What Is Blocking U.S. Power Expansion for AI — and What Could Unlock It by 2030?

RAND published this quantitative research brief on June 10, 2026, mapping 66 barriers to U.S. power expansion that AI data-centre demand is now stressing, then prioritising the 17 that federal policymakers could realistically address by 2030. The authors organise the 17 barriers into four clusters and estimate capacity gains from resolving each: permitting delays (16–54 GW), interconnection bottlenecks (65–130 GW), underutilised transmission (up to 7 GW), and limited incentives for supplemental generation (11–106 GW). In aggregate, coordinated federal action across all four clusters could unlock approximately '92 to 297 gigawatts of additional capacity' by 2030. Interconnection reform — enabling faster grid access for new generators without full conventional studies — is identified as 'the most important near-term bottleneck' with the largest potential payoff.

AI infrastructure planning is increasingly constrained by energy availability, not compute. Boards and CIOs building or procuring data-centre capacity in the U.S. should factor in the structural grid constraints RAND quantifies, and policy teams should monitor which of the 17 identified federal actions advance in the current legislative session.

Enterprise infrastructure and real-estate teams should incorporate RAND's grid-constraint analysis into U.S. data-centre siting and capacity-planning assumptions for 2027–2030; government affairs teams should track federal action on permitting and interconnection reform as leading indicators of AI infrastructure availability.