Can The Power Grid Handle AI And Wildfires At The Same Time?

Federal regulators have told the North American Electric Reliability Corporation to create mandatory reliability standards for AI data centers, setting a Dec. 31 deadline after the Federal Energy Regulatory Commission issued the order on July 16. The directive reflects rapid growth in data-center electricity demand: Bloom Energy estimates U.S. data center IT load at roughly 80 gigawatts today, rising toward 150 gigawatts by 2028, and Gartner warns that supply shortages could limit operations at a large share of AI facilities by 2027.

Industry specialists say the concern is not only the size of these loads but how abruptly they can change. Tom Eyford, Oracle’s global industry specialist for utility operations solutions, likens the effect of a major data center to losing a large power plant: a campus can draw hundreds of megawatts one moment and drop off the grid the next, forcing operators to rebalance generation and load in real time to avoid instability. Arun Nimmala, Oracle’s global head of grid operational technology products and services, argues utilities should treat data centers as active grid participants rather than ordinary customers — a reclassification FERC’s order will accelerate.

Wildfires add a distinct and legally fraught complication. Courts and state legislatures have been moving in different directions: an Oregon appeals court in April tossed a $1 billion verdict against PacifiCorp, citing a flawed jury instruction; South Dakota in March passed a law that bars strict liability claims against utilities in wildfire suits; and California’s governor has proposed a “fast pay” plan to speed victim payouts while limiting subsequent litigation. Eyford says liability is what sets wildfires apart from other extreme weather events and has driven outcomes as severe as bankruptcies for some utilities.

Utilities are turning to AI and data fusion to reduce wildfire risk and make operations more resilient. Combining weather forecasts, LiDAR-based vegetation mapping, asset age and outage histories can produce equipment-level risk scores, and Nimmala said deep-learning approaches have been associated with up to a 35% reduction in load shedding during extreme weather. Many of the necessary tools — advanced distribution management systems, fault-location and restoration systems, metering, and virtual power plant programs — are already being rolled out, but whether utilities can assemble and scale these capabilities quickly enough remains uncertain as both AI load growth and wildfire risk continue to accelerate.