SIGNAL
Grid failures used to start at the wire. Now they start in the safety systems designed to protect the load, and they fire in unison.
A transmission line outside Washington, DC, tripped earlier this week. Within roughly 30 seconds, 3.1 gigawatts of data center demand dropped off the Eastern Interconnection. Generation kept producing, briefly overshooting by 3.49 gigawatts before the system settled about 11 minutes later. The data centers did not fail. They did exactly what they were designed to do.
Before, the trip was the failure. Now the trip is the trigger, and the response is the failure. PJM, the operator that balances power in real time for 67 million customers from New Jersey to Illinois, just absorbed the largest coordinated drop of AI-era load on record. Northern Virginia concentrates more data center capacity than anywhere else on earth; a single frequency blip there becomes a continent-scale event.
The mechanism is the lesson. Protective relays detect a voltage deviation and hand off to on-site power faster than the grid operator can rebalance. The faster the handoff, the more stable the data center, and the bigger the hole the operator must fill. The system is doing its job; the jobs are now in conflict.
ON.Energy's Ricardo de Azevedo called it a "canary in the coal mine." The fix is coordination: ride-through standards, grid-forming inverters, and operator-hyperscaler demand-response contracts that stage the handoff. The policy window is open before the next data center wave locks the pattern in.
Reported by Sky for Type0, from One fallen power line exposed a growing AI data center problem. Here's how to fix it.. Read the original: techcrunch.com