When PJM Interconnection published the results of its 2026/2027 Base Residual Auction on July 22, the headline was the price. Every zone in the footprint cleared at $329.17 per megawatt-day of unforced capacity, the cap agreed with Pennsylvania's governor and approved by the Federal Energy Regulatory Commission. The total bill came to about $16.1 billion, up from $14.7 billion a year earlier and from $2.2 billion for the 2024/2025 delivery year. The question that followed was how much of that cost is a direct consequence of data centers.
On October 1 the Independent Market Monitor for PJM, Monitoring Analytics, gave a precise answer. In the first part of its analysis of the auction, it re-ran the clearing without the existing and forecast data center load that PJM had built into its 2026 peak forecast. That load amounted to 11,993 MW. Total auction revenues in the actual results were $16,124,370,889. Removing the data center load cut them by $7,271,197,971. In other words, data centers accounted for 82.1% of the difference between what the auction cost and what it would have cost without them, by the monitor's method.
How the monitor reached its figure
The calculation is a counterfactual, and its assumptions matter. The monitor took the same supply offers that cleared in July and replaced the demand side with a version of the variable resource requirement curve built from a load forecast that excluded data centers, including both load already embedded in history and forecast load above that embedded level. Because supply in a three-year forward auction is relatively inelastic and the demand curve slopes downward, a lower requirement both reduces the quantity bought and slides the clearing point down to a lower price.
The monitor also pointed to growth in the data center adjustment itself. In the forecast used for the 2025/2026 auction, existing and planned data center load was 7,892 MW. For 2026/2027, the figure was 11,993 MW. PJM's own planning parameters show the RTO forecast peak rising from 153,883 MW to 159,329 MW between the two auctions, an increase of 5,446 MW or 3.5%. PJM's news release attributed that increase largely to data center expansion, electrification and economic growth. The monitor's calculation isolates the first of those.
The cap did a lot of work
The clearing price would have been higher without the collar. PJM's auction report notes that the BGE and Dominion zones had cleared at $466.35 and $444.26 in the previous auction, and that the rest of the RTO had cleared at $269.92. For 2026/2027, the cap of $329.17 was binding everywhere, so there were no constrained zones with separate prices. The cap derives from the $325 per megawatt-day figure in the agreement between PJM and Governor Josh Shapiro, converted into unforced capacity terms using the accredited capacity factor of a dual-fuel combustion turbine. The floor was $177.24.
The margin was thin. PJM said the cleared volume, plus capacity committed by utilities under the Fixed Resource Requirement alternative, was just 139 MW above the reliability requirement. The installed reserve margin target was 19.1%. In practice the system bought almost exactly what it needed and no more, at the maximum allowed price. That is the signal a capacity market is designed to send when supply is short. The difficulty is that new supply cannot respond inside a three-year window when interconnection studies, turbine orders and permitting all take longer than that.
What did respond
There was some supply response. PJM's presentation to its Markets and Reliability Committee noted 2,669 MW of unforced capacity from new generation and uprates offered and cleared, reversing a decline over the previous three auctions. About 1,100 MW of capacity interconnection rights had withdrawn planned retirements. The cleared mix was 45% natural gas, 22% coal, 21% nuclear, 4% hydro, 3% wind and 1% solar.
That mix says something about how data center load is being met in the near term. It is not mainly met by new wind and solar, which earn modest accredited capacity values under PJM's effective load carrying capability framework. It is met by keeping existing coal and gas units running and by squeezing more out of the nuclear fleet. Some of the coal capacity cleared because PJM itself offered reliability-must-run units into the auction, 1,637 MW of unforced capacity, at a zero price.
Who pays
PJM estimated that the cap price would translate to year-over-year increases of 1.5% to 5% in some customers' bills, depending on how utilities and states pass through wholesale costs. Customers in the BGE and Dominion zones could see small decreases because their prices fell. Those estimates are averages. Customers on default service in competitive retail states, where capacity costs flow through quickly, tend to feel auction results within a year.
The distributional question is the one regulators are now wrestling with. If data centers drive most of the increase, should data centers pay most of it? Under PJM's current design, the capacity cost is spread across all load in a zone according to each customer's share of peak demand. A data center pays its share, but so does every household, and the household's share of the price increase is set by a demand surge it did nothing to create. That is the logic behind large load tariffs with minimum demand charges, separate rate classes for large customers, and proposals that new large loads bring their own capacity.
Limits of the counterfactual
The monitor's figure is not a bill that can be sent to any one customer. A world without data center load would also have had different supply offers, different retirement decisions and possibly a different cap. Some resources might have retired earlier. The counterfactual holds supply fixed, which almost certainly overstates the effect of demand alone over a longer period, and understates it in the sense that it does not capture the effect of forecast data center load on future auctions. It is best read as a measure of the marginal pressure data centers put on a tight market in a single auction.
Even so, it is the clearest public quantification yet. It also arrives with the next auction, for 2027/2028, scheduled for December, and with PJM's board considering proposals on how to integrate large loads without compromising reliability. The cap remains in place for one more auction. After that, unless a new agreement is reached, the market will clear on its standard demand curve, and the data center share of the bill could be larger still.