A power line outage near Washington, DC, recently exposed a significant challenge for AI data centers and the electrical grid. When the power line fell, data centers in Northern Virginia, the world's largest concentration of such facilities, rapidly switched to backup power. This mass shift caused a sudden drop of over 3 gigawatts in electricity demand on the PJM Interconnection grid, triggering voltage spikes that affected a broad area from Northern Virginia to Chicago and caused regional lights to flicker. Although no blackout occurred, the event highlighted the fragile balance between supply and demand on the grid and the disruptive impact of concentrated data center power usage.

The PJM grid, serving 67 million customers across much of the U.S., experienced a surge in supply as data centers collectively disconnected within seconds, worsening the initial disturbance caused by the fallen power line. Experts point out that data centers acted simultaneously to protect their operations during the voltage dip, but this “all-at-once” response made matters worse by rapidly removing a large load from the grid. The incident echoes a similar event two years ago and signals mounting risks as data centers grow—from contributing 6% of PJM’s load in 2024 to a projected 24% by 2040.

Industry players are intensifying efforts to prevent these problems. One company, ON.Energy, is deploying advanced uninterruptible power supply systems across multiple data center campuses totaling 3 gigawatts. These systems use banks of batteries and power conversion technology to smooth out power demand and can absorb grid fluctuations by charging batteries or supplying power in milliseconds, thus avoiding sudden shutdowns. This approach creates a more stable load profile that benefits both data centers and grid operators.

Grid managers are also adapting to the evolving landscape. For example, ERCOT plans to mandate that large users like data centers “ride through” grid disruptions rather than disconnect instantaneously. However, time is pressing because data center energy demands are skyrocketing as AI workloads expand. Without coordinated solutions to sequentially manage power disconnections and reconnections or enhanced on-site energy systems, the grid faces increasing instability and risks greater disruptions as this sector continues to grow.