Fallen power line exposes fragility of AI data centers’ grid interactions

A power line failure in Northern Virginia this week caused more than 3 gigawatts of data centers to disconnect from the grid within about 30 seconds, according to PJM Interconnection data. The sudden loss of load led to a voltage spike across PJM’s territory, which stretches from Northern Virginia to Chicago, and caused lights to flicker across the region.

The incident highlighted how data centers’ near-instantaneous decisions to switch to backup power can destabilize regional electricity networks. Experts note that such events are becoming more frequent as data centers proliferate. Northern Virginia hosts the world’s highest concentration of data centers, making it a bellwether for similar challenges elsewhere.

When the voltage dip reached data centers, many disconnected within seconds of each other, removing large blocks of demand from the grid. The disconnections represented roughly 3% of total demand on PJM at the time, a share that can trigger protective mechanisms if fluctuations grow too large.

The problem is expected to intensify: data centers accounted for about 6% of PJM’s load in 2024 and are projected to reach 24% by 2040, according to Synapse Energy Economics. A similar event in 2024 involved 60 data centers simultaneously disconnecting, pulling 1.5 gigawatts of load from the grid.

Startups and grid operators are exploring solutions. ON.Energy, a company developing uninterruptible power systems for data centers, has installed systems totaling 3 gigawatts across four campuses. Its approach uses large battery arrays and power conversion equipment to present a consistent load to the grid, allowing data centers to absorb fluctuations rather than disconnecting.

ON.Energy’s system can follow grid signals within milliseconds, charging batteries when excess power is available and discharging to support servers during dips, thereby smoothing grid interactions. Grid operators are also taking notice: ERCOT, for example, plans to require large loads like data centers to ‘ride through’ disruptions, avoiding sudden disconnections.

Experts argue that more orderly disconnection and reconnection processes, or grid-aware infrastructure, could prevent future incidents. Without changes, the growing scale of AI data centers risks more frequent and severe grid disturbances as their collective impact on demand grows.