An oil major and a software giant just agreed to build a power plant together — and the reason says everything about the state of the AI boom.
Chevron and Microsoft announced on June 22, 2026, that they have signed a 20-year power purchase agreement to co-locate a dedicated natural-gas plant with a new Microsoft AI data-center campus near Pecos, in Reeves County, West Texas. The development, called Project Kilby, is structured around Energy Forge One LLC, a wholly owned Chevron subsidiary, and is being developed in collaboration with investment firm Engine No. 1. It is expected to begin producing power in 2028 and to ramp to roughly 2.67 gigawatts — enough electricity, Chevron says, to supply more than 530,000 Texas homes.
The economics are as large as the machinery. Chevron pegs the total capital outlay at $7 billion to $9 billion, and estimates the plant will generate more than $10 billion in state and local tax revenue while supporting almost 2,000 jobs. A final investment decision is expected before the end of 2026, subject to remaining conditions. A majority of the generation will come from large GE Vernova turbines, with additional capacity supplied by Solar Turbines, a Caterpillar subsidiary. The gas itself will be produced locally, in the Permian Basin that surrounds the site.
"AI is reshaping the global economy, and abundant, affordable, reliable energy is essential to fueling that transformation," said Jeff Gustavson, Chevron's president of New Energies, in announcing the deal.
The critical detail is how the power gets to the servers. Project Kilby is a behind-the-meter installation, meaning the plant sits directly alongside the data center and feeds it power without routing through the public ERCOT grid. That design lets Microsoft avoid the interconnection queues that have become the single hardest constraint in the industry. It also means the campus does not compete with local ratepayers for electricity — at least not directly.
"There's really no competition with local electricity consumers," Gustavson told CNBC. "In fact, over time, as we have excess power, we plan to push that into the grid to help stabilize it." The plant is being built through a phased, modular approach, allowing capacity to be added incrementally rather than all at once. Microsoft has separately outlined plans for a campus at Pecos approaching 2 gigawatts of IT load, placing it among the largest single AI sites announced anywhere in the country.
Why It Matters
The AI race is no longer bottlenecked by chips. It is bottlenecked by electrons. Across the roughly 84 large AI data-center facilities tracked in the United States, planned capacity runs to nearly 43 gigawatts — and what now decides how fast that capacity comes online is the grid, not the buildings. In the markets carrying most of the 2026 buildout, including Dallas, Phoenix, and Northern Virginia, utility interconnection queues stretch four to seven years. A campus that joins a Northern Virginia queue this spring cannot realistically expect grid power before 2030, no matter how quickly the structure goes up.
That math is why behind-the-meter gas has become the dominant strategy for building AI capacity in a hurry. A dedicated gas plant can be stood up in as little as 18 months, against a multi-year wait for grid interconnection. Roughly 50 gigawatts of behind-the-meter gas projects were announced in 2025 alone. Project Kilby is the largest and most visible expression yet of that pivot — and it turns Chevron, historically an upstream oil-and-gas producer, into something closer to an AI-infrastructure company.
The trade-offs are real. Behind-the-meter gas locks in fossil-fuel generation for the 20-year life of the contract, sitting awkwardly against Microsoft's carbon-negative pledge. It also raises questions regulators are only beginning to grapple with: when a hyperscaler builds its own gigawatts outside the grid, who bears the cost of the shared transmission system it still relies on, and what happens to local emissions and water in a region already stretched thin. Texas, home to about a third of the nation's tracked AI capacity, is fast becoming the test case for all of it.
What to Watch
The first hard signal is Chevron's final investment decision, due by year-end 2026; until then, the $7-to-$9 billion figure is a commitment on paper. Watch whether GE Vernova turbine lead times — a chronic industry chokepoint alongside a multi-year transformer backlog — hold the 2028 start date. Watch, too, whether Chevron actually exports surplus power to ERCOT as promised, and how Texas regulators treat behind-the-meter megaprojects that consume Permian gas while sidestepping the grid. If Kilby delivers on schedule, expect a wave of oil majors to follow Chevron's lead — pairing wellheads with server halls as the defining infrastructure bet of the decade.
"AI is reshaping the global economy, and abundant, affordable, reliable energy is essential to fueling that transformation."- Jeff Gustavson, President of New Energies, Chevron