A power line failure outside Washington, DC, turned into a larger grid lesson this week. The line itself was not the only issue: more than 3 gigawatts of data centers stopped drawing power almost at the same time, and the PJM grid took more than 10 minutes to settle.
The result was not a blackout. But voltage spiked across a wide area, lights flickered, and the event showed how AI data centers can affect a power system that depends on constant balance.
What happened on the PJM grid
Under normal conditions, the grid would have needed only a few seconds to recover from a downed power line. This time, the disruption spread because nearby data centers reacted quickly and in similar ways.
According to data collected by Ting Labs, voltage across the PJM grid spiked from Northern Virginia to Chicago. Ting Labs runs an IoT sensor network from people’s electrical sockets, giving it visibility into how voltage changes show up inside homes and buildings.
Northern Virginia matters because it sits inside PJM’s territory and is home to the highest concentration of data centers in the world. PJM Interconnection manages grids from New Jersey to Illinois and serves 67 million customers, making it the largest grid operator in the United States.
When the power line failed, data centers switched to backup power. PJM data showed that about 3.1 gigawatts of load disappeared in about 30 seconds. A short time later, more loads dropped away, and at the peak PJM’s grid had an extra 3.49 gigawatts of electricity on it.
The disconnected data centers represented around 3% of total demand on PJM at the time, according to Reuters. That may sound small, but the grid is not built to treat supply and demand as rough estimates. It needs them closely matched.
Why a few percent can matter
Electric grids operate best when electricity supply and electricity demand stay in near-perfect balance. If demand suddenly falls while supply remains in place, voltage can rise. If supply drops faster than demand, voltage can sag.
Small movements are expected and manageable. Larger movements can trigger protective systems, either inside the grid or inside individual facilities. Those systems are designed to prevent damage, but when many large facilities respond at once, the protective reaction can amplify the original disturbance.
That appears to be the core problem in this event. The failed line created a voltage dip. Data centers in Northern Virginia sensed the change and moved to backup power. As they left the grid, they removed more demand, which pushed the grid in the other direction.
In practical terms, a supply problem became a demand problem. The grid had extra electricity, voltage rose, and lights across the region flickered.
Ricardo de Azevedo, CTO at ON.Energy, described the event to TechCrunch as “the canary in the coal mine.” He said events involving large loads like data centers are “happening more and more.”
The coordination problem
Most data centers make power decisions extremely quickly. That is useful for keeping servers online, but it creates a new challenge when many facilities are located near one another and respond to the same signal in the same time window.
Ali Zain Banatwala, senior market models specialist at the Independent Electricity System Operator, told TechCrunch that when the voltage dip reached these facilities, they all appeared to disconnect within a few seconds of each other.
“We need to figure a way for these loads that are located next to each other to sequentially either disconnect or reconnect,” he said.
The point is not that data centers should ignore grid trouble. The point is that simultaneous action by large loads can be destabilizing. A more orderly sequence would give grid operators a clearer process to plan around.
This matters for reconnection as well as disconnection. If large loads leave the grid together, the grid must absorb the sudden excess. If they return together, the grid must handle a sudden jump in demand. Sequential behavior could make both sides of that cycle easier to manage.
How batteries could change the response
One proposed fix is to make data centers better at riding through disruptions instead of immediately stepping away from the grid. ON.Energy has been developing a system aimed at that problem.
The company has created an uninterruptible power supply for an entire data center campus. The system covers servers, chillers and other equipment. It uses batteries and power conversion equipment so the grid sees one steadier load instead of the ups and downs created by different parts of the data center.
That design could let data centers ramp computing workloads up and down, including AI training, without bothering the grid. It could also help during a disturbance. If extra power appears, the system can charge batteries. If power flow dips, it can send power to servers.
ON.Energy’s system can follow the grid’s lead within milliseconds, according to the source article. De Azevedo said the company is currently installing a total of 3 gigawatts worth of its systems at four different data center campuses.
Grid managers are also starting to respond. ERCOT, for example, is going to require large loads like data centers to “ride through” disruptions, de Azevedo said.
Why this is becoming urgent
This week’s mass disconnection was not an isolated warning. A similar event happened in 2024 on PJM’s grid, when 60 data centers simultaneously disconnected and pulled 1.5 gigawatts of load from the grid.
The latest event was twice as large. That matters because data centers are expected to become a much bigger part of PJM’s load. According to Synapse Energy Economics, data centers accounted for about 6% of PJM’s load in 2024. By 2040, they are expected to make up 24%.
The implication is straightforward: as AI data centers grow, the grid consequences of fast, uncoordinated disconnections can grow too. The fixes described in the source are not abstract. They are operational changes: better sequencing, ride-through requirements and battery-backed systems that make data centers look less volatile to the grid.
The Washington, DC, power line failure did not bring down the grid. But it showed that the relationship between data centers and power infrastructure is changing. Keeping servers online is no longer only a facility-level problem. It is becoming a regional grid problem as well.