Hyperscalers lock in long‑term power contracts to secure capacity for AI growth
- Microsoft, Amazon and Google have each signed multi-year power deals tied to new nuclear capacity, while the White House said seven major AI companies pledged on March 4 to cover data-center power needs. - Microsoft’s Constellation agreement backs an 835-megawatt Pennsylvania reactor restart; Amazon’s X-energy deal scales to 960 megawatts; Google’s Kairos pact targets up to 500 megawatts from small modular reactors. - Data-center electricity demand is projected to more than double by 2030, pushing power procurement from back-office function to expansion bottleneck. (iea.org)
The biggest cloud companies are no longer treating electricity like a routine utility bill. They are signing decade-long deals to secure new power for artificial intelligence data centers. (microsoft.com) (aboutamazon.com) (blog.google) Microsoft said on September 20, 2024 that it signed a power purchase agreement with Constellation to help restart the 835-megawatt Crane Clean Energy Center in Pennsylvania, a reactor retired in 2019. Microsoft said the plant is expected to return to service in 2027 and feed the PJM grid. (microsoft.com 1) (microsoft.com 2) Amazon said on October 16, 2024 that it would support X-energy and Energy Northwest on small modular reactor projects in Washington state, with 320 megawatts in the first phase and an option to expand to 960 megawatts. Amazon said the output is meant to support future cloud and artificial intelligence demand in the Pacific Northwest in the early 2030s. (aboutamazon.com 1) (aboutamazon.com 2) Google said on October 14, 2024 that it signed what it called the first corporate agreement to buy power from multiple small modular reactors, with Kairos Power targeting up to 500 megawatts by 2035. In June 2025, Google added a Tennessee Valley Authority arrangement tied to 50 megawatts on the grid serving its data centers in Tennessee and Alabama. (blog.google 1) (blog.google 2) The shift reflects a simple constraint: new AI systems need chips, buildings and transmission lines, but they also need firm electricity that shows up every hour. The International Energy Agency said global data-center electricity use is set to rise from 460 terawatt-hours in 2024 to about 945 terawatt-hours in 2030. (iea.org) (iea.org) In the United States, the Electric Power Research Institute estimated data centers could account for 4.6% to 9.1% of annual electricity generation by 2030, up from about 4% today. The International Energy Agency said U.S. data centers are on course to account for almost half of the country’s electricity-demand growth through 2030. (restservice.epri.com) (iea.org) Washington has started treating that demand as a grid issue, not just a tech issue. The White House said on March 4, 2026 that seven AI companies and hyperscalers agreed to build, procure or fund enough new generation for their data centers and to pay for related grid upgrades. (whitehouse.gov) (powermag.com) Power companies and pipeline operators are also trying smaller on-site supply plays while big nuclear projects move slowly. Tallgrass and Turboden said on March 12, 2026 they would add waste-heat-to-power systems at compressor stations in Columbus and Chandlersville, Ohio, and St. Paul, Indiana, bringing their combined portfolio to 46.1 megawatts. (turboden.com) (pgjonline.com) Turboden said those Ohio and Indiana units will use its Turboanalytics cloud system for anomaly detection, predictive maintenance and forecasting. The projects show how operators are trying to squeeze electricity out of existing industrial assets while waiting for new plants and transmission to clear permitting and construction bottlenecks. (turbomachinerymag.com) (iea.org) The result is that power contracts are starting to look like capacity reservations. The companies that lock in generation first can build more artificial intelligence infrastructure while rivals wait for electricity, not servers. (microsoft.com) (aboutamazon.com) (blog.google)
