PJM locked in $325/MW-day power costs on demand projections it cannot verify until 2028
PJM's capacity auction cleared at the price cap with a 6.8-GW shortfall, but the AI demand driving that outcome won't be confirmed or refuted until after costs are already locked in for 67 million people.
An Oilprice.com analysis published Saturday (2026-08-22) identified a timing problem embedded in PJM's capacity market: if the AI-driven electricity demand underpinning this summer's auction proves overstated, the correction arrives as canceled slot reservations in 2028 and 2029 — not before. By then, power costs for 67 million people across 13 states will already have been repriced upward.6
That pricing already happened. PJM Interconnection's capacity auction for the 2028-2029 delivery year, concluded on Tuesday (2026-07-14), cleared at the $325/MW-day price cap across the entire footprint. It was the third consecutive auction in which PJM failed to attract enough supply commitments to meet its reliability standard, producing a 6.8-GW shortfall, Utility Dive reported.4,3
The market's read has been straightforward: data centers are consuming power faster than new generation can be built, the grid is short, and prices will stay elevated. PJM's board reinforced that view on Monday (2026-07-27), proposing a backstop capacity auction to launch in September, along with new provisions allowing curtailment of data center load under tight supply conditions.5
What that framing sets aside is PJM's own near-term capacity picture. The grid operator's Winter Outlook from November 2025 projected 180,800 MW of operational capacity against a forecast peak of approximately 145,700 MW. PJM described its resources as adequate under expected conditions. The all-time winter peak reached 143,700 MW on January 22 (2026-01-22), roughly 2,000 MW below the season's forecast ceiling.2
Near-term adequacy and multi-year shortfall can coexist. PJM's current reserve margins reflect capacity contracted years ago, while the 6.8-GW gap reflects what generators won't commit to for 2028. But the distance between those two numbers is worth holding alongside the demand projections driving the auction result. PJM added approximately 4,800 MW of nameplate generation since the prior winter season, mostly solar, contributing around 1,000 MW of effective winter capacity — a modest increment against the load growth embedded in the auction.2
The accelerated backstop timeline adds weight to the urgency argument. PJM moved the auction forward from 2027 to September after concluding, as it stated in a release dated May 21 (2026-05-21), that waiting would leave insufficient time to close the reliability gap. That urgency rests on the same demand projections now in question. The September auction will clear against interconnection queue data that includes data center requests which are not firm load commitments.1
The Oilprice.com analysis published Saturday (2026-08-22) noted that the supply-side constraint — gas turbine lead times — holds regardless of how the demand question resolves. New turbines cannot arrive before 2027 even if ordered now. The near-term capacity gap is real on the supply side. The demand side is another matter.6
Analysts do not expect the backstop mechanism to stay exceptional. Utility Dive cited market observers who see it becoming a recurring feature. "It's hard to see how we can return to normal," one analyst said, reflecting a broader view that PJM's standard auction process has broken down structurally.4
What would clarify the demand picture is movement in PJM's interconnection queue. A material increase in queue withdrawals or deferrals by data center developers ahead of or following the September backstop auction would suggest load growth is being revised down. If instead the backstop clears near the $325/MW-day cap with a residual shortfall similar to July's result, the supply deficit is genuine and the 2027 constraint arrives on schedule regardless of what the AI build-out ultimately delivers. The backstop results are the first hard data point capable of separating those two outcomes.5,1