PJM is weighing reliability requirements for data centers after nearly 3,800 MW of Northern Virginia demand transferred to onsite generation during a grid disturbance, the second event of its kind in two years.

The July disturbance began with a mechanical failure on a 230 kV transmission line, according to Utility Dive. Rather than riding through the voltage dip, a large block of computing load disconnected from the grid and moved to its own generation. PJM said on 11 August that it is considering interconnection reliability requirements for data centers and crypto-mining facilities, and that these could cover voltage and frequency ride-through, how onsite generation behaves during a disturbance, and how a facility reconnects afterwards.

This is the second time, and the first one was investigated

The pattern is documented. On 10 July 2024, a lightning arrestor failed on a 230 kV line in the Eastern Interconnection, and NERC’s incident review records roughly 1,500 MW of load associated with data centers dropping simultaneously across 60 separate points and 25 substations. The auto-reclosing scheme was configured for three attempts at each end of the line, which produced six successive faults in 82 seconds. Voltage in the affected area fell to between 0.25 and 0.40 per unit.

NERC’s own conclusion in that review was that the grid had not historically experienced simultaneous load loss of that magnitude in response to a system fault. Two years later the same interconnection, the same voltage class of line and a similar failure produced a larger version of the same event.

What it means for operators

The distinction that matters is between protecting the facility and protecting the grid. Equipment that disconnects on a voltage dip is doing what it was specified to do. At 3.8 GW, thousands of those individually correct decisions become a single grid event, and that is the behaviour PJM is now proposing to write rules about.

Two things follow for anyone running load in a constrained interconnection. Ride-through settings on UPS and transfer equipment are about to stop being a purely internal choice, and a facility that cannot say what its load actually did during the last disturbance will struggle to answer for it. NERC’s 2024 review reconstructed the event from the grid side. Operators who can reconstruct it from their own side, with time-aligned voltage and current at the transfer point, will be in a considerably better position when the requirements arrive.

Sources: Utility Dive, Data Center Knowledge, NERC incident review