Gartner expects global data center electricity consumption to rise sharply in 2026, reaching 565 TWh, up from 447 TWh in 2025. The firm says the increase is being driven primarily by AI workloads, which are taking a growing share of data center power budgets.gartner+1
The same forecast shows worldwide data center power demand climbing from 105 GW in 2025 to 133 GW in 2026, a gain of about 27%. Gartner says demand could reach 291 GW by 2030, reflecting the scale and speed of AI infrastructure expansion.telecompaper+1
AI Servers Are Becoming the Main Power Driver
AI-optimized servers are no longer a niche category. Gartner estimates they will account for 31% of total data center power consumption in 2026, and by 2027 they are expected to consume more electricity than conventional servers.tomshardware+1
That contrast is striking. Conventional servers are projected to remain almost flat, rising only from 193 TWh in 2025 to 200 TWh in 2027, while AI-optimized servers are forecast to jump from 95 TWh in 2025 to 175 TWh in 2026 and then to 258 TWh in 2027.fourweekmba+1
The U.S. Is a Big Share of the Load
Gartner says the U.S. will account for about 204 TWh of the expected 565 TWh global total in 2026, or roughly 36% of worldwide data center electricity use. Of that U.S. total, dedicated AI data centers are projected to consume 68 TWh, or about one-third.networkworld+1
The implication is that AI data centers have gone from zero to a major share of U.S. power consumption in just a few years. Gartner also expects total data center electricity use to exceed 1,200 TWh by 2030, at which point grid supply may no longer be enough to support continued expansion.theregister+1
Power Security Becomes the Bottleneck
Gartner says the challenge is no longer just building more compute capacity — it is securing enough power to run it. Linglan Wang, Gartner’s director analyst and lead economist, said rising demand for compute-intensive AI workloads is making power availability the key constraint in the global AI race.gartner+1
Wang recommends that infrastructure leaders focus on efficiency upgrades, grid access, advanced cooling, and edge computing to reduce pressure on central facilities. In other words, the next bottleneck for AI growth may be electricity itself, not chips or software