One fallen power line exposed a growing AI data center problem; here’s how to fix it
A power line failure near Washington, D.C., triggered over 3 GW of data centers to disconnect simultaneously, causing voltage fluctuations across the U.S. eastern grid. Experts warn such events will become more frequent as AI workloads grow and propose a battery-based system to help data centers ride through disruptions.
A power line fell outside Washington, D.C., this week. Normally, the grid would recover in seconds, but this time it took more than 10 minutes because over 3 gigawatts (GW) of data centers stopped drawing power nearly simultaneously. The event caused a voltage spike across the PJM grid, stretching from Northern Virginia to Chicago. Although no blackout occurred, lights flickered across the region.
PJM Interconnection, the largest grid operator in the U.S. serving 67 million customers, saw about 3.1 GW of load vanish within 30 seconds, and later additional loads dropped off, peaking at 3.49 GW of excess electricity on the grid. The disconnected data centers represented about 3% of PJM’s total demand at the time. While a few percent may seem small, the grid requires near-perfect balance between supply and demand; otherwise, voltages can sag or spike.
Data centers in Northern Virginia, the world’s highest concentration of such facilities, sensed the voltage dip and switched to backup power, removing their load from the grid. This triggered a cascade: a small drop in supply turned into a larger drop in demand. According to Ali Zain Banatwala of the Independent Electricity System Operator, the data centers all decided to disconnect within seconds of each other, and a method is needed to have large loads reconnect or disconnect sequentially.
Startup ON.Energy has developed an uninterruptible power supply for an entire data center campus, hiding the facility behind a bank of batteries and power conversion equipment. The grid sees only a consistent, well-behaved load. The system allows data centers to ramp AI training and other workloads up and down without disturbing the grid and can absorb fluctuations by charging batteries from excess power or dispatching stored energy during dips. ON.Energy is currently installing 3 GW of its systems at four data center campuses.
Grid operators like ERCOT are now requiring large loads to “ride through” disruptions. This week’s incident was twice as large as a similar event in 2024, when 60 data centers disconnected simultaneously, pulling 1.5 GW off the grid. In 2024, data centers accounted for about 6% of PJM’s load; by 2040, they are expected to reach 24%. If the problem is not addressed soon, consequences could become much more severe.
