US Power Grid Strained by Rising AI Data Center Energy Demand

A severe heatwave sweeping the United States is exposing the limits of the national power grid as artificial‑intelligence (AI) data centres consume an increasingly large share of electricity. Data centres already account for about 4 % of U.S. power use and analysts project the share could rise to 9 % by 2030. Grid operators such as PJM have asked the U.S. Department of Energy to require data centres to switch to backup generators within 15 minutes of an emergency alert, and politicians from Texas to New York are urging tighter oversight of new facility siting.

Field trials in Arizona and the United Kingdom show that AI data centres can reduce their power draw by more than 30 % during periods of grid stress by dynamically shifting workloads. The trials, led by Nvidia, Oracle and Emerald AI, used a software platform called Emerald Conductor to pause or reschedule GPU tasks in response to real‑time grid signals, delivering up to 35 % peak‑load reductions without hurting performance.

Research from MIT and market analyses suggest that greater flexibility could lower consumer electricity prices – by as much as 5 % in Texas – if data centres move a substantial portion of their load to off‑peak hours. Meanwhile, utilities and investors are preparing for a surge in capital spending, with forecasts of $240 billion of utility investment by 2026 to meet AI‑driven demand, and firms such as Taiwan’s Hua‑Cheng reporting more than $200 billion in AI data‑center transformer orders for the North American market.