Gartner warns AI power crunch

- Gartner said on June 10 global data-center electricity use will rise 26% in 2026 to 565 terawatt-hours as AI workloads drive demand. (gartner.com) - Gartner said AI-optimized servers will account for 31% of data-center power use in 2026 and surpass conventional servers in 2027. (gartner.com) - Meta said on July 8 it is building a 1-gigawatt Alberta data center, while blocked U.S. projects are searching for new sites. (cnbc.com)

Gartner’s warning about an AI power crunch is a story about electricity, not chips. On June 10, the research firm said global data-center electricity consumption is projected to rise 26% in 2026 to 565 terawatt-hours, up from 447 terawatt-hours in 2025. Gartner said the increase is being driven by compute-intensive AI workloads and that power availability is becoming the main constraint on new capacity. (gartner.com) Linglan Wang, a director analyst at Gartner, said power security is becoming “the new battleground” for scaling AI while protecting margins. ### Why is Gartner focusing on electricity now? Gartner said the immediate issue is how fast AI servers are changing the power mix inside data centers. (cnbc.com) The firm said AI-optimized servers will account for 31% of data-center power consumption in 2026, even before they overtake conventional servers in 2027. Linglan Wang said the demand surge is coming from compute-heavy AI workloads rather than ordinary enterprise processing. Gartner’s forecast also said AI capacity is now constrained by power availability, which shifts the bottleneck from semiconductors and buildings to grid access and secured electricity supply. (gartner.com) ### How big is the jump from AI servers alone? Gartner said the power required for incremental AI-optimized servers will reach 500 terawatt-hours a year in 2027. In an April 15 release, the firm said that figure would be 2.6 times the level recorded in 2023. (gartner.com) A November 2024 Gartner release set out the operational risk more directly. The firm said 40% of existing AI data centers could be constrained by power availability by 2027 as electricity demand from AI and generative AI outpaces utility capacity. ### What are companies doing when power is scarce? (gartner.com) Meta said on July 8 that it is building its first large data center in Canada, a 1-gigawatt facility in Alberta. CNBC reported the project will cost about $9 billion, take two to three years to build and become Meta’s 33rd data center overall. The Alberta project shows how site selection is shifting toward places that can offer power at scale. (gartner.com) Gartner’s forecast said companies should prioritize power-secure regions and utility-integrated infrastructure, and Meta’s move north of the border fits that pattern. That link is an inference based on Gartner’s forecast language and Meta’s announced project details. (gartner.com) ### Why are some projects stalling instead of moving ahead? OilPrice.com reported on July 9 that communities across the United States blocked or delayed more than $130 billion in AI data-center projects in the first three months of 2026. The report said developers are now looking for locations where permits, power access and local backing are already in place. (cnbc.com) CNBC reported in March that hyperscalers were already facing backlash over electricity costs as households and policymakers questioned whether data centers were pushing up power bills. That adds a second pressure point beyond engineering: even where companies can finance the projects, they still need community approval and utility arrangements. (gartner.com) ### What comes next if the grid stays tight? Gartner said in April that small modular nuclear reactors could become one answer for large data centers when electricity demand exceeds grid supply. The firm framed that as a possible dedicated on-site power source rather than a near-term replacement for grid connections. (oilprice.com) The next milestones are already dated. Gartner’s forecast points to 2027 as the year AI-optimized servers surpass conventional hardware in power consumption, and Meta said its Alberta facility will take two to three years to construct after the July 8 announcement. (gartner.com 1) (gartner.com 2) (cnbc.com)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.