The Energy Cost of AI and Data Centers

We study the spillover effects of rising AI and data center electricity demand on the rest of the U.S. economy over the next decade. Combining project-level data on planned data center developments with regional power supply curves built from individual power plants, we calibrate a multi-region spatial equilibrium model to quantify effects on electricity prices, output in non-data-center sectors, and emissions. Despite a substantial load increase, we predict only a modest increase in national electricity prices on average, because existing fossil-fuel power generation has considerable spare capacity. Aggregate output losses in non-data-center sectors are less than 0.1% of GDP, while the energy sector emerges as a large winner. However, effects are highly uneven across regions: price increases and output losses are concentrated in a small number of exposed areas, like Texas, Virginia, and the Carolinas. Emissions increase as new electricity demand is met largely by ramping up existing gas plants. Investment in local power supply can offset these spillovers, but doing so requires large capital expenditures. Building a larger share of renewables has little effect on electricity prices but reduces emissions and redistributes profits within the energy sector away from fossil fuel producers. The main economic cost of AI's electricity demand is not the average price effect itself, but the amplification of two structural risks: diminished grid reliability and exposure to volatile fossil fuel prices.