AI-driven data center expansion is straining the U.S. electric grid at a pace grid operators were not prepared for. The Electric Power Research Institute projects AI power demand currently stands at approximately 5 gigawatts and could reach 50 gigawatts by 2030, a figure larger than Pennsylvania's total generating capacity of roughly 49 gigawatts.
Wood Mackenzie estimates that data centers already account for 6 percent of U.S. electricity consumption in 2024, with that share potentially reaching 13 percent by 2030. PJM Interconnection, which manages the grid for 65 million people across 13 states, has warned that the concentration of data center load is producing what the grid operator describes as a massive wealth transfer from ordinary ratepayers to large technology companies.
PJM forecasts peak load increases of 30 gigawatts by 2030, driven primarily by data center construction concentrated in Virginia, Ohio, and the Mid-Atlantic corridor. The grid operator's independent market monitor, Monitoring Analytics, has recommended that data centers be required to bring their own new generation capacity rather than drawing against shared grid resources built for general public use.
The scale of the demand shift becomes clearer in historical context. ENIAC, the first general-purpose electronic computer, consumed 174 kilowatts when it operated in 1946 and reportedly dimmed lights across Philadelphia. A single modern hyperscale AI training facility can draw 100 to 500 megawatts, and utilities are approving dozens of such facilities simultaneously.