases and Inside Lines posts for the three capped years. No figure here comes from secondary coverage.

That distinction earned its keep three times during the compilation.

Secondary reporting gave the BGE zonal price as $466.36; PJM's Table 1 says $466.35. A cent, on a number nobody would check.

PJM's July 2026 release describes the previous year's shortfall as "approximately 6,500 MW". The 2027/2028 auction's own release, seven months earlier, states 6,623 MW. The rounded figure is PJM's, not a reporter's — which is the point: a primary source summarising its own earlier work is still a secondary account of it.

And Table 2 publishes two reserve margins, an RPM one and a total one that PJM's footnote defines as including FRR. The capped-year figures PJM gives in its releases — 14.8% and 14.7% — are the system-wide kind, quoted alongside FRR resources. They belong in the total column, not the RPM column, and an earlier draft of the dataset had them in the wrong one. Two numbers under one heading is how a spreadsheet tells a lie without containing a false figure.

It is the reason the compiled series in data/pjm-capacity/ records a source for every row, and why the columns that PJM does not publish on a consistent basis are left empty rather than filled in from somewhere else.