Outage Oracle: Universities Build AI to Outsmart Storms

UConn and UAlbany launch a $550K AI initiative to forecast storm-driven grid outages before they hit

27 Mar 2026

Americans lost power 1.42 billion times in 2023, more than any other developed country. The usual response is reactive: trucks dispatched after the lines fall, crews scrambling in the dark while customers wait. A new academic initiative, announced in February, proposes a different approach. Rather than fixing outages, predict them.

Researchers at the University of Connecticut and the University at Albany have launched the North American Forecasting Weather, Outage, Load and Damage Initiative, backed by $550,000 from a coalition of major utilities and the National Science Foundation. The model fuses high-resolution weather data with artificial intelligence and historical outage records to pinpoint, before a storm arrives, where the grid is likely to fail. Pilots will begin in New England, New York, and California.

The timing is not incidental. US regulators approved 43 utility rate increases in 2025, adding $11.6 billion to customer bills. New York electricity costs have risen 33% over the past decade. Faster recovery would reduce the operating costs that justify those rises, or so the argument goes. Whether utilities will translate better forecasts into meaningfully shorter outages, rather than into leaner staffing, remains an open question.

The project draws on more than a decade of outage-prediction research at WISER, a joint centre established in 2023. Its advisory board reads like a roll-call of North American grid operators: Avangrid, Con Edison, Eversource, National Grid, Pacific Gas and Electric, Southern California Edison, the New York Power Authority, and HydroQuébec. NASA is also involved, which gives the initiative a useful gloss of scientific credibility.

Extreme weather is intensifying; electricity demand is rising, driven by data centres and electric vehicles; and the grid was not designed for either. Smarter forecasting will not fix ageing infrastructure, nor resolve the financing disputes that slow repairs. But knowing where the next outage will strike, hours before it does, is at least a more honest starting point than pretending storms are surprises.