A single failed power line outside Washington, D.C., this week led to a grid disturbance that took more than 10 minutes to stabilize, after over 3 gigawatts of data center load abruptly disconnected. The event caused voltage spikes from Northern Virginia to Chicago and made lights flicker across the region, revealing how clustered large-scale computing facilities can amplify minor disruptions.
The incident began when the downed line triggered data centers to switch to backup power almost simultaneously. According to grid operator PJM, about 3.1 GW of demand vanished within 30 seconds. At its peak, the grid carried 3.49 GW of excess electricity, representing roughly 3% of total PJM load at the time, and took 11 minutes to return to normal.
Northern Virginia hosts the world’s densest concentration of data centers. These facilities rely on sensitive protective relays that disconnect at the first sign of voltage sags or surges. In this case, the initial loss of supply caused a voltage dip; data centers reacted in split seconds, paradoxically creating a massive oversupply that further destabilized the grid.
“It’s the canary in the coal mine,” Ricardo de Azevedo, CTO at ON.Energy, told TechCrunch, adding that such events are “happening more and more.” Ali Zain Banatwala, senior market models specialist at the Independent Electricity System Operator, noted the need for a more orderly disconnection and reconnection sequence, which would allow grid managers to develop better contingency plans.
The episode echoes a 2024 event when 60 data centers in the same region dropped 1.5 GW of load, then about 6% of PJM’s demand. By 2040, data centers are projected to account for 24% of PJM’s load, making the current vulnerability a pressing concern.
One startup, ON.Energy, has developed a campus-wide uninterruptible power supply that hides entire data centers behind a bank of batteries and sophisticated power electronics. The system allows facilities to absorb voltage fluctuations without disconnecting from the grid, while also enabling smooth ramping of AI training workloads. ON.Energy is currently installing 3 GW of its systems across four data center campuses. Grid operators are taking notice: ERCOT, for example, plans to require large loads to “ride through” disruptions rather than disconnecting.
Without such measures, the risk will grow. The latest disconnection was twice as large as the 2024 occurrence. As AI and cloud services continue to expand, the grid’s ability to handle clustered, instantaneous load changes will become critical.
See an error? Read our corrections policy or email [email protected].
