Data Center Volatility and Capacity Illusions Cloud PJM's Winter Reliability Picture
A 3,800 MW data center trip in July put computational loads under PJM scrutiny, while NERC's winter capacity headline conceals thin firm generation and declining wind credits.
About 3,800 MW of data center load tripped offline in northern Virginia on July 22, the largest such event in PJM's history, the grid operator announced Tuesday (2026-08-11). The incident unfolded in two waves, pulling PJM's systemwide load from 99,984 MW to 96,205 MW, a fall of 3.8%, before recovering. Two earlier incidents in Dominion's zone, in 2025 and 2024, each caused sudden load transfers of roughly 1,500 MW. The July event exceeded both by a considerable margin.6
PJM said Tuesday (2026-08-11) it is now considering "ride-through" requirements and other interconnection reliability standards specifically for computational loads, a category that includes data centers and cryptocurrency mining facilities.6 Northern Virginia hosts the world's largest concentration of data center capacity, and the region's rapid electricity demand growth has underpinned bullish load assumptions across PJM's footprint for years.
Market positioning in PJM real-time remains broadly bullish, in part because NERC's most recent winter reliability assessment, published in June 2026, projected total bulk power system capacity rising 20.2 GW, or 2.5%, over the previous season's forecast.1 A contrarian demand signal has emerged alongside that bullish consensus, pointing with relatively high confidence in the opposite direction.
But the capacity headline is harder to read than it appears. Of the 9,445 MW increase in on-peak demand capacity in NERC's assessment, generation contributes just 1,335 MW. The larger share comes from demand response programs, which depend on voluntary customer curtailment and have a weaker track record in the deep-cold events that push the grid hardest.1
Battery additions look substantial on paper. NERC counted 19,659 MW of new battery nameplate capacity translating to 11,121 MW of credited on-peak demand capacity. Solar shows the conversion problem more sharply: 11,097 MW of nameplate yields only 1,176 MW of credited peak capacity, a reflection of how little solar generates during winter demand peaks. Wind moved in reverse, with nameplate capacity declining 562 MW and on-peak capacity values revised downward in several assessment areas because of reassessed output at peak demand hours.1
That wind revision is getting insufficient attention. Grid operators have been systematically reassessing how much output wind resources actually deliver when the system is under maximum stress, and NERC's language suggests those revisions have already spread across multiple assessment areas. If they continue, the capacity buffer embedded in winter outlooks erodes without a headline event to prompt a market reaction.1
MISO's operational data from recent cold periods puts numbers on what stress actually looks like. During a January cold event at 6.5°F, MISO's peak demand reached 108 GW while renewable output registered 19 GW, roughly 18% of that peak total, with incremental outages over the three-day period reaching 9 GW. Thermal generation and demand response absorbed the shortfall.3
Post-Uri reforms have produced real changes: weatherization requirements, cold weather critical component inventories, dual-fuel conversions, and PJM's practice of running unannounced operational tests on generators that have not cycled for several weeks before cold events.4,2 Still, a Utility Dive review found that power plants retained under DOE emergency orders were producing significantly less electricity than before those orders were issued, which raises questions about whether administrative retention of capacity translates into available megawatts when temperatures drop.5
Back to the data center side: new ride-through or interconnection requirements, if adopted by PJM with hard curtailment limits, would constrain load growth in northern Virginia. That reduces the demand baseline underpinning forward capacity needs — a bearish demand signal at a moment when reliability planning still assumes continued aggressive expansion by computational loads.6
PJM's release of the specific terms of any proposed reliability standards for computational loads is the clearest near-term test. A sustained cold event across PJM and MISO this winter 2026-2027, one that exposes the gap between wind nameplate and deliverable capacity while data center behavior adds an unpredictable layer to the demand curve, would put both sides of NERC's capacity equation under simultaneous pressure.1,6