A single transmission fault near Ashburn, Virginia, knocked more than three gigawatts of data center load off the PJM Interconnection on July 22, 2026, and the grid absorbed the shock without a reliability breach on the bulk power system. Data centers across the region shifted to backup power almost simultaneously, an uncoordinated response that grid operators had to manage in real time. The events have pushed ride-through from a specialized engineering concern toward an explicit regulatory priority, with NERC developing national reliability standards and regional markets moving ahead with requirements for large computational loads. What began as a technical footnote inside PJM’s operations reports has become the defining reliability story of 2026 for anyone building or financing large-scale compute infrastructure.

A Single Fault, Three Gigawatts Gone

The Ashburn event stands out because of its size, its speed, and its geography, occurring in Northern Virginia, the largest concentration of data centers in the United States. Three gigawatts represents enough load to power a mid-sized American city, and it disappeared from the grid within seconds of the fault occurring. PJM’s data show the bulk power system held steady throughout the disturbance, which grid planners see as a reassurance and a warning in equal measure. The reassurance comes from the fact that transmission-level reliability never came under threat. The warning comes from the realization that a single, localized fault can trigger a multi-gigawatt customer response that no operator explicitly designed for. Data centers in Northern Virginia transferred to backup generation in response to the disturbance, producing a rapid and unusually large customer-side reduction in PJM load. 

A lightning arrester failure on a 230-kV line in the Eastern Interconnection triggered roughly 1.5 gigawatts of data center load to drop off the grid that day. NERC investigated the incident and found that customer-side protection and control schemes shed the load entirely, with no utility equipment disconnecting anywhere in the sequence. The sustained loss, measured at 1.26 gigawatts, traced back to an interaction between the line’s reclosing sequence and data center protection schemes designed to trip after repeated voltage disturbances. Facilities stayed on backup power for hours rather than seconds, which is the detail that alarmed reliability planners most. That documented interaction from 2024 provided an important technical precedent as regulators examined the larger customer-initiated load reductions emerging in 2026.

A Pattern Regulators Can No Longer Ignore

NERC’s subsequent work widened the lens beyond the 2024 incident and identified multiple large customer-initiated load reductions during 2024 and 2025, including events involving 1,000 MW or more of unexpected large-load reduction. NERC’s tracking of large-load events during 2024 and 2025 showed that the risk was not confined to a single incident or interconnection. The Electric Reliability Council of Texas has responded to the same emerging reliability concern by establishing frequency and voltage ride-through requirements for Large Computational Loads. As data center campuses grow larger and cluster more tightly around shared substations, each individual protection trip carries a bigger systemic footprint. Frequency and voltage stability, once discussed as a theoretical concern tied to hyperscale growth, now appear as line items in regulatory filings. Consequently, NERC’s work has shifted from investigating and monitoring large computational-load behavior toward interim reliability actions and formal standard development.

Registered entities and large computational-load operators now face a series of requirements and implementation milestones across NERC and regional markets during the same calendar year, leaving little room for delayed engineering work. NERC issued a Level 3 “Essential Actions” alert on computational load on May 4, 2026, with formal responses due back from load entities by August 3, 2026. In Texas, the Public Utility Commission approved ERCOT’s Nodal Operating Guide Revision Request 282, establishing ride-through requirements that took effect on August 1, 2026. The Federal Energy Regulatory Commission then directed NERC, on July 16, 2026, to file new or modified reliability standards specific to computational loads no later than December 31, 2026. Registered entities and large computational-load operators now face a series of requirements and implementation milestones across NERC and regional markets during the same calendar year, leaving little room for delayed engineering work.

Who Falls Under NERC’s New Threshold

These requirements add a new layer of engineering, modeling and documentation work for entities responsible for integrating large computational loads into the bulk power system. The draft covers aggregate connected loads of 20 megawatts or more at a single point of interconnection, operating at 60 kilovolts or above, provided the site hosts at least one megawatt of computational load. If adopted, that threshold would extend the proposed computational-load registration framework beyond the largest hyperscale campuses to some colocation and enterprise facilities that meet the specified technical criteria. The alert requires applicable entities to improve their understanding of computational-load behavior through modeling, studies, instrumentation, commissioning, operations, protection and control information. These requirements add a new layer of engineering, modeling and documentation work for entities responsible for integrating large computational loads into the bulk power system.

That shift requires coordinated protection and control settings that allow facilities to remain connected through specified disturbances while avoiding unnecessary protective disconnections that can magnify grid impacts. Operators must coordinate protection and control settings so facilities can remain connected through specified disturbances while avoiding unnecessary protective disconnections that can magnify grid impacts. That shift requires sharing protection settings directly with grid planners, installing fault recorders capable of capturing disturbance data, and running commissioning tests that simulate real voltage swings under controlled conditions. Coordination with transmission and distribution operators, including documented operating procedures and facility performance information, is becoming an important part of the emerging reliability framework rather than an informal courtesy. Vendor firmware adjustments, equipment tuning and, where existing equipment cannot meet the required performance, hardware or design changes are among the measures facilities may need to consider.

Reconnection Is the Overlooked Half of the Problem

If a facility cannot demonstrate controlled reconnection and predictable load recovery, grid planners have less certainty about how that facility will behave during restoration following a disturbance. Data centers increasingly need controlled reconnection sequences so that large blocks of load do not return to the grid simultaneously without an agreed ramp profile. An unmanaged return of gigawatt-scale load can create new voltage and balancing events of its own, effectively trading one disruption for another. Grid operators view uncontrolled load shedding as a direct threat to the industry’s broader credibility on flexibility programs. If a facility cannot demonstrate controlled reconnection and predictable load recovery, grid planners have less certainty about how that facility will behave during restoration following a disturbance. Ride-through capability and flexibility, in this framing, are not separate initiatives but two outputs of the same underlying commissioning process and asset base.

Shalin Savalia, a senior electrical engineer at Amazon Web Services, says the data center industry should not wait for regulators to force these changes. Savalia emphasizes that operators should coordinate their protection settings with utility practices, particularly around reclosing sequences that can interact with campus-level protection logic. He also says facilities must document their actual, measured ride-through capabilities rather than allowing planners to default to conservative assumptions that penalize the entire compute load category. Accurate data, in his view, protects well-designed facilities from being lumped in with poorly protected ones. His position reflects an engineering argument that facilities can reduce uncertainty by documenting their actual protection and ride-through performance before regulators establish the final national standards.

The Strategic Stakes for Compute Infrastructure

Large increments of new AI and cloud demand now carry greater reliability considerations than they did several years ago, particularly where computational loads can disconnect rapidly during grid disturbances. Site selection decisions are increasingly shaped by grid reliability, interconnection conditions and the behavior of the transmission systems serving prospective campuses, alongside power price and fiber access. As computational-load reliability requirements develop, financing and investment assessments may increasingly consider the compliance and grid-integration risks associated with new campuses alongside power purchase agreements. Meanwhile, operators that move early on fault recording, protection coordination and commissioning tests can give utilities and planners better evidence of how their facilities will behave during disturbances. The December 31, 2026 deadline set by FERC gives NERC a national timetable for submitting computational-load reliability standards and related registry changes, while regional requirements such as ERCOT’s ride-through rules continue to develop alongside it.