AI data center power demand strains US grid: Morgan Stanley forecasts 49 GW power shortfall by 2028

Morgan Stanley Research forecasts that U.S. data center demand will reach 74 GW by 2028, but projects a shortfall of approximately 49 GW in available power access. This mismatch between demand and infrastructure capacity underscores an acute bottleneck: grid modernization has not kept pace with AI deployment. Gartner predicts power shortages will restrict 40% of AI data centers by 2027, forcing geographic consolidation and spurring hyperscalers to pursue "bring your own power" (BYOP) strategies.

The underlying constraint is structural. AI inference—now 80–90% of AI compute load—runs continuous, high-wattage draws that differ fundamentally from training bursts. A single AI task consumes up to 1,000 times more electricity than a traditional web search. Northern Virginia, Texas, Ireland, and East Asia data center clusters are already pushing local grid capacity to operational limits; a July 2024 voltage fluctuation in Northern Virginia disconnected 60 data centers simultaneously and freed a 1,500 MW power surplus.

Hyperscalers are responding with distributed on-site generation: natural gas, fuel cells, and microgrids. Morgan Stanley estimates $1+ trillion in capex for 2025–2026 alone, with financing becoming critical—utilities and grid operators now cite data centers as the primary driver of planned capital expenditure. Grid enhancement technologies (dynamic line rating, advanced switching) and small modular reactors (SMRs) are shifting from policy debates to production readiness.