Future Tech & AI Wonders · Sam Patel · 25 July 2026

One fallen power line exposed a growing AI grid problem

One fallen power line exposed a growing AI grid problem

One fallen power line near Washington, D.C., triggered more than 3 gigawatts of Northern Virginia data centers to switch to backup power almost at once, spiking voltage across the PJM grid for over 10 minutes. The fix: sequential disconnects or ride-through systems so AI campuses absorb shocks instead of dumping load.

Key Takeaways

What happened after one fallen power line failed?

This week, a power line went down outside Washington, D.C. A typical grid blip clears in seconds. This one dragged on for more than 10 minutes because more than 3 gigawatts of data centers stopped drawing power nearly simultaneously.

According to TechCrunch, PJM data show about 3.1 gigawatts of load vanished in about 30 seconds as facilities flipped to backup power. Extra supply peaked at 3.49 gigawatts before the system stabilized after another 11 minutes.

Ting Labs sensors recorded a voltage spike stretching from Northern Virginia to Chicago. The disconnected campuses represented around 3% of PJM demand at the time, per Reuters reporting cited in the coverage. Northern Virginia hosts the world's densest data center cluster inside PJM, which serves 67 million customers from New Jersey to Illinois.

Why do AI data centers make a small fault worse?

Grids need near-perfect balance between supply and demand. When the line failed, Northern Virginia centers sensed the dip and switched to backup together. Removing that load turned a supply hiccup into a larger demand plunge, pushing voltage up and making lights flicker.

Ali Zain Banatwala of Ontario's Independent Electricity System Operator told TechCrunch the facilities decided within seconds of each other. Ricardo de Azevedo, CTO at ON.Energy, called the episode a "canary in the coal mine," saying large-load events are "happening more and more."

The mass disconnection was twice as large as a 2024 PJM incident that pulled 1.5 gigawatts when 60 data centers dropped off. Synapse Energy Economics put data centers at about 6% of PJM load then; they are expected to reach 24% by 2040. For more on the AI buildout, see our Future Tech & AI Wonders hub.

How can operators fix the AI data center problem?

Banatwala argued neighboring loads should disconnect or reconnect sequentially so operators can plan robust procedures. Another path is building campuses that ride through disruptions instead of abandoning the grid.

ON.Energy's campus-scale uninterruptible supply hides servers, chillers, and other gear behind batteries and power conversion. The grid sees one steady load; surplus charges batteries, and dips are met from storage within milliseconds. The startup is installing 3 gigawatts of systems across four campuses, de Azevedo said. ERCOT, he added, plans to require large loads like data centers to ride through disruptions.

Without those changes, experts warn flicker events could grow into deeper stability risks as AI demand climbs.

← Open in blast feed