Microsoft, Google and Amazon are backing power projects as the expansion of artificial intelligence puts growing pressure on electricity infrastructure.
The race to build more powerful artificial intelligence is becoming a race to secure electricity. Alongside advanced chips and data centres, technology companies increasingly need access to dependable generation, substations and grid connections capable of supporting their expansion.
The International Energy Agency’s updated 2026 projections put global data-centre electricity consumption at approximately 950 terawatt-hours by 2030, up from 485 terawatt-hours in 2025. That would represent around 3% of global electricity demand. These figures cover all data-centre activity, including AI, rather than AI training alone.
The pressure is particularly acute where large facilities cluster together. Even when sufficient electricity can be generated across a country, delivering it to a specific location requires adequate transmission and distribution infrastructure. In its 2025 assessment, the IEA estimated that grid constraints could delay around one-fifth of planned data-centre projects unless those risks were addressed.
That challenge is drawing technology companies deeper into energy development.
Microsoft’s 20-year power purchase agreement with Constellation underpins the planned restart of Three Mile Island Unit 1 in Pennsylvania, renamed the Crane Clean Energy Center. Constellation expects the facility to return in 2027, restoring approximately 835 megawatts of generating capacity, subject to the required approvals. The arrangement makes Microsoft a long-term customer supporting the project’s economics; Constellation remains the operator.
Google is expanding its geothermal commitments. On September 1, Fervo Energy announced a 396-megawatt power purchase agreement with Google for its Cape Station development, with the contracted capacity expected online in 2028. Fervo said the electricity would support a potential Google data centre in Utah.
Amazon, meanwhile, has invested in small modular reactor developer X-energy. Their collaboration targets more than five gigawatts of new nuclear projects in the United States by 2039. That is a deployment ambition extending well into the next decade, rather than capacity available to serve today’s demand.
Together, these deals show technology companies taking a more active role in financing and securing power. Long-term purchase commitments can give developers confidence to proceed with expensive projects, while equity investments support emerging technologies.
But generating electricity and delivering it are separate challenges. A power contract cannot, by itself, eliminate a congested grid connection or accelerate the manufacture of essential equipment. The IEA identifies growing pressure on supplies of components such as transformers as electricity systems expand.
Calling this a universal energy crisis would overstate the evidence. Constraints differ by region, and electricity has joined a wider set of requirements that includes processors, networking, cooling and construction capacity.
The commercial implication is nevertheless substantial: buying computing hardware is only part of building an AI business. Companies must also secure the infrastructure that lets it operate. Electrical engineers, utilities and grid developers are becoming increasingly important to determining where—and how quickly—the next generation of AI can be deployed.
Sources:
