PJM's computational load gap complicates NERC's winter capacity picture
A 3,800 MW load-trip event and a wind valuation markdown put pressure on a headline reserve increase built largely on demand response.
PJM announced on Tuesday (2026-08-11) it was evaluating new interconnection reliability requirements for data centers and cryptocurrency-mining facilities after about 3,800 MW of computational load tripped offline in a single event, Utility Dive reported.5
The review points to a growing mismatch between how computational loads are modelled in winter reliability planning and how they actually behave under grid stress.5
NERC's Winter Reliability Assessment, published in early June 2026, projects the bulk power system will carry 20.2 GW more capacity into the 2025–2026 winter than the prior season, a 2.5% increase. Power markets have broadly read that figure as adequate. But the composition of the increase deserves closer examination.1
Battery storage accounts for 19,659 MW of nameplate additions and 11,121 MW of effective peak demand capacity. Solar contributes 11,097 MW nameplate but just 1,176 MW of peak demand capacity — a steep derate driven by limited winter solar hours. Wind is a net negative in nameplate terms at -562 MW, and on-peak capacity values for wind have been revised lower across several assessment areas due to updated resource capability valuations at peak demand hours.1
The source of the 9,445 MW improvement in total usable capacity is equally instructive. Just 1,335 MW comes from generation additions. The dominant contributor is demand response programs.1
Demand response is not the same as firm generation. Winter Storm Fern, the most recent cold-weather stress event to test PJM and MISO, left the system running "very close to the edge" by the assessment of the five-year Uri retrospective, with demand curtailment programs historically prone to underperformance when gas supply and load stress align.4
MISO's seasonal readiness data from January 2026 adds granularity. During the cold snap of January 20–22, 2026, with temperatures falling to 6.5°F, MISO recorded peak demand of 108 GW against just 19 GW of renewable output, with incremental outages reaching 9 GW and three-day uplift costs of $1.5 million.3
The PJM computational load incident sits on top of those supply-side uncertainties. Data centers and crypto-mining facilities have grown rapidly in PJM territory without any obligation to maintain ride-through capability during grid emergencies. The 3,800 MW that disconnected on Tuesday (2026-08-11) would constitute a significant forced generation outage if it were on the supply side. In planning models, it is demand that can vanish without notice.5
PJM's Seasonal Outlook notes the grid operator plans unannounced operational tests of generators that have not run for several weeks ahead of winter, specifically to screen for cold-weather performance limitations. That practice reflects a hard lesson: nameplate capacity and deliverable winter capacity diverge. No equivalent screen exists for large, uncharacterized demand blocks.2
The 20.2 GW headline still represents a real increase in available resources. But it rests on demand response accounting for the bulk of usable capacity gains, wind on-peak values already marked down once, and a rapidly expanding computational load base that carries no mandatory reliability obligation. Any two of those conditions performing worse than planned simultaneously would make the headline margin look considerably thinner than it reads on paper.1,5
PJM's timeline for finalizing computational load ride-through standards and MISO's unforced outage totals during the first sustained cold snap of the 2026–2027 winter are the two data points that will either support or erode the current adequacy assessment.3,1,5