US Data Center Capacity — June 2026 Update

The AI build-out has a conversion problem: the money is there, but the megawatts aren't showing up.

Hyperscaler capital spending keeps climbing — the Big Four are now on track for roughly $725 billion in 2026, up about 77% on last year. Yet only a fraction has turned into operational capacity, with barely 18% of the year's planned spend deployed by the end of Q1.

The physical build-out is lagging its own schedule, and the gap is widening rather than closing. Against the capacity that was supposed to be online by now, the shortfall has grown from about 7.5 GW at the start of 2026 to an estimated 10.5 GW by mid-year.

Independent trackers tell the same story — only a small share of announced gigawatts for 2026 and 2027 is actually under construction — as projects stall on power interconnection, transformer and supply-chain bottlenecks, and local permitting fights.

Demand, meanwhile, is moving the other way. Goldman projects US data center power demand more than doubling, from 31 GW in 2025 to 66 GW by 2027 — pushing data centers from roughly 4% to over 8% of peak summer power use.

The clearest signal that supply can't keep up came in early June. Google — a company with one of the world's largest data center fleets — agreed to pay SpaceX nearly $1 billion a month to rent compute at xAI's facilities, explicitly as "bridge capacity," and tapped the equity markets for $80 billion to fund more build-out.

When the best-capitalized players are renting rivals' GPUs and raising record capital just to keep pace, the trajectory points one way: surging demand running into a build-out that can't deliver fast enough — and a tightening compute shortage as the result.

One caveat: the demand forecast and several of the build-out figures are projections (Goldman's, plus my own estimates for the mid-2026 and 2027 points), not settled facts — so treat the 10.5 GW gap and the "shortage" framing as a well-supported direction of travel rather than a precise certainty.