North American grid regulators are paying closer attention to the electricity habits of hyperscale AI data centers after a series of large, abrupt load disconnections rattled grid operators during 2025. In its 2026 State of Reliability report, the North American Electric Reliability Corporation (NERC) documents multiple incidents in which more than 1 GW of data center demand dropped off the grid within moments of transmission disturbances.
The most significant event, recorded in February 2025, shed roughly 1,800 MW of data center load following a transmission fault. A second event in June 2025 involved approximately 1,300 MW. NERC also recorded additional data center load reductions ranging from around 200 MW to more than 500 MW throughout the year. The Electric Reliability Council of Texas (ERCOT) separately logged nine cryptocurrency mining load-loss events each exceeding 100 MW.
Why This Is Different From Traditional Industrial Load
Conventional industrial facilities shed load gradually. Large computational facilities do not. Their power systems can disconnect rapidly when transmission disturbances occur, and as individual campuses approach gigawatt scale, those sudden changes become system-wide events rather than localized blips. NERC noted directly in the report that if large load facilities continue to grow and cluster regionally, “this could lead to measurable frequency or voltage stability issues.”
The Electric Power Research Institute (EPRI) described the problem as a longstanding technical issue that AI campus growth has pushed into material territory. Parag Mitra, a senior principal technical leader at EPRI, explained that existing data center equipment was not designed to ride through normally cleared grid disturbances, and that the rapid increase in data center load capacity has magnified the problem to the point where it affects grid reliability. Mitra also noted that utilities still lack both the detailed operational data and the standardized validation methods needed to accurately model large data center behavior.
NERC’s Response
NERC’s reaction goes beyond documenting what happened. During 2025 and early 2026, the organization took several concrete steps:
- Issued industry alerts on large computational load behavior
- Developed guidance for modeling computational loads in planning studies
- Published a technical reference for representing data centers in transient stability analysis
- Endorsed the PERC1 performance model for simulating data center behavior during disturbances
- Advanced draft Rules of Procedure to create a new registered entity category specifically for computational load facilities
Vikhyat Chaudhry, co-founder and CTO of Buzz Solutions, put the operational shift plainly: a single AI campus can now represent demand equivalent to an entire city, which changes how utilities plan, monitor, and operate the grid assets that serve them. He emphasized that earlier coordination between utilities and hyperscale developers, combined with better data sharing, will be increasingly important as new campuses come online.
Conventional Generation Under Pressure Too
The report does not focus solely on data centers. NERC also flagged rising stress on conventional generation. Weighted equivalent forced outage rates climbed to 9.2% in 2025, above the historical range of 7% to 8%. Coal plants contributed the largest share of unavailable generation, followed by combined-cycle natural gas units. NERC recommends that utilities and regulators consider whether reserve margin requirements need to be raised to account for this reduced conventional generation availability.
For hosting and cloud infrastructure professionals, the broader takeaway is straightforward: electricity access and grid coordination are no longer background concerns for hyperscale operators. They are active variables in site selection, campus design, and long-term capacity planning.