Updated
Updated · Utility Dive · May 28
AI Hyperscalers Lock Up 10.5 GW+ of Power as Grid Shortfalls Force Billion-Dollar Generation Bets
Updated
Updated · Utility Dive · May 28

AI Hyperscalers Lock Up 10.5 GW+ of Power as Grid Shortfalls Force Billion-Dollar Generation Bets

3 articles · Updated · Utility Dive · May 28
  • Microsoft, Meta, Amazon and Google are securing their own electricity supplies as AI-driven data center demand outpaces grid expansion, turning major cloud groups into de facto power developers.
  • 1,050 TWh of global data-center electricity use is approaching nearly triple 2024 levels, while AI racks draw 40-100+ kW versus 5-15 kW for traditional racks and interconnection timelines still run for years.
  • Microsoft agreed to buy all output from a restarted Three Mile Island under a 20-year contract, while its Brookfield partnership adds more than 10.5 GW over a $10 billion-plus investment horizon.
  • Meta locked up 6.6 GW through TerraPower, Oklo and Vistra, Amazon secured 1.92 GW from Susquehanna, and Google targeted 500 MW from Kairos SMRs by 2035 alongside a $40 billion Texas buildout.
  • The build-your-own-power push is extending into 800-volt DC systems and critical components, where more than 70% dependence on Chinese dielectric film is emerging as a strategic supply-chain risk.
With AI's power grid built on Chinese parts, what is the plan if the supply chain is cut?
As tech giants build private grids, who will ultimately control the world's energy supply?

The 10 GW AI Data Center Surge: How Hyperscalers Are Reshaping Global Energy, Costs, and Climate by 2026

Overview

By mid-2026, the explosive growth of AI hyperscalers is dramatically reshaping the global energy landscape. Their insatiable demand for computing power is pushing electricity grids to their limits and forcing a redefinition of infrastructure priorities. Power availability has become the main barrier to data center expansion, as the rapid development of new server farms outpaces the ability to build new grid and generation capacity. As a result, the total power load from AI-driven data centers is expected to hit 10 GW globally by the end of 2026—a ceiling set by physical constraints rather than a slowdown in demand.

...