When a single power line went down outside Washington, D.C., this week, the regional electricity grid would normally have recovered within seconds. Instead, it took more than ten minutes to stabilize — because more than 3 gigawatts of data centers stopped drawing power nearly simultaneously, sending voltage surging across a network that stretches from Northern Virginia to Chicago.
The incident, first detailed by TechCrunch and corroborated by PJM Interconnection data, has laid bare a growing fault line in the AI boom: the massive data centers powering artificial intelligence can destabilize the very grid they depend on. For ongoing AI infrastructure coverage and analysis of how the buildout is straining power systems, the details below explain why a routine outage became a regional event.
What Happened on the PJM Grid
A transmission line failure triggered an automatic response among the hyperscale data centers clustered in Northern Virginia, home to the highest concentration of data centers in the world. Sensing the voltage dip, the facilities switched to backup power, and roughly 3.1 gigawatts of load vanished from the grid in about 30 seconds, according to PJM data reported by Reuters.
The grid partially recovered, but a short time later additional loads dropped off. At its peak, PJM's network was carrying an extra 3.49 gigawatts of electricity with nowhere for it to go. It took another 11 minutes before operators brought the system back into balance. The disconnected data centers represented around 3% of total demand on PJM at the time.
While the event did not cause a blackout, it caused lights across the region to flicker. Data collected by Ting Labs, a startup that runs an Internet-of-Things sensor network installed in people's electrical sockets, showed voltage spikes propagating across the territory. The PJM Interconnection manages grids from New Jersey to Illinois and serves 67 million customers, making it the largest grid operator in the United States.
A Canary in the Coal Mine
Energy experts warned that the event is a sign of things to come. "It's the canary in the coal mine," Ricardo de Azevedo, chief technology officer at ON.Energy, told TechCrunch. Events involving large, concentrated loads like data centers are "happening more and more," he added.
The episode echoes a similar disturbance that occurred two years ago on the same PJM grid, and it could foreshadow larger failures if data centers are not engineered to handle power disruptions more gracefully. The core problem is one of balance: the electrical grid must keep supply and demand matched almost perfectly. When they drift apart, voltages sag or spike. The grid and connected devices tolerate small fluctuations, but larger ones trip failsafes that force facilities offline.
When the downed power line caused the initial voltage dip, neighboring data centers all made the same split-second decision to disconnect from the grid and fall back on their own backup systems. As each facility dropped off, it removed more load, which in turn pushed supply even higher. Ali Zain Banatwala, a senior market models specialist at the Independent Electricity System Operator, told TechCrunch that the facilities "all decided to disconnect within a few seconds of each other."
Coordinated Disconnects and 'Ride-Through' Systems
The path forward, grid specialists argue, is to make data center behavior more predictable during disturbances rather than allowing hundreds of facilities to react independently at the same instant. "We need to figure a way for these loads that are located next to each other to sequentially either disconnect or reconnect," Banatwala said. A more orderly, staged process would let grid operators develop robust procedures in advance instead of scrambling to absorb sudden swings.
An alternative is to build data centers that can absorb disruptions rather than abandoning the grid the moment voltage wavers — a capability the industry calls "ride-through." ON.Energy is among the startups developing products to help data centers, and the grid itself, weather events like the one that struck this week. Rather than cutting ties instantly, ride-through systems would keep facilities connected long enough for grid operators to rebalance supply and demand.
Why AI Makes the Problem Worse
The AI buildout is accelerating precisely the conditions that produced this week's incident. Training and running large language models demands enormous, continuous power draw, and the facilities that deliver it are increasingly concentrated in a handful of regions where land, fiber, and electricity are cheap. Northern Virginia alone hosts thousands of megawatts of computing capacity, and PJM's queue of pending interconnection requests is dominated by data center developers.
That concentration means a single local fault can now cascade across a multi-state region. A few percentage points of total demand may sound modest, but when that load disappears in 30 seconds, the grid has no time to ramp down generation to match. The result is surplus power with nowhere to go — the mirror image of the shortages that cause rolling blackouts, but no less destabilizing.
Grid operators and regulators are now weighing whether hyperscale facilities should be treated differently from ordinary industrial customers, with stricter rules for how quickly they can disconnect and reconnect. As AI computing demand is projected to multiply over the coming years, the stakes of getting that framework right are rising fast.
Stay Ahead of AI
The collision between AI computing and the power grid is only beginning, and the next disruption could be larger. To read more AI news and track how operators, regulators, and startups respond, keep following AI Buzz Wire.
Read more AI news →
