Data Center Power Risks Hit US Grid

A single fallen power line near Washington, DC, recently triggered an unexpected reaction across the regional power network. While the grid is typically designed to absorb such minor incidents within seconds, this event caused significant instability that lasted for more than 10 minutes, revealing the outsized influence of concentrated AI infrastructure on regional power stability.

Synchronized Load Shedding Events

When the initial fault occurred, automated systems within local data centers responded by simultaneously cutting their connection to the grid and shifting to backup power. According to data from the PJM data, approximately 3.1 gigawatts of electrical load vanished in roughly 30 seconds. The grid’s inability to immediately stabilize led to a peak excess of 3.49 gigawatts of power, resulting in flickering lights across the region.

• 3 gigawatts of data centers stopped drawing power almost simultaneously.
• 3.49 gigawatts of extra electricity hit the PJM grid at peak imbalance.
• 3% of total demand on PJM was represented by the disconnected data centers.
• 1.5 gigawatts of load was pulled from the grid during a similar 2024 event.
• 24% of PJM’s load is expected to be comprised of data centers by 2040.

Grid Stability and Automation

The core issue lies in how these facilities are programmed to respond to voltage fluctuations. When the grid experiences a minor dip, current safety protocols trigger internal failsafes that force a rapid disconnection, effectively removing large chunks of demand from the system. This creates a secondary surge in supply, which can perpetuate instability.

“We need to figure a way for these loads that are located next to each other to sequentially either disconnect or reconnect,”

— Ali Zain Banatwala, senior market models specialist at the Independent Electricity System Operator

Potential Technical Mitigations

Industry experts suggest that shifting from reactive disconnections to “ride-through” capabilities could mitigate these cascading effects. Firms like ON.Energy are developing uninterruptible power supply systems that act as a buffer, allowing facilities to absorb grid fluctuations by either charging batteries or dispatching power back into the grid in milliseconds. This approach aims to present the grid with a consistent, manageable load rather than unpredictable spikes and valleys.

Implications for Infrastructure

The frequency of these events appears to be increasing as the density of AI compute facilities grows, particularly in regions like Northern Virginia. For grid operators, the challenge is no longer just about meeting base demand, but managing the volatile behavior of large, automated consumers. If industry standards do not evolve to require more orderly load management, regional grids may face progressively larger disruptions as the total power draw from data centers continues to climb toward projected 2040 levels.