3 Gigawatts of Data Centers Dropped Off the PJM Grid After a Power Line Failed

A single transmission line failure outside Washington, D.C., on the morning of July 22 set off a cascade that pulled more than 3 gigawatts of data center load off the largest power grid in the United States, sending a voltage disturbance that was felt from Northern Virginia to Chicago. There was no blackout, but as TechCrunch reported, the incident is a stark example of how the AI data center boom can stress grid stability faster than operators adapt.

The event hit the PJM Interconnection, the grid operator whose territory stretches from New Jersey to Illinois and serves 67 million customers. Reuters reported that roughly 3 percent of the grid’s total demand vanished in seconds, and PJM registered both the sudden drop and a shift in grid frequency, the measure of how closely electricity supply and demand are matched. A downed line is normally a non-event that the grid corrects within seconds. This time it took more than 10 minutes to stabilize.

How a Routine Line Fault Became a Regional Event

When the line went out of service, data centers in Northern Virginia, home to the world’s largest concentration of them, sensed the resulting voltage dip and did exactly what they are built to do: they switched to backup power. That decision was made by the data centers’ own control systems rather than by any grid operator, Dominion Energy, the regional utility, said in a statement.

That self-protective reflex is the core problem. As each facility shed its load almost simultaneously, TechCrunch reported that about 3.1 gigawatts disappeared in roughly 30 seconds. The grid partially recovered, then a second wave of load fell away. At its peak, PJM was left carrying an extra 3.49 gigawatts with nowhere to send it, which pushed voltages up and made light bulbs flicker across the region before the system was brought back into balance.

The grid runs on a near-perfect match between supply and demand, so when a large block of demand disappears at once, supply surges and voltage spikes. Those swings can trip failsafes that force still more equipment offline. Bob Marshall, chief executive of Ting Labs, whose network of 1.4 million socket-level sensors recorded the disturbance across the eastern United States, called it “a massive event” and said customers reported flickering lights and noises from their air conditioners and refrigerators. Grid wobbles are usually corrected in milliseconds, he noted; this one lasted about 10 minutes.

“The Canary in the Coal Mine”

Experts describe the event as a preview rather than an anomaly. “It’s the canary in the coal mine,” Ricardo de Azevedo, chief technology officer at ON.Energy, told TechCrunch, adding that these large-load disruptions are “happening more and more.” The disturbance echoed a 2024 event on the same PJM grid, when roughly 60 data centers disconnected at once and pulled 1.5 gigawatts offline. This week’s drop was about twice as large.

The trend line is steep. Data centers accounted for about 6 percent of PJM’s load in 2024, according to Synapse Energy Economics, and that share keeps climbing as AI training and inference workloads expand. DataCenterDynamics noted that PJM is already staring at an unprecedented power shortfall over the coming decade driven by data center demand, a gap between new load and new generation that PJM has warned could pressure grid reliability later this decade.

Two Ways to Fix It

The proposed fixes fall into two categories: coordinate how data centers leave and rejoin the grid, and build them to ride through disturbances instead of fleeing them.

The coordination challenge is synchronization. “We need to figure a way for these loads that are located next to each other to sequentially either disconnect or reconnect,” said Ali Zain Banatwala, a senior market models specialist at the Independent Electricity System Operator, Ontario’s grid operator. Staggering those transitions would give operators room to plan procedures in advance rather than absorbing thousands of megawatts vanishing in a single breath.

On the hardware side, ON.Energy offers a campus-scale uninterruptible power supply that covers not just servers but chillers and other equipment. The system parks the entire data center behind a bank of batteries and power conversion gear, so the grid sees one smooth, well-behaved load instead of the peaks and valleys of individual components. Rather than disconnecting during a fluctuation, the batteries absorb excess power or dispatch it to servers, tracking the grid within milliseconds. The company said it is installing 3 gigawatts of the systems across four data center campuses.

Regulators are moving in the same direction. Federal and state authorities are studying rules to better manage sudden demand drops from data centers and crypto miners, and in Texas, grid operator ERCOT plans to require large loads to “ride through” disruptions rather than dropping offline, de Azevedo said.

A Near Miss With a Warning Attached

“Within minutes of the incident occurring, our system operations team was able to stabilize the situation and return to normal operating conditions,” Dominion Energy said. PJM, for its part, reported no impact to the grid’s overall reliability. Still, the episode showed how an ordinary equipment fault can now ripple into a regional voltage event, simply because so much synchronized, self-protective computing load now sits in one place. The warning signs, as de Azevedo put it, are already flashing.