US Grid Expansion Needed to Meet AI-Driven Data Center Surge, Study Warns

McKinsey & Company argues that the most pressing near-term threat to the U.S. power system is building too little capacity to satisfy soaring data-center needs tied to AI compute, which could account for roughly 75 % of electricity demand growth in the next ten years. The firm projects an annual requirement of almost 30 GW, split between IT equipment and supporting infrastructure such as cooling and distribution. Even if AI demand stalls, the newly built generation and transmission assets are expected to remain useful for other loads, enhancing grid resilience.

A separate Bain & Company study warned of a looming supply-demand gap, estimating a national shortfall of 30-55 GW by 2030 after accounting for retirements of 50-75 GW of coal and gas capacity. Utilities are temporarily extending the life of retired plants, and the Trump administration recently authorized up to $500 million to keep 13 coal plants operating. Nonetheless, McKinsey concludes that underinvestment remains the larger risk compared with the possibility of overbuilding amid an AI bubble.