US Data-Center Power Delivery Emerges as the Critical Constraint on AI Build-Out

US data-center development is being reshaped by a shortage of power that can be delivered within AI campus construction timelines. Rather than waiting for conventional utility interconnection, operators and hyperscalers are seeking sites with available power, pursuing co-location with generation, using interim local generation and participating directly in energy projects.

The timing gap is central. S1 cites typical US data-center construction periods of roughly three to six years, compared with five to 15 years for new grid infrastructure from planning through delivery. Even connections to the existing grid typically required one to three years of waiting in 2024, according to the IEA data cited by S1, with waits of up to seven years at certain nodes.

Forecasts cited in the source illustrate the scale of expected demand, although they use different methodologies and should not be treated as directly comparable. Goldman Sachs reportedly raised its forecast for actual US data-center electricity demand in 2030 from 83 GW to approximately 108 GW, following estimates of about 31 GW in 2025 and 66 GW in 2027. The DOE forecast cited by S1 puts US data-center annual electricity consumption at 649 TWh in 2030, equivalent to an average load of around 74 GW. BNEF, approaching the market from the project pipeline, expects 118 GW of data-center capacity to enter operation by 2030, a 52% increase from its earlier forecast.

On supply, Morgan Stanley estimates that US data centers could still face an electricity shortfall of about 33 GW by 2028, even after accounting for grid and local-generation solutions. TrendForce, as cited by S1, estimates a gap of more than 170 GW in 2030 between data-center demand and grid capacity available for delivery. These are not directly comparable deficit forecasts: they cover different time horizons and definitions.

The response is increasingly visible in investment and development models. Alphabet acquired energy developer Intersect for $4.75 billion, according to S1. Intersect’s energy and data-center projects under construction and in development are expected to represent 10.8 GW of power capacity by 2028; this is expected capacity, not commissioned capacity. Separately, OpenAI and SoftBank each invested $500 million in SB Energy, which is tasked with building a planned 1.2 GW OpenAI data center in Texas alongside supporting energy infrastructure.

Policy is also being directed toward larger projects. S1 says accelerated approval applies to data centers and associated energy infrastructure with more than 100 MW of new load or capital expenditure above $500 million.

For developers, the practical implication is that a project’s competitiveness could increasingly depend on a confirmed power-delivery date. That may favour campuses with available grid capacity and suppliers of distributed gas generation, fuel cells, gas infrastructure, substations and interconnection equipment, while potentially shifting future development toward regions where power can be delivered sooner.