Chevron and Microsoft have struck a 20-year agreement to build and supply natural gas power for a massive artificial intelligence data center in West Texas, a project that crystallizes how far the energy demands of AI have pushed two of America’s largest corporations into each other’s arms. The deal, dubbed Project Kilby, would consume nearly 2.7 gigawatts of electricity, enough to power roughly two million homes.
What the Deal Looks Like
The project will sit on more than 2,000 acres in Reeves County, deep in the Permian Basin where Chevron already operates extensive natural gas fields. Chevron is partnering with Joulent, an energy company backed by activist investment firm Engine No. 1, to construct a power generation complex fueled entirely by local gas production. GE Vernova will supply the large gas turbines, with Caterpillar providing additional units, TechCrunch reported.
Total costs for the project are estimated at roughly $7 billion. Chevron and Microsoft expect to reach a final investment decision by the end of 2026, subject to remaining approvals, with initial power delivery targeted for 2028.
That timeline matters. Microsoft’s AI infrastructure needs are scaling faster than the grid can keep up, and securing dedicated power generation two years in advance is the kind of planning horizon that separates companies with real AI deployment roadmaps from those still making promises.
The Energy Calculus Behind the AI Boom
The Chevron-Microsoft deal is the latest and one of the largest in a growing series of direct power agreements between tech hyperscalers and energy producers. Amazon signed a multibillion-dollar fiber and infrastructure deal with Corning earlier this month. Meta announced a $200 billion data center in Louisiana. Google and Anthropic committed tens of billions to SpaceX’s Colossus data center for computing capacity.
What all these deals share is a recognition that AI compute at frontier scale requires dedicated energy infrastructure that the existing grid cannot provide. A Reuters poll published earlier this month found 57% of Americans oppose new AI data centers in their communities, primarily over energy cost concerns. Building behind-the-meter generation, the way Chevron and Microsoft are doing, sidesteps that grid congestion entirely.
Why Natural Gas, and Why Now
The choice of natural gas over nuclear, solar, or wind is pragmatic, not ideological. Gas turbines can be deployed and scaled within two years. Nuclear plants take a decade or more to permit and build. Renewables require storage infrastructure that does not yet exist at data-center scale. For Microsoft, which has also invested in nuclear restart projects and geothermal, the Chevron partnership is about speed to capacity.
For Chevron, the deal represents a meaningful diversification of its revenue base. Supplying power to AI data centers turns natural gas fields into recurring infrastructure contracts rather than commodity plays subject to price cycles. Engine No. 1, the activist firm behind Joulent, won board seats at ExxonMobil in 2021 by arguing oil majors needed a transition strategy. Backing Chevron’s AI power play is that thesis made real.
The broader pattern is unmistakable: AI is creating a new category of energy customer that is willing to sign 20-year contracts, commit billions upfront, and build dedicated generation capacity rather than compete for grid allocation. World Oil noted this is Chevron’s first direct power supply agreement of this scale with a technology company.
What This Means for the Market
Chevron shares rose 2.49% on Monday, making it the top gainer in the Dow Jones Industrial Average. The market clearly likes the recurring revenue profile of a 20-year infrastructure contract with one of the world’s most creditworthy counterparties.
For Microsoft, the deal secures capacity for AI workloads that are growing faster than most analysts projected even six months ago. The company is building one of the largest AI data centers in the United States, and it now has the power supply locked down before a single server rack goes live.
The question going forward is whether deals like this become the template for AI infrastructure, or whether they remain the exception. If the next generation of large language models requires 10 to 50 times the compute of current ones, as several research labs have suggested, 2.7 gigawatts may turn out to be the floor, not the ceiling.