Key facts
- FERC ordered NERC to create mandatory reliability standards for AI data centers by year-end.
- US data center IT load is projected to reach 150 gigawatts by 2028.
- Data centers can cause grid instability due to their rapid load fluctuations.
- Wildfire liability poses an existential risk to some utilities.
- AI is being used to reduce load shedding during extreme weather events by up to 35%.
US utilities face a growing challenge in managing the strain of AI data centers on the power grid, coupled with the significant financial and operational risks posed by wildfires. The Federal Energy Regulatory Commission (FERC) has set a December 31 deadline for the North American Electric Reliability Corporation (NERC) to establish mandatory reliability standards for AI data centers, which are increasingly large and volatile power consumers. Gartner forecasts that power shortages could limit 40% of existing AI data centers by 2027, as US data center IT load is projected to nearly double to 150 gigawatts by 2028.
Oracle's Tom Eyford noted that the impact of a data center on the grid can be comparable to losing a large power plant, with the added concern that data centers can rapidly disconnect, forcing grid operators to match generation to load in real time to avoid destabilization. Arun Nimmala, also from Oracle, suggests utilities should view data centers as active grid participants rather than passive customers.
Wildfires present a different, yet equally critical, challenge. Unlike other extreme weather events, wildfires carry significant liability for utilities, potentially leading to billions in judgments and even bankruptcy. Recent legal developments, such as an Oregon appeals court ruling requiring fire-by-fire causation analysis and a South Dakota law barring strict liability claims, highlight the evolving legal landscape. California is exploring a 'fast pay' proposal to expedite victim payouts while limiting future claims.
Utilities are turning to AI and advanced technologies to mitigate these risks. By integrating weather forecasts, LiDAR data, asset age, and historical outage patterns, utilities can generate granular risk scores for equipment. Nimmala stated that deep learning frameworks have demonstrated up to a 35% reduction in load shedding during extreme weather events, with the necessary building blocks like advanced distribution management systems already in place.