2030 means 273GW! Goldman Sachs drastically raised global data center capacity and predicted that capital expenditure over the next 5 years could reach 8.5 trillion US dollars

The Zhitong Finance App learned that Goldman Sachs said in a newly released research report that it has updated the global data center capacity forecast based on the latest data from the 451 Research database. Currently, it is estimated that by 2030, the global data center supply will reach 273 GW (previously the forecast was 217 GW at the time of the last update in July 2026), which means that 170 GW of capacity will be added after 2025, and the net additional capacity in 2026, 2027, and 2028 will be 25 GW, 32 GW, and 35 GW, respectively.

This increase is mainly due to Goldman Sachs raising 451 Research's capacity estimates for the latter part of the forecast period to include unannounced projects. Goldman Sachs anticipates that the net increase in data center capacity will not slow down in the medium term. According to empirical estimates of investing 50 billion US dollars per GW, this additional capacity of 170 GW means that capital expenditure will reach 8.5 trillion US dollars over the next 5 years. But at the same time, Goldman Sachs also noticed that IT infrastructure costs continue to face upward pressure.

As far as the US market is concerned, Goldman Sachs expects the data center capacity to reach 152 GW by 2030 (previously predicted 125 GW), which means an additional 103 GW of capacity will be added after 2025. This forecast increase is mainly due to Goldman Sachs raising estimates for the latter part of the forecast period to include unannounced projects. In addition, net additional capacity from cloud services, emerging cloud service providers (neocloud), cryptocurrency, wholesale and retail data center operators also contributed to a partial increase (cumulative increase of 11.8 GW from 2026 to 2030). Goldman Sachs believes this reflects that emerging cloud service providers are rapidly scaling up to meet GPU demand.

Goldman Sachs anticipates that high-performance computing (HPC) data centers (usually associated with AI workflows) will drive most of the net increase in US data center capacity, and that by 2028, projects of at least 75MW will account for 75% of the net increase in capacity. The US will add a total of 103 GW of data center capacity over the next 5 years. According to empirical estimates of investing 50 billion US dollars per GW, this will mean a capital expenditure of 5.2 trillion US dollars.

Goldman Sachs believes that the current market's general expectations for the capital expenditure of US hyperscale cloud service providers are sufficient to cover this portion of expenditure. The average annual capital expenditure is about 1.51 trillion US dollars. Historically, this forecast has been raised several times, and capital expenses for private AI laboratories and capital expenses for data center platforms have not been included, but capital expenses in the international market have not been excluded from this.

Goldman Sachs pointed out that implementation risks facing data center construction and delivery are rising due to factors such as power supply issues, regulatory uncertainty, and high leverage. Despite this, Goldman Sachs believes that data center capacity is expected to maintain continuous and unconventional growth for projects with high certainty about power supply and licensing.

Additionally, Goldman Sachs added that recently, investors have raised more and more questions about data center directives issued by the governors of Texas and Pennsylvania that may cause delays in the project. Both directives effectively suspended the expedited approval process within their respective states.

Goldman Sachs believes that the main impact will fall on speculative construction projects, but we still need to be wary of the ongoing risks brought about by related negative news and the possibility that project development progress may slow down. The Goldman Sachs Sustainability Research Team (GS Investigations) recently wrote a report detailing the main reasons the community is opposed to data center construction, and possible mitigation options. The team notes that these solutions may not completely allay the concerns of the community. Community perceptions of the social and/or economic impact of AI, transparency in individual data center development processes, and opposition from the broader “Neighborhood Avoidance Effect” (NIMBY, “don't build in my backyard”) may all take precedence over innovation and solutions.

High performance computing (HPC) data center construction usually has a high leverage ratio, and the loan cost as a percentage of the total cost (loan-to-cost) usually reaches 80% to 90%. This means that once debt yield spreads narrow, project returns will be highly sensitive to effects.

Goldman Sachs will continue to pay attention to further pressure on hyperscale cloud service provider bond yields, as well as pressure on industry yields caused by negative industry news and increased project delays. As credit costs rise, Goldman Sachs expects the market to pay more attention to the lease term of data center assets and the credit quality of tenants. Asset-backed debt financing from high-quality data center projects can obtain contract cash flow from investment-grade tenants for 15 to 30 years.

Additionally, as projects mature, capital recycling (capital recycling) provides developers with a way to monetize existing projects, thereby funding new project development at lower capital costs.