A RAND study finds that rapidly growing electricity demand from AI data centers could strain the US power grid, especially because new capacity is uncertain and unevenly distributed across regions. Although thousands of gigawatts of generation projects are proposed, only a small fraction is likely to become reliable, usable capacity once delays, retirements, and technical constraints are considered.
By 2030, the U.S. is projected to add about 82 gigawatts of net reliable capacity—33 GW from traditional grid-connected projects and 49 GW from behind-the-meter resources like solar and storage. However, much of this growth is concentrated in a few regions, while others face flat or declining available capacity.
The report, How Much More Power Can the US Grid Provide for AI?, concludes that AI-driven electricity demand may outpace reliable supply in certain areas, making grid planning, regional coordination, and realistic project forecasting critical to avoiding future capacity shortfalls.
