Generative AI Begins to Redefine Power Grid Decisions

New DOE and NREL research explores how generative AI could help utilities plan faster, analyze data better, and boost grid resilience with caution

19 Dec 2025

A technology famed for writing poems is edging into a far less forgiving arena. In North America’s power sector, generative artificial intelligence is beginning to attract serious attention, not as a cure-all, but as a possible aid to an increasingly complex grid.

Studies from the National Renewable Energy Laboratory (NREL), supported by the Department of Energy (DOE), strike a careful tone. They urge utilities to experiment through pilots and partnerships rather than sweeping deployments. The restraint reflects hard reality. Power systems are critical infrastructure, and mistakes can be costly.

The appeal of AI is clear. Modern grids generate torrents of data from sensors, markets, weather models and control rooms. Extreme heat, storms and wildfires add further strain. Human operators, however skilled, struggle to absorb it all in real time. NREL researchers argue that generative AI could help by filtering noise, summarising risks and highlighting priorities during outages or emergencies.

Yet the research draws a firm line. The aim is not to replace operators, but to support them. DOE-backed work repeatedly stresses “human-in-the-loop” systems, where people retain final authority. In electricity networks, reliability matters more than speed, and accountability more than novelty.

Planning offers another opening. Expanding transmission or upgrading substations can take years of modelling and scenario testing. AI-assisted tools might compress that process, allowing planners to test how future demand, from electric vehicles, data centres or broader electrification, could strain the grid. Faster analysis could, in theory, mean faster investment.

The barriers are substantial. Generative AI can hallucinate errors. Greater digitalisation widens the attack surface for cyber threats. Rules on data use, accountability and liability remain unsettled. For now, DOE and NREL stop short of calling for new regulation. Instead, they advocate early engagement with regulators to build trust before AI becomes embedded in daily operations.

The message is cautious but not dismissive. Generative AI will not solve the grid’s problems on its own. But as electricity systems grow more complex, ignoring a tool that might help manage that complexity carries its own risks. For utilities, the challenge is to move slowly enough to stay safe, yet quickly enough to stay relevant.