Updated October 2026: this analysis corrects the July edition’s description of the Duke study and of the NERC and FERC records.
A thesis has been circulating in energy and data center circles: that a third of U.S. grid capacity sits unused, and that flexible demand could unlock 100–300 GW of new data center load without building a new power plant.
The Duke University study behind it is more careful than the popular framing. Its finding: if new large loads accepted curtailment for about 0.5% of their running time, roughly 177 hours a year and mostly partial, the largest U.S. balancing authorities could take on about 98 GW of new load before exceeding what planners already prepare for. That’s a meaningful finding, and its authors are clear about one limit: they did not model transmission.
There is a second limit. The study’s curtailment is planned: scheduled, ramped and partial. The load losses NERC is now acting on were not. On July 10, 2024, six normally cleared faults on a Virginia transmission line caused about 1,500 MW of data center load to disconnect at once as the facilities switched to backup power. In July 2026, a load reduction of nearly 4 GW followed in the same region. NERC says the grid has “not experienced, nor planned or operated for” simultaneous load losses on this scale.
This is the gap between the administrative map and the physical terrain.
Regulators are working both sides of it. FERC’s June 18, 2026 show-cause orders asked all six FERC-regulated RTOs/ISOs to fix how they study and connect large loads, including through flexible large-load services. NERC has issued a Level 3 alert, a reliability guideline that calls for power quality monitoring before commissioning, and new standards requiring disturbance recording at large sites.
A capacity study tells you what the spreadsheet says. Measurement at the point of connection tells you what the physics says. The full analysis looks at three limits of the headroom thesis and what the regulatory record shows.
Leave a Reply