A Goldman Sachs analysis highlights the necessity for more data centers and an even greater power supply to keep pace with the rapid expansion of artificial intelligence.
The global energy consumption attributed to data centers is projected to rise by 50% by 2027 and could surge by as much as 165% by the decade’s end compared to 2023, according to Goldman Sachs Research.
Despite the potential of DeepSeek—an innovation claiming to enable large-scale AI modeling on low-end hardware—the industry remains focused on efficiency rather than alternative methods, according to James Schneider, a senior equity research analyst specializing in U.S. telecom, IT services, and digital infrastructure. However, concerns remain regarding DeepSeek’s scalability, training procedures, and overall infrastructure viability, Schneider noted.
“In the long term, enhanced efficiency leading to reduced capital expenditures—whether from hyperscale operators or new market entrants—could lessen the risks of oversupply beyond 2027. This, in turn, could enhance market stability and reduce volatility in the data center industry,” Schneider explained.
On the demand front, large cloud service providers and major enterprises are developing increasingly complex large language models (LLMs), requiring substantial computational resources for training and operation.
At the same time, hyperscale cloud firms, data center managers, and investors are committing substantial capital to constructing high-capacity data centers. However, according to Goldman Sachs Research, the balance between supply and demand is expected to tighten significantly in the coming years.
As a result, data center utilization rates are forecasted to climb from approximately 85% in 2023 to a peak of over 95% by late 2026. A gradual easing is anticipated from 2027 onward, as new data centers become operational and AI-driven demand expansion slows, Schneider stated.
Goldman Sachs Research estimates the current global data center power consumption at roughly 55 gigawatts, with cloud computing accounting for 54%, traditional business applications using 32%, and AI contributing 14%.
By 2027, energy consumption is expected to reach 84 GW, with AI’s share increasing to 27%, cloud workloads declining to 50%, and traditional business functions dropping to 23%, according to Schneider.
Projections indicate that data center capacity will expand to 122 GW by 2030, with power density within these facilities rising from 162 kilowatts per square foot to 176 KW per square foot, largely driven by AI-driven workloads, Schneider noted.
“Over the past 18 months, data center expansion has been restricted by several factors,” Schneider wrote. These obstacles include delays in utility infrastructure upgrades, supply chain disruptions, and the high costs and long timelines associated with modernizing transmission networks.
