Electricity demand is rising again after years of relative stagnation, with data centers emerging as a major source of uncertainty for utilities, regulators and lawmakers. National electricity consumption reached 4,195 terawatt-hours in 2025, 3. 2% above its 2022 level, but no federal agency yet maintains a comprehensive, regular measure of data-center electricity use. Estimates put data centers below 5% of U.S.
Forecasts have shifted sharply in regions attracting large computing facilities. Georgia Power’s projection of less than 400 megawatts of peak-load growth through winter 2030-31 in its 2022 plan rose to 6,600 megawatts a year later and roughly 8,500 megawatts in its approved 2025 plan. PJM attributed that revision to changes in electric-vehicle assumptions, economic conditions and large-load adjustments.
The fiscal impact of the buildout has also been difficult to estimate. Thirty-eight states provide some tax exemption for data-center equipment or electricity.
Federal regulators are moving to speed connections for large electricity users while requiring grid operators to address reliability and cost allocation. The Federal Energy Regulatory Commission unanimously directed six regional grid operators serving about two-thirds of its jurisdiction to explain how they will connect large loads more quickly, prevent grid-upgrade costs from shifting to households and manage power supplied from facilities located alongside data centers. Operators were given 30 days to respond on supply adequacy and 60 days to address integration rules. The commission’s order requires data centers to pay the full cost of grid upgrades needed for their connections, but it does not by itself resolve shortages when new generation cannot be built as quickly as demand is arriving.
Policy proposals now include making data-center load a permanent category in federal energy statistics, requiring states to publish the actual cost of tax incentives, creating large-load tariffs and allowing privately financed, physically separated power systems to serve new industrial customers under contract. A voluntary federal pilot launched in March covered 196 identified companies in Texas, Washington, Northern Virginia and the District of Columbia. Separately, a House-passed ratepayer measure would have required states to consider making facilities with peak demand of at least 100 megawatts pay the incremental generation, transmission and distribution costs they create, but it failed to advance in the Senate on Sept. As of Sept. 15, lawmakers in 16 states were considering temporary pauses or longer prohibitions on new data centers, making measurement, permitting, interconnection and cost responsibility central issues in the next phase of the AI infrastructure expansion.