Massive AI data center blackout test reveals grid vulnerability as power demand doubles

A single downed power line near Washington, DC this week exposed a hidden risk in how America's electrical grid operates. In seconds, more than 3 gigawatts of data centers yanked themselves offline at the same time, sending shockwaves across the PJM grid—the nation's largest electric utility serving 67 million people—and causing lights to flicker from Northern Virginia to Chicago.

The grid survived, but the incident was a stark reminder that artificial intelligence infrastructure is growing faster than the power systems supporting it can adapt.

The Cascade Effect#

When the power line failed, the downed connection triggered data centers scattered across Northern Virginia to sense the voltage dip. Each facility made the same rational choice in seconds: switch to backup generators and disconnect from the grid. But that collective decision spiraled into a larger problem.

As 3.1 gigawatts of power demand suddenly vanished, the grid lurched in the opposite direction. Extra electricity—3.49 gigawatts at peak—flooded the system with nowhere to go, destabilizing voltages across an eight-state region. The recovery took over 11 minutes instead of the usual few seconds.

"It's the canary in the coal mine," said Ricardo de Azevedo, CTO of ON.Energy, a startup developing power solutions for data centers. These kinds of disruptions are "happening more and more."

A Problem That's Getting Bigger#

This event dwarfed a similar incident just two years ago, when 60 data centers disconnected and pulled 1.5 gigawatts from the grid. Data centers now consume roughly 6% of PJM's total load; by 2040, they're expected to consume 24%. Without intervention, synchronized disconnections will become catastrophic.

"Most data centers make decisions in a split second, and those that disconnected this week appear to be no different," said Ali Zain Banatwala, senior market models specialist at the Independent Electricity System Operator. The problem: they all make the same decision at once.

Building Smarter, Not Just Bigger#

Two paths exist to solve this. One approach is coordination: grid operators could require data centers to disconnect and reconnect sequentially rather than in unison, giving operators time to manage the load shifts. But that requires both technology and buy-in from multiple utilities.

The second approach is redesign. ON.Energy has built what amounts to a giant battery backup for entire data center campuses—not just servers, but cooling systems and all supporting equipment. To the grid, it looks like one stable consumer rather than a volatile load. The system absorbs power fluctuations within milliseconds, charging batteries during surges and discharging during dips, allowing AI workloads to ramp without destabilizing the broader network.