1,500 MW in 82 seconds: what the July 2024 data center load loss teaches

At about 7:00 p.m. Eastern on July 10, 2024, a lightning arrester failed on a 230 kV transmission line in the Eastern Interconnection. The line's auto-reclosing was set for three attempts, staggered at each end, so a single permanent fault became six successive faults in 82 seconds before the line locked out.

Protection did its job. Every fault cleared normally, in 42 to 66 milliseconds, with voltages in the affected area dipping to 0.25 to 0.40 per unit. And yet roughly 1,500 MW of voltage-sensitive load, largely data centers, disconnected at the same time. The grid had never seen a simultaneous load loss of that size caused by a normally cleared fault.

The cause was inside the fence

NERC's incident review found the load loss was customer initiated. Utility equipment did not trip the facilities. Their own power supply logic did: transfer schemes moved load to backup power during the dips, and a rule that counts disturbances within a short window kept them there once the count was reached. A second instance in the same review involved facilities without backup supply simply dropping offline.

To each facility, that behavior is sensible protection for its servers. To the grid, sixty or so facilities making the same decision at the same moment looks like a 1,500 MW generator in reverse: frequency rises, voltage rises, and operators have to act.

Why the planning models missed it

Planning models for large loads are typically built from design data. A UPS datasheet tells you a nominal transfer threshold. It does not tell you how the installed settings, the disturbance counter, the lockout timer and the transfer back interact during a reclose sequence. Those are exactly the parameters that decided the outcome on July 10.

That is the gap the NERC CLO standards and utility large load requirements are now trying to close, through ride-through data, disturbance recording and model verification under CLO-001.

What operators should measure

• Voltage and current at the point of interconnection during every disturbance, at a resolution that captures transfer and recovery.
• UPS and transfer events with timestamps that line up with the grid record.
• Recovery behavior: when and how load returns, and in what steps.
• Settings changes, so the model is updated when the behavior changes.

Each recorded disturbance is an experiment the grid runs once. Facilities that keep the record can prove how they behave. Facilities that do not will have their behavior assumed for them.

Source

NERC, Incident Review: Considering Simultaneous Voltage-Sensitive Load Reductions.