When a single power line failed outside Washington, DC in late July, it exposed a growing crisis in how AI data centers interact with the electrical grid. The incident revealed that massive computing facilities can inadvertently destabilize power systems when they all react to disruptions at the same time. As AI infrastructure continues to expand, grid operators and data center builders are racing to prevent larger cascades.

What Happened When That Power Line Went Down?

On July 25, a routine power line failure near Washington triggered an unusual chain reaction. Normally, the electrical grid recovers from such events in seconds. This time, it took more than 10 minutes. The reason: data centers in Northern Virginia, home to the world's highest concentration of data centers, simultaneously switched to backup power and disconnected from the grid.

When the voltage dip hit, approximately 3.1 gigawatts of load vanished in about 30 seconds. To put that in perspective, that's roughly equivalent to the power consumption of 3 million homes. The sudden loss created an imbalance on the PJM Interconnection, which manages grids serving 67 million customers across New Jersey to Illinois. At its peak, the grid had an extra 3.49 gigawatts of electricity flooding through it, causing lights to flicker across the region from Northern Virginia to Chicago.

The disconnected data centers represented around 3 percent of total demand on PJM at the time. While that may sound modest, electrical grids operate in a state of near-perfect balance. Even small fluctuations can trigger voltage sags or surges that damage equipment or trigger automatic shutdowns.

Why Are Data Centers Becoming a Grid Problem?

The incident wasn't isolated. A similar event occurred two years earlier on the same PJM grid, but this time the problem was twice as severe. In 2024, about 60 data centers simultaneously disconnected, pulling 1.5 gigawatts from the grid. Back then, data centers accounted for roughly 6 percent of PJM's total load. Projections suggest that by 2040, they will consume 24 percent of the grid's capacity.

The core issue is that data centers make split-second decisions. When they sense a voltage fluctuation, they all switch to backup power within seconds of each other. This synchronized response, while protecting individual facilities, creates a cascading problem for the broader grid.

"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, senior market models specialist at the Independent Electricity System Operator.

Ali Zain Banatwala, Senior Market Models Specialist at the Independent Electricity System Operator

Ricardo de Azevedo, CTO at ON.Energy, described the July incident as "the canary in the coal mine." These sorts of events involving large loads like data centers are "happening more and more," he noted.

How Can Data Centers Stop Breaking the Grid?

Experts and engineers have identified several approaches to prevent future disruptions:

• Orderly Disconnection Protocols: Grid operators can develop procedures that stagger when data centers switch to backup power, preventing simultaneous load drops that destabilize voltage levels.
• Battery-Backed Power Systems: Data centers can be equipped with sophisticated uninterruptible power supplies that absorb grid fluctuations instead of disconnecting, allowing them to ride through disruptions smoothly.
• Dynamic Load Management: Systems that allow data centers to ramp computing workloads up and down without creating peaks and valleys that stress the grid.
• Regulatory Requirements: Grid operators like ERCOT are beginning to require large loads such as data centers to "ride through" disruptions rather than automatically disconnecting.

ON.Energy has developed a product that essentially hides an entire data center campus behind a bank of batteries and sophisticated power conversion equipment. From the grid's perspective, it sees one consistent, well-behaved load rather than the chaotic peaks and valleys from individual servers and cooling systems. The system can absorb power fluctuations within milliseconds, charging batteries when power surges and dispatching stored energy when supply dips.

"ON.Energy's system allows data centers to ramp computing workloads up and down, including AI training, without bothering the grid," explained Ricardo de Azevedo.

Ricardo de Azevedo, CTO at ON.Energy

ON.Energy is currently installing a total of 3 gigawatts' worth of its systems across four different data center campuses. Meanwhile, grid managers have begun waking up to the problem. ERCOT, which manages the Texas grid, is implementing requirements for large loads to ride through disruptions rather than automatically disconnecting.

Why Does This Matter for AI's Future?

The stakes are rising rapidly. Data centers already consume enormous amounts of power, and AI training and inference are among the most energy-intensive computing tasks. As companies race to build larger AI models and deploy them at scale, power consumption will only increase. If the grid cannot reliably support these facilities, it could become a bottleneck for AI development itself.

The July incident demonstrated that the problem isn't theoretical. It's happening now, and it's getting worse. The mass disconnection this week was twice as large as the 2024 event. Without intervention, similar incidents could cascade into larger blackouts as data center density increases.

The good news is that solutions exist. They require coordination between data center operators, equipment manufacturers, and grid managers. The clock is ticking, though. If the problem isn't addressed soon, the explosive growth of AI infrastructure could strain regional power systems to the breaking point.