The power problem for AI infrastructure just got a corporate solution at scale. Microsoft confirmed it’s building a new data center campus in West Texas, with the project expected to deliver approximately 2.67 gigawatts of capacity to meet growing AI demand. Supplying that power: a partnership between Chevron, Engine No. 1, and GE Vernova, three companies building what they describe as scalable, reliable power generation for U.S. data centers. GE Vernova’s role is confirmed via Chevron’s own announcement, which notes that a majority of the generation will come from large GE Vernova turbines and associated electrical infrastructure, with additional capacity provided by Solar Turbines, a wholly owned subsidiary of Caterpillar Inc. Engine No. 1 has been collaborating on the development alongside Chevron.
The structure matters more than the headline. Electricity will flow directly to the Microsoft campus, designed to deliver reliable, dispatchable electricity to Microsoft while aiming to mitigate impacts on the regional grid. That behind-the-meter framing aligns with the deal’s logic: a hyperscaler co-locating generation with consumption to guarantee uptime at a scale that public utilities can’t currently queue fast enough to serve.
The project is referred to as “Project Kilby,” involving a 20-year power purchase agreement signed between Energy Forge One LLC, a wholly owned Chevron subsidiary, and Microsoft. The agreement represents an important milestone toward Chevron’s Final Investment Decision (FID), which is expected by the end of 2026, subject to the completion of other necessary conditions. First power delivery is anticipated in 2028. The project is targeting mid-teen returns and is expected to generate diversified cash flow independent of oil and gas price cycles.
The real story is the procurement model, not the project cost. Hyperscalers have been running into the same wall across every major grid: interconnection queues measured in years, not months. ERCOT, PJM, MISO — the line to get firm grid capacity is long, and AI data center build-out timelines don’t bend to it. Behind-the-meter generation, co-located with the facility, sidesteps that constraint entirely. Microsoft doesn’t wait for a queue. It builds the generation.
This is part of a broader pattern of energy incumbents partnering directly with hyperscalers on dedicated AI power. The energy industry isn’t selling power to Microsoft. It’s building a company with Microsoft to produce it.
What to Watch
What to watch
the final investment decision, targeted by end of 2026, is the real commitment signal. Watch Chevron and Engine No. 1 investor communications in Q3 and Q4 for any confirmation. If FID clears on schedule, first power in 2028 makes this a 2027-2028 capacity story, relevant for anyone modeling AI infrastructure availability at hyperscale. The ERCOT bypass angle also has a regulatory dimension: FERC’s ongoing pressure on behind-the-meter arrangements means this structure isn’t without policy risk.
The catch is that the FID-pending status is doing a lot of work here. The partnership is real and confirmed. The financial commitment is still conditional. Don’t read any reported cost figure as money already in the ground.