Power Grid Meltdown: Fallen Line Exposes AI Data Center's Achilles' Heel

A recent power grid disruption outside Washington, D.C., highlighted a growing vulnerability stemming from the rapid expansion of data centers. When a power line went down, the PJM grid, which manages electricity for 67 million customers from New Jersey to Illinois, experienced an unusual and prolonged recovery. Instead of the typical few seconds, it took over ten minutes for stability to return, during which voltage across the network spiked significantly, causing lights to flicker from Northern Virginia all the way to Chicago.

The root cause was the nearly simultaneous disconnection of more than 3 gigawatts of data centers from the grid. According to PJM data, approximately 3.1 gigawatts of load vanished within about 30 seconds as these facilities switched to backup power. This initial drop was then followed by additional load disconnections, resulting in an excess of up to 3.49 gigawatts of electricity on the PJM grid. Such massive and sudden load fluctuations are highly problematic for electrical grids, which require a near-perfect balance between supply and demand to operate efficiently. Even a few percent imbalance can lead to severe voltage sags or spikes, triggering failsafes within the grid infrastructure or individual facilities, causing further disconnections.

Northern Virginia, home to the world's highest concentration of data centers and located within PJM's territory, is increasingly becoming a critical point for these grid challenges. Experts view this incident as a "canary in the coal mine," signaling a trend that is expected to intensify. Indeed, this event echoes a similar disruption two years prior, also on the PJM grid, and was twice as large as an incident in 2024 where 60 data centers simultaneously pulled 1.5 gigawatts of load. The urgency to address this issue is underscored by projections that data centers, which accounted for roughly 6% of PJM's total load in 2024, are expected to consume 24% by 2040.

The core problem lies in how current data centers are designed to react to power disturbances. When a voltage dip occurs, they make split-second decisions to disconnect and switch to backup power to protect their sensitive equipment. While logical for individual facilities, when numerous data centers act simultaneously, they create a cascading effect where a small initial supply issue transforms into a much larger demand deficit, further destabilizing the grid.

Addressing this requires innovative solutions and a shift in data center operational philosophy. One proposed approach involves developing mechanisms for data centers located near each other to either disconnect or reconnect sequentially, providing grid operators with more predictable load management. Another, more proactive solution, is for data centers to be built with the capacity to absorb disruptions rather than turn their backs on them.

Startup ON.Energy is at the forefront of this latter approach, developing a sophisticated uninterruptible power supply (UPS) system designed for entire data center campuses. This system integrates a bank of batteries with advanced power conversion equipment, effectively shielding the grid from the individual peaks and valleys of a data center's power consumption. The grid perceives the data center as a consistent, well-behaved load. Crucially, ON.Energy's system allows data centers to ride through power fluctuations: it can use excess grid power to charge its batteries and dispatch stored power to servers when grid supply dips, all while following the grid's lead within milliseconds to prevent sags or surges. The company is currently installing a total of 3 gigawatts worth of its systems across four different data center campuses.

Grid managers are also taking notice and implementing new requirements. For example, ERCOT, another major grid operator, will soon mandate that large loads, including data centers, must be able to "ride through" disruptions rather than immediately disconnecting. With the scale and frequency of these events increasing, and data center demand projected to surge, proactive measures and technological advancements are critical to preventing more severe grid stability issues in the future.