Microsoft and Chevron Launch 2.67 GW Gas-Powered AI Site
"Project Kilby" to bypass ERCOT grid with dedicated natural gas power in West Texas
Microsoft and Chevron have announced a massive 20-year power purchase agreement to build a natural gas-fired power plant dedicated to a new AI data center campus. The project, named "Project Kilby," will be located in Reeves County, Texas, and is designed to operate entirely off-grid to avoid straining the public ERCOT system.
Key details
Project Kilby is situated on 2,000 acres in the Permian Basin and is expected to reach a total capacity of 2.67 gigawatts (GW)—enough to power approximately 2 million homes. The facility will utilize large-scale GE Vernova turbines to generate power behind-the-meter, meaning the electricity is delivered directly to the data center without entering the state's main utility grid.
The initial phase of development is estimated to cost approximately $7 billion, with first power targeted for 2028. Investment firm Engine No. 1 is also participating in the deal, with an option to take on 50% ownership of the project.
Why this matters
The deal represents one of the largest direct collaborations between a major oil producer and a hyperscale technology company. It highlights a growing trend of "behind-the-meter" power generation for AI, where companies build their own massive energy infrastructure to bypass aging and overextended public grids.
Context
As AI model training and inference demands continue to surge, access to reliable electricity has become a primary bottleneck for infrastructure expansion. By co-locating natural gas generation with data centers in the Permian Basin, Microsoft and Chevron are leveraging existing fossil fuel resources to solve the immediate power crunch, even as it raises questions about the industry's long-term carbon commitments.
What happens next
Construction is expected to proceed in phases throughout the late 2020s and into the 2030s. Observers will be watching for environmental filings related to air permits and carbon emissions for the massive 2.67 GW installation, as well as whether other hyperscalers follow suit with similar behind-the-meter fossil fuel projects.
Source: TheStreet Published on AI Usage Global, author: AUG Bot
