Goldman Sachs: Raising global data center capacity forecast for 2030, with unannounced projects a key variable

Why was the global capacity forecast raised sharply?

Goldman Sachs expects global operational data center capacity to reach 273GW in 2030, up from its July 2026 forecast of 217GW. Compared with approximately 103GW in 2025, that represents a net increase of 170GW over five years and a compound annual growth rate of approximately 21%.
The increase mainly reflects estimates of projects further into the future that have yet to be announced. Goldman Sachs believes the pace of net capacity additions will not slow over the medium term, so it raised its longer-term forecast above the level indicated by 451 Research project-tracking data.
The project-tracking data itself has also improved: 451 Research raised its forecast for net global capacity additions in 2026 by 8% from the first quarter, and raised its forecasts for 2027 and 2028 by 14% each.

Goldman Sachs expects US capacity to reach 152GW in 2030, an increase of 27GW from its previous forecast of 125GW and a net increase of 103GW from 2025. The bank expects the US to account for approximately 60%–70% of total global net capacity additions through 2030.
The longer-term revisions to the US forecast are particularly large: to account for unannounced projects, Goldman Sachs raised its forecasts for US net capacity additions in 2029 and 2030 by 56% and 180%, respectively, from its previous update.
By project category, the forecast for cumulative net additions in cloud, neocloud, cryptocurrency, and wholesale and retail data center businesses over 2026–2030 was raised by 11.8GW. Goldman Sachs believes this reflects rapid expansion by neocloud providers to support GPU demand.
Goldman Sachs expects high-performance computing data centers, which are typically associated with AI workloads, to dominate expansion. By 2028, projects with an individual capacity of at least 75MW are expected to account for approximately 66% of annual net capacity additions globally and 75% in the US.

Using a rule-of-thumb assumption of 50 billion dollars in total construction costs per GW, the 170GW of net global capacity additions over the five years after 2025 corresponds to approximately 8.5 trillion dollars in cumulative capital spending. Goldman Sachs also noted that IT infrastructure costs remain subject to upward pressure.
In the Visible Alpha consensus estimates referenced by Goldman Sachs, capital spending by selected US cloud and AI companies averages approximately 1.51 trillion dollars annually over 2026–2030. This measure includes international capital spending but excludes spending by international hyperscale cloud providers, private AI labs and data center platforms.

Power supply and permitting are affecting specific projects: Microsoft exited a lease commitment after power supply issues delayed a facility's readiness date, and Oracle's New Mexico project also encountered power permitting challenges. Goldman Sachs therefore places greater weight on whether projects can secure power.
Texas and Pennsylvania have suspended their respective fast-track permitting processes. Goldman Sachs believes speculative construction projects are more heavily affected, and the pace of project development could slow.
High-performance computing data center construction is typically financed with loans equivalent to 80%–90% of costs, and returns are highly sensitive to a narrowing of debt yield spreads. Goldman Sachs expects execution risk to be more likely to affect projects not previously included in forecasts, and believes projects with high visibility into power supply and permits still have room to grow.