Goldman Sachs this week raised its estimate for worldwide data center capacity through the end of the decade, citing an acceleration in AI infrastructure spending and stronger-than-expected project activity. The bank now projects global installed capacity will reach 217 gigawatts (GW) by 2030, up from its earlier projection of 168 GW and more than double the 101 GW it expects for 2025.
The incremental 116 GW of capacity implied by Goldmans revision would require approximately $6 trillion in capital investment. The bank says that level of spending appears alignable with current hyperscaler investment plans, as major cloud providers and large-scale customers rush to secure computing resources.
Goldman identified several power and infrastructure firms that it expects to be among the primary beneficiaries of the accelerating buildout. The bank flagged utilities including FirstEnergy, Xcel Energy, Duke Energy and Sempra, and called out independent power producers Talen Energy, Vistra and NRG Energy for their exposure to markets where data center demand is forecast to grow most strongly. Among data center operators, Goldman left a Buy rating in place on Digital Realty, citing ongoing supply-demand tightness and sustained AI-related capital expenditure as supports for long-term growth.
Drivers and regional concentration
Goldman said data center additions are being revised higher in part based on project tracking from 451 Research, which shows stronger activity across the 2026-2030 window than had been expected in prior quarters. The bank projects a 17% compound annual growth rate (CAGR) in global capacity between 2025 and 2030. It also expects that 60% to 70% of new capacity additions over that period will be located in the United States.
In Goldmans U.S.-specific view, the bank forecasts roughly 125 GW of installed data center capacity by 2030, with annual demand estimated at about 108 GW. That expansion is expected to lift overall U.S. electricity demand growth to a 3.5% CAGR through 2030, creating material opportunities for utilities and independent power producers operating in major data center hubs.
Pricing, supply tightness and the neocloud
Goldman also pointed to a more constructive stance on the growing "neocloud" category after several large leasing and GPU-as-a-service agreements were announced in recent months. The bank cited average lease rates of about $166 per kilowatt per month on contracts with 15-to-20 year terms, a signal of persistent scarcity for large-scale computing capacity. Goldman noted some market participants expect pricing per kilowatt to approach $250 later this year, compared with roughly $70 in 2021.
Even with a higher supply outlook, Goldman said conditions remain exceptionally tight: vacancy rates continue to fall and pricing power is strengthening for operators. The firm also raised its view of data center power consumption, estimating global demand will increase 170% by 2030 versus 2025 levels as AI adoption accelerates. Goldman expects enterprise deployment of AI agents and other agentic workloads to underpin ongoing demand growth even as computing efficiency improves.
Risks and constraints
Goldman warned that several non-demand factors could constrain expansion. These include power availability challenges, permitting delays, labor shortages, equipment constraints and local opposition to new developments. The bank said such factors remain key risks that could slow the pace of future additions despite the strength of demand signals.
Overall, Goldmans updated outlook underscores a materially larger buildout of compute infrastructure over the remainder of the decade and highlights specific segments of the power and infrastructure complex that may see rising revenue streams tied to long-term data center power contracts.