PJM real-time power markets underestimate demand elasticity, July data suggest
A sudden 3,000 megawatt demand drop in July and price-cap distortions in the capacity auction complicate the data-center demand narrative driving PJM prices.
PJM Western Hub real-time power traded at $78.03 per megawatt-hour on Thursday (2026-08-20), holding elevated levels as summer heat sustains the bull narrative that surging data center load has permanently tightened the Mid-Atlantic grid. Capacity auction results and grid operator filings have reinforced that view: the 2028/2029 delivery year auction came up 6.8 gigawatts short, prices cleared at the regulatory ceiling of $554.72 per megawatt-day across most of the footprint, and PJM pulled forward its backstop reliability auction to September this year rather than waiting until 2027.3,1
But on Wednesday (2026-07-22), something happened that does not fit the demand-as-baseload narrative. WattClarity reported a frequency spike in PJM accompanied by a roughly 3,000 megawatt demand drop just before 08:00 local time — a steep, near-instantaneous fall consistent with large industrial load interruption or widespread automated demand response triggering simultaneously. PJM had published no explanation as of WattClarity's report on July 23 (2026-07-23).5
The absence of a disclosed cause matters more than the event itself. If a single occurrence can shed 3,000 megawatts within seconds, effective peak demand is more elastic than the forward capacity framework assumes. The entire bull thesis is built on the premise that data center load, once energized, is effectively firm: it runs around the clock, does not curtail, and cannot shift. One unexplained event does not disprove that premise. It raises a question the market has not priced.5
The capacity auction numbers carry their own distortion. OilPrice.com reported that the 2028/2029 auction cleared at the $554.72 per megawatt-day cap across most of PJM, with the COMED local delivery area reaching a higher ceiling of $776.69. Without the regulatory cap, the outlet reported, prices would have been approximately 70% above the clearing level. Markets cannot price scarcity efficiently when the signal is truncated at a ceiling. Generators are responding to an artificially bounded price, and the supply response could undershoot actual requirements or overshoot if caps are eventually lifted — either way, the current clearing price tells traders less than it appears to.3
PJM's own resource adequacy data complicates the tightness story further. The grid operator's November 2025 seasonal outlook projected 180,800 megawatts of operational capacity for the 2025/2026 winter, against a forecast peak of roughly 145,700 megawatts — a reserve buffer exceeding 35,000 megawatts. That forecast peak, if reached, would set a new winter record, surpassing the all-time winter high of 143,700 megawatts set on January 22, 2025 (2025-01-22).2
The arithmetic creates tension. A capacity auction short 6.8 gigawatts yet a reserve margin running above 35,000 megawatts signals something structural about when and where tightness occurs — not that tightness is absent, but that it is more localized and time-specific than the aggregate bull narrative implies. Part of the explanation is mechanical: PJM's capacity market procures for summer peak conditions roughly three years forward, while reserve margins reflect available winter resources. Still, the buffer scale argues for more precision in how forward tightness is characterized.2,3
PJM's latest policy direction adds to the ambiguity. Rather than centrally procuring additional capacity to absorb data center growth, PJM has indicated it expects states to manage demand directly at peak moments, Canary Media reported on August 5 (2026-08-05). Qualifying resources include demand-response aggregations and virtual power plants — controllable loads, not firm generation. The grid operator is simultaneously running an early backstop auction and directing states toward demand-side solutions. That combination suggests PJM's internal assessment of load firmness differs from the market narrative.6
Bilateral contracts are also pulling large data center loads outside the centralized auction mechanism, Canary Media reported on July 17 (2026-07-17). Direct agreements between data centers and generation developers mean capacity clearing prices become progressively less representative of full grid stress, which further muddies forward price signals.4
Demand growth in PJM is real. The dispute is about its hourly shape, its flexibility, and how much is interruptible — none of which is resolved by the capacity auction shortfall headlines. The September backstop auction result, and specifically whether it clears above or below the regulatory cap, will be the first hard data point capable of testing whether ceilings or genuine scarcity drove the 2028/2029 outcome. Before that, any PJM disclosure on the cause of the July 22 (2026-07-22) demand drop would do more to resolve the bull case than anything in the forward curve.5,1