Overview
- Gartner’s July 14 forecast projects global data center electricity use will rise to 565 terawatt‑hours in 2026 and shows AI‑optimized servers consuming 175 TWh in 2026 and an estimated 258 TWh in 2027, a level that would exceed conventional server power use.
- Power availability is now the main bottleneck for new AI capacity because utilities must complete interconnection work and permitting while lead times for transformers, chillers and other equipment remain long.
- Developers have responded by activating capacity with on‑site gas turbines, adding battery storage, and signing long‑lead deals for firm power including nuclear offtakes, but those measures do not remove short‑term grid constraints.
- Cooling demand is rising with denser AI racks, and Gartner forecasts data center cooling electricity will climb to about 195 TWh in 2026, increasing overall site loads and strain on local infrastructure.
- Fact checks citing LBNL, EPRI, the EIA and other studies find that U.S. data center demand is far below President Trump’s public claim, with projections ranging roughly from 6% to 25% of U.S. electricity in various future scenarios, and policymakers should expect more local delays, higher rates, and tighter scrutiny.