A RAND Technology and Security Policy Center report extrapolates two exponential trends, total AI chip supply and frontier training compute, to estimate that global AI data center power capacity could reach 68 GW by 2027 and 327 GW by 2030, versus roughly 88 GW for all data centers globally in 2022. Single-site training clusters are projected to require up to 1 GW by 2028 and 8 GW by 2030, levels that strain or exceed what any existing regional grid can reliably deliver. The report identifies permitting timelines, grid interconnection queues, and renewable-power intermittency as the binding near-term constraints, and flags that failure to resolve them could push U.S. firms to site AI infrastructure abroad, eroding domestic compute concentration. For financial institutions, the key structural read is that power availability, not chip supply or capital, is increasingly the rate-limiting variable on AI infrastructure buildout timelines.
382 other tracked developments touching the same claims.