Power line failure reveals AI data center grid risks and solutions

{ "title": "Data Center Cascade After Power Line Failure Exposes Grid Risk as AI Demand Surges", "body": "A routine power line failure outside Washington, DC this week triggered a cascade of nearly simultaneous disconnections by data centers, causing a voltage spike that flickered lights from Northern Virginia to Chicago and exposed the growing risk that AI-driven computing poses to the stability of the US electrical grid.\n\nThe event, which took over 10 minutes to resolve—compared to the usual recovery time of seconds—saw data centers stop drawing over 3 gigawatts of power almost at once. The resulting surge of extra electricity on the PJM Interconnection grid, the largest grid operator in the US, peaked at 3.49 gigawatts before the system stabilized. No blackout occurred, but the incident is the second such event in two years and experts warn it could foreshadow much larger disruptions as data center demand skyrockets.\n\n## What Happened When the Line Went Down\n\nPJM Interconnection serves 67 million customers across a region stretching from New Jersey to Illinois. Northern Virginia, within PJM’s footprint, has the highest concentration of data centers in the world. When a power line failed outside Washington, DC this week, the grid experienced a voltage fluctuation. Most data centers are programmed to sense such fluctuations and switch to backup power within a split second, and they all disconnected within seconds of each other.\n\nAbout 3.1 gigawatts of load vanished in about 30 seconds. The grid recovered somewhat, but then additional loads dropped off. At its peak, PJM had an extra 3.49 gigawatts of electricity that had nowhere to go. It took another 11 minutes before the grid stabilized. The disconnected data centers represented around 3% of total PJM demand at the time.\n\nThe electrical grid requires a near-perfect balance of supply and demand. Small fluctuations are tolerated, but large ones trigger fail-safes. Ricardo de Azevedo, CTO at ON.Energy, a startup developing uninterruptible power supply systems for data centers, described the event as a warning sign. "It's the canary in the coal mine," he said, adding that events involving large loads are "happening more and more."\n\n## A Growing Pattern of Disruptions\n\nThis week’s mass disconnection was twice as large as a similar event in 2024, when 60 data centers simultaneously disconnected, pulling 1.5 gigawatts from the PJM grid. That event also occurred on the same grid, and the pattern is clear: as data centers proliferate, their collective behavior during grid disturbances is becoming a stability risk.\n\nAccording to Synapse Energy Economics, a consultancy, data centers accounted for about 6% of PJM load in 2024. By 2040, they are expected to make up 24% of PJM load. A few percent load change may not sound like much, but the grid needs near-perfect balance. The problem is compounded by the fact that data centers are often clustered in the same regions, meaning they all react to the same grid signals at the same time.\n\nAli Zain Banatwala, a senior market models specialist at the Independent Electricity System Operator, which manages Ontario’s grid, highlighted the need for coordination. "We need to figure a way for these loads that are located next to each other to sequentially either disconnect or reconnect," he said.\n\n## The Technical Challenge and Proposed Solutions\n\nThe chain reaction is straightforward: a power line fails, causing a voltage fluctuation. Data centers sense the fluctuation and switch to backup power, removing their load from the grid. This sudden drop in demand creates a supply surge, which causes voltage spikes and light flickering across the region. The grid’s control systems then work to rebalance, but the process can take minutes.\n\nON.Energy is developing a system that could prevent this cascade. The company’s uninterruptible power supply is designed for entire data center campuses, covering servers, chillers, and other equipment. The system hides the data center behind a battery bank with power conversion equipment, so the grid sees a consistent load. It allows data centers to ramp computing workloads up or down without bothering the grid, and it can absorb power fluctuations: charging batteries on extra power, dispatching power on dips, and following grid commands within milliseconds.\n\nON.Energy is currently installing a total of 3 gigawatts of these systems at four data center campuses. De Azevedo said the technology allows data centers to ride through disruptions rather than disconnecting, which would help stabilize the grid. The cost of such systems can be significant, with a single installation for a large campus running around $330 per kilowatt, meaning a 100-megawatt facility could face a $33 million investment. ON.Energy was founded in 2018 and has focused on grid-scale battery solutions since its inception.\n\n## Grid Managers Are Waking Up\n\nGrid managers have woken up to the problem. ERCOT, the grid manager for Texas, is planning to require large loads like data centers to "ride through" disruptions, meaning they must stay connected and manage their power draw rather than dropping off instantly. This approach would prevent the kind of sudden load loss that caused this week’s voltage spike.\n\nThe Independent Electricity System Operator is also considering rules that would require sequential disconnection and reconnection of large loads. Banatwala’s comment about needing a way for loads located next to each other to sequentially disconnect or reconnect reflects a broader recognition that the current behavior of data centers is incompatible with grid stability.\n\nIf the problem is not addressed soon, things could get a lot worse. The data center share of PJM load is projected to grow from 6% in 2024 to 24% by 2040, meaning the potential for simultaneous disconnections will multiply. A similar event two years ago was smaller; this week’s was larger. The trend is clear.\n\n## The Broader Implications for AI Infrastructure\n\nThe event demonstrates the effect data centers can have on the grid, and experts believe such incidents will become more frequent. Northern Virginia’s data center cluster is the world’s largest, and its growth shows no signs of slowing. The AI boom is driving demand for computing power that requires massive, continuous electricity consumption.\n\nThe current approach—where data centers prioritize their own uptime by disconnecting from the grid at the first sign of trouble—works for the data centers but not for the grid. As de Azevedo put it, the event could foreshadow larger events if data centers are not built to handle disruptions elegantly. The solution, he argues, is to make data centers part of the grid’s solution rather than a source of instability.\n\nON.Energy’s system is one example of how that might work. By using batteries and power conversion equipment to present a consistent load, data centers can absorb power fluctuations rather than disconnect. This allows them to stay online while helping the grid maintain balance. The cost of such systems, at around $100 per kilowatt for smaller installations, can vary widely depending on scale and configuration.\n\n## Related on Neura Market\n\n- PJM Interconnection Market Overview\n- Data Center Energy Infrastructure Analysis\n- Grid Stability and Renewable Integration" }