Microsoft and Chevron Sign $7 Billion, 20-Year Agreement to Power AI with Natural Gas in Texas

The Kilby project will consume 2.7 GW in Reeves County, with off-grid natural gas generation from the Permian Basin. First delivery set for 2028, featuring turbines from GE Vernova.
Chevron announced on Monday that its subsidiary Energy Forge One has signed a 20-year power purchase agreement with Microsoft to supply a dedicated data center campus in Reeves County, West Texas. The project, named Kilby, will consume approximately 2.7 gigawatts (enough power for 2 million homes) and will operate off the ERCOT grid, utilizing natural gas extracted from Chevron wells in the Permian Basin. People familiar with the deal have estimated the total cost to be around $7 billion.
The Engineering of Off-Grid
The decision to remain unconnected to the Texas grid addresses a central bottleneck in AI expansion in the United States: interconnection queues that already exceed five years in markets like Virginia. By placing 2.67 GW of dedicated generation on over 2,000 acres adjacent to its own load, Microsoft circumvents the regulatory wait and captures economies of scale that would be difficult to achieve in a traditional PPA. Joulent, an energy company backed by activist fund Engine No. 1, will operate the plant, with GE Vernova supplying the majority of the gas turbines and Caterpillar fulfilling part of the contract.
For Chevron, Kilby serves as a test case for a new revenue model. The oil company will leverage associated gas from Permian production, which was previously partially burned in flares, and transform the excess volume into predictable revenue over two decades. The final investment decision will be made in 2026, with the first energy delivery expected in 2028 and a gradual ramp-up to full capacity of 2.67 GW. The $190 billion in capex that Microsoft projects for the calendar year 2026 now gains an anchor point in dedicated firm generation with Kilby.
The Permian Against Frankfurt, Tokyo, and São Paulo
The agreement positions the Permian on a short list of regions capable of delivering gigawatts in a short timeframe, and the contrast with other markets explains the premium Microsoft is paying. In Ireland, EirGrid has maintained a near-moratorium on new data center connections in Dublin since 2022 and is only now discussing additional capacity windows. In Germany, even after the KRITIS package, the electricity price for large industrial consumers hovers around EUR 0.28 per kWh, more than double what Microsoft is expected to pay at Kilby when considering the associated gas cycle. In Japan, a regional hub for Southeast Asian clients, OCCTO conducts firm capacity auctions that have forced NTT Group to shift projects from Tokyo to Osaka.
The message to CIOs in Frankfurt, Tokyo, and São Paulo is direct: when the frontier of AI starts to rely on combining self-generation from gas, land, turbine supply, and water infrastructure into one package, regions lacking this combination fall to the second tier. In Brazil, replicating this equation would require a partnership with an integrated gas producer like Petrobras or Eneva and the long-term capital traction that BNDES alone does not provide. Without this, the waitlist for firm power in data centers, currently nearing 24 months, is likely to extend rather than shorten.
Carbon, Capture, and Replicability
Kilby arrives at a time when Microsoft is facing criticism for the disconnect between its climate goals (net-zero emissions by 2030) and the pace of fossil generation addition. Chevron itself admits that carbon capture at the plant will only be partial. For Microsoft’s investors, however, the trade-off has already been priced in: the agreement locks in marginal electricity costs and removes regulatory risk from the AI roadmap, precisely what competitors relying on the grid lack. Amazon and Google, still dependent on wind and solar PPAs to deliver Project Rainier and the Gemini data centers, will have two quarters to decide whether to accept recalibrating their carbon balance or fall behind in the race for firm capacity.
The less discussed question is how far this model can be replicated. The Permian combines cheap gas, open land, and an oil company willing to operate outside its historical core, along with an activist fund willing to absorb operational risk over two decades. Not every market has this combination, and the recent rally in natural gas at Henry Hub shows that the economic equation of 2026 may not be repeated in 2028.