PJM’s Largest Load Transfer Took 3,800 MW of Data Centers Offline After a Fault That Cleared Normally
- Load transferred to backup: 3,800 MW
- First wave: 2,970 MW
- Second wave: 1,099 MW
- Fault current: 28.8 kA
- NERC filing deadline: December 31, 2026
Roughly 3,800 MW of data center load left the PJM grid on July 22, 2026. PJM Interconnection presented the event to its Operating Committee on August 6. The load transferred to backup generation across northern Virginia. A 230 kV line fault inside Dominion Energy’s zone started it at 07:56. PJM calls the event the largest load transfer in its history. The fault itself cleared properly in two to three cycles, with no relay misoperations and no stuck breakers.
Dominion traced the fault to a single sub-conductor of a double bundle. It pulled free from its mechanical compression sleeve on a dead-end structure outside the Roundtable substation. The apparent cause was overheating at the sleeve, from improperly torqued bolts on the pad-to-pad connection. Fault current reached 28.8 kA. A formal root cause analysis lands in September 2026.
The load left in two steps. Dominion’s timeline puts the initial loss of 2,970 MW at 07:56:33, during the reclosing sequence. That drop raised system voltage. The high voltage then pushed a second block of 1,099 MW off the grid at 07:57:40.
PJM system load fell from 99,984 MW to 96,205 MW. Reporting Area Control Error hit plus 3,928 MW. Frequency reached 60.092 Hz. PJM recovered its Balancing Authority ACE Limit in nine minutes, against the 30 minutes that BAL-001-2 R2 allows. Operators tripped overvoltage capacitor banks to pull voltage back down. Customers returned their load to the grid within 30 minutes.
No utility equipment disconnected any of this load. The data centers tripped themselves, on their own protection. Operating Committee chair Emanuel Bernabeu put it plainly. “This was a normally cleared fault. They should not disconnect from the grid.” Dominion’s slides also record one data center back-feeding the fault, across three attempts.
PJM lists two earlier events of the same type, on July 10, 2024 and February 17, 2025. NERC investigated the 2024 case. It found roughly 1.5 GW of load leaving on customer-side protection. About 1.26 GW then stayed off the grid for hours, because a counting relay latched after three voltage disturbances inside one minute.
The Project 2026-02 drafting team at NERC works to a FERC deadline of December 31, 2026, under docket RD26-7-000. Phase 1 sets interconnection and modeling rules for what NERC calls a Computational Load Site. PJM’s slides put the registration threshold at 50 MW or more, connected at 100 kV or above. That criterion is still in draft.
Phase 1 does not set a minimum voltage or frequency ride-through. The NERC Large Load Working Group posted a whitepaper on that floor for comment on July 30, and expects it inside a standard in 2027.
Why It Matters
Dominion’s slides show why the sequencing hurts. Data center load already equals 27 percent of the previous peak load record in that zone. The Loudoun County pocket tops out at 4,247.6 MW. Prince William tops out at 2,186.2 MW. These are not diverse loads spread across a system. They answer the same fault in the same second.
PJM executive director David Souder framed the problem directly. “You really need these requirements in place before the load comes on the system.” One rule addresses what happened on July 22. That rule is the minimum ride-through floor, and it sits in the 2027 phase. Every interconnection agreement signed before then locks in whatever protection settings the customer picked.
Critical Perspective
Bernabeu’s line puts the fault on the customer. The record in the same deck complicates that. No utility relay misoperated, no breaker stuck, and the data centers did what their protection was built to do, which is protect the load. Calling that behavior too sensitive reframes a design choice as misconduct. It also skips the setting that would have been correct. Phase 1 declines to name it.
The hardware that failed belonged to Dominion. An improperly torqued bolt overheated a compression sleeve on a dead-end structure. That root cause analysis is not due until September 2026, well after the standards conversation takes its shape. The remedy under discussion lands on the loads, not on the asset owner.
Judge the event by the operator’s own metric and it reads as a pass. PJM cleared its BAAL excursion in nine minutes against a 30-minute standard, and reported no reliability impact. The case for new interconnection requirements rests on an event that satisfied the rule it is measured by.
Two gaps sit inside the draft. The 50 MW and 100 kV registration threshold leaves smaller sites outside the registry. So does splitting a site across two points of connection. And one data center back-fed the fault three times. That is generation behavior from a facility registered as load, and Phase 1 does not reach it.
Sources
- PJM Operating Committee, July 22, 2026 Dominion Load Transfer Event (presentation)
- PJM Inside Lines, PJM, Dominion Review Large Load Transfer Event
- Utility Dive, PJM eyes data center, crypto reliability requirements
- POWER Magazine, 3 GW Off PJM, Again
- Data Center Knowledge, Fault in Data Center Alley Triggered 3 GW Load Drop
