AI data centers expose a new grid architecture challenge
A transmission-line fault in Ashburn, Virginia, on July 22, 2026, abruptly removed more than 3 gigawatts of data-center demand from the grid. A similar incident two years earlier followed the failure of a single surge arrester and disconnected roughly 60 Virginia facilities, representing 1,500 megawatts. The incidents highlighted a reliability risk caused not by insufficient generation, but by many large facilities responding to disturbances in the same way.
The article says AI campuses can change their electricity use by 70% in milliseconds and disconnect rapidly when upstream problems occur. Conventional data-center systems, built around low-voltage UPS units and protection settings for much smaller loads, may therefore transmit disturbances or shut down at the wrong moment.
ON.energy proposes moving power equipment to medium voltage, placing it in modular enclosures near substations, and routing all electricity through an inline system rather than relying mainly on bypassed batteries. The company says it tested such a system in early 2026 at the National Laboratory of the Rockies against AI-scale load changes and simulated grid faults, including a zero-voltage event. According to the article, the system met ERCOT ride-through requirements. The content was produced by ON.energy, not by the publication’s editorial staff.
Written by Sakhanda Wire AI from reporting by MIT Tech Review. Facts, quotes and full context are in the original article below.