A power line failed outside Washington, DC, earlier this week. Typically, the grid requires just a handful of seconds to recover from such incidents. However, this particular situation took over 10 minutes as more than 3 gigawatts of data centers ceased drawing electricity almost at once.

This incident triggered a voltage spike across the PJM grid, reaching from Northern Virginia to Chicago, according to insights gathered by Ting Labs, a startup that operates an IoT sensor network installed in residential sockets.

While the event didn’t lead to a blackout, it did cause lights in the area to flicker. This occurrence highlighted the impact that data centers can exert on the grid — an outcome that experts anticipate will become more common.

Northern Virginia, a part of PJM’s jurisdiction, hosts the highest density of data centers globally.

“It’s a warning signal,” remarked Ricardo de Azevedo, CTO at ON.Energy, to TechCrunch. He noted that such occurrences involving significant loads like data centers are “becoming increasingly frequent.”

The event mirrors one that transpired two years ago on PJM’s grid, which could indicate larger issues if data centers aren’t equipped to manage power supply interruptions effectively. The PJM Interconnection oversees grids stretching from New Jersey to Illinois, servicing 67 million customers, making it the largest grid operator in the U.S.

During this week’s power line failure, data centers were prompted to switch to backup power, resulting in approximately 3.1 gigawatts of load disappearing in roughly 30 seconds, based on PJM data. The grid showed signs of recovery, but shortly thereafter, additional loads dropped off. At its highest, PJM’s grid had an additional 3.49 gigawatts of electricity. It took an additional 11 minutes for stabilization. The disconnected data centers represented about 3% of total demand on PJM at that time, as reported by Reuters.

While a few percent may not seem significant, the electrical grid must operate in a state of nearly perfect equilibrium, with supply and demand closely aligned. When they diverge, voltages may sag or spike. The grid and its connected devices can endure minor fluctuations, but excessive deviations can trigger failsafes within the grid or individual facilities, leading them to disconnect. 

When data centers in Northern Virginia detected the fluctuation due to the fallen power line, they switched to backup power, reducing their load from the grid. As more data centers switched, the load removal increased. What started as a relatively minor reduction in supply resulted in a more substantial decline in demand, causing supply to surge and lights to flicker. 

Most data centers respond in an instant, and those that disconnected this week appeared to follow suit. Ali Zain Banatwala, senior market models specialist at the Independent Electricity System Operator, informed TechCrunch that as soon as they experienced the voltage drop, they all made the decision to disconnect within moments of one another. 

“We need to establish a method for these loads, which are situated next to each other, to disconnect or reconnect in a sequential manner,” he stated. A more organized approach would enable grid operators to develop stronger procedures in advance.

Alternatively, data centers could be designed to withstand disruptions instead of disengaging from them. One startup, ON.Energy, is working on a solution to assist data centers — and the grid — in weathering events similar to the one that occurred this week. 

The company has created an uninterruptible power supply for an entire data center campus, covering not just servers but also chillers and other machinery. Essentially, the company situates the data center behind a bank of batteries linked to advanced power conversion equipment. All the grid “perceives” is one consistent, smoothly operating load as opposed to the fluctuations from each segment of the data center. ON.Energy’s system enables data centers to adjust computing workloads, including AI training, without impacting the grid. 

More crucially, it means that data centers can absorb power fluctuations from the grid. Instead of disconnecting, ON.Energy’s system can utilize any surplus power to recharge its batteries, and if power flow decreases, the system can allocate power to servers. Additionally, it can adapt to changes in the grid within milliseconds, averting sags or surges like those that contributed to this week’s issue for PJM.

ON.Energy is presently installing a cumulative 3 gigawatts worth of its systems at four different data center locations, de Azevedo mentioned.

Grid managers have also begun to acknowledge the issue.

For instance, ERCOT will mandate that large loads like data centers “ride through” disruptions, as noted by de Azevedo.

However, time is of the essence. The mass disconnection that occurred this week was double the size of a comparable event in 2024, when 60 data centers disconnected simultaneously, removing 1.5 gigawatts of load from the grid. At that time, data centers constituted roughly 6% of PJM’s load, as reported by Synapse Energy Economics. By 2040, they are projected to account for 24%. If the situation isn’t proactively addressed, it could deteriorate significantly.