An AIxEnergy.io analysis of a recent 3 GW data center demand loss in Virginia argues that grid reliability rules must address sudden load-shedding operational behavior rather than focusing solely on tariff costs.

The loss of more than 3 GW of data center demand in Virginia demonstrates why grid policy must address operational behavior rather than focusing solely on costs, writes Brandon Owens, founder of AIxEnergy.io, a platform that provides data on energy infrastructure and artificial intelligence. At 7:56 a.m. EDT on July 22, 2026, Ting Labs detected the onset of a major transmission disturbance in Northern Virginia. PJM Interconnection subsequently reported that more than 3 GW of power demand—approximately 3% of system demand at the time—went offline as affected data centers rapidly transferred to backup power. Ting’s sensor analysis recorded a frequency rise and voltage effects across the Eastern Interconnection. While PJM stated the disturbance caused no reliability impact and Dominion Energy confirmed operators stabilized conditions and returned the system to normal within minutes, the event remains highly significant from a reliability standpoint. It serves less as an isolated incident and more as a harbinger of what lies ahead, underscoring the urgent need to address the escalating risks posed by the global influx of massive electrical loads.

This Northern Virginia disturbance was far from unprecedented. Between November 2020 and March 2023, the Electric Reliability Council of Texas (ERCOT) identified eight events in which faults near a large Texas Gulf Coast industrial load triggered repeated demand reductions of approximately 400 MW to 700 MW. ERCOT recorded system frequencies reaching as high as approximately 60.11 Hz, noting that subsequent adjustments to variable-frequency-drive settings and internal controls later improved the facility’s ride-through performance. Similarly, on December 7, 2022, multiple faults and a delayed 19-cycle clearing following a breaker failure resulted in an approximate 1,560 MW load reduction in West Texas. Of that total, ten large power-electronic loads accounted for roughly 162 MW, oil-and-gas production, processing, and delivery facilities contributed about 420 MW, and 112 MW of thermal generation also tripped offline.

Such disturbances are not confined to North America. EirGrid and SONI documented four major data center demand reductions linked to Irish 220-kV transmission events: 74 MW on January 7, 2022; 204 MW on December 13, 2022; 321 MW on January 26, 2025; and 387 MW on May 8, 2025. In response, Irish system operators proposed Grid Code Modification MPID345, which introduces rate-of-change-of-frequency thresholds, voltage fault ride-through mandates, and post-fault active-power-recovery requirements. Under the proposal, a facility could transfer demand to backup systems during a voltage dip but would generally be required to restore at least 90% of its pre-fault demand within 500 milliseconds after fault clearance and voltage recovery. The modification remains under regulatory review.

Collectively, these incidents in Virginia, Texas, and Ireland highlight the pressing need for a more comprehensive large-load architecture tailored to grid operators. Historically, most large-load debates have centered on whether sufficient generation and transmission infrastructure can be built, when projects may energize, what financial security developers should provide, and who should bear the cost of infrastructure upgrades. While critical, these questions only address whether the system can physically connect to and serve the load—not how that load behaves once operational. A data center may successfully procure generation, fund network upgrades, and satisfy all applicable capacity and interconnection obligations, yet still present a poorly modeled common-mode transfer risk during a grid disturbance.

What might a credible operating architecture for such large loads entail? A robust framework should: (1) identify the largest plausible simultaneous demand reduction at both individual facility and electrical-cluster levels; (2) establish performance-based voltage and frequency ride-through requirements; (3) mandate verified as-built models covering information-technology loads, cooling systems, uninterruptible power supplies, protection settings, backup generation, transfer logic, and reconnection timing; (4) provide grid operators with timely telemetry on real and reactive power, voltage, transfer status, and expected restoration behavior; (5) specify clear procedures for notification, ramping, restoration, and battery recharging; and (6) preserve a shared forensic record following material events. Where a facility’s scale, concentration, or control behavior necessitates additional instrumentation, reactive support, reserves, or protection modifications, associated costs should be allocated according to established cost-causation principles. North American reliability authorities are already beginning to adopt this approach.

Regulatory momentum is building. NERC’s May 2026 Level 3 Alert calls for enhanced computational-load modeling, studies, instrumentation, commissioning, operational coordination, protection, and control, though it does not constitute an enforceable Reliability Standard. In July, the Federal Energy Regulatory Commission (FERC) directed NERC to develop new or modified computational-load reliability standards and registration criteria, with compliance filings due by December 31, 2026.

For most of the power industry’s history, reliability planning has concentrated on the sudden loss of supply. The computational era is creating its mirror image: the sudden loss—and return—of demand. The next major contingency may not originate with the trip of a power plant or transmission line. Instead, it may begin behind the meter, driven by thousands of power-electronic devices responding simultaneously to the same disturbance, each safeguarding its own facility exactly as engineered.