The plant will not connect to the Texas grid, letting the two companies commit to multi gigawatt AI power without going through the public planning process a normal power plant would face.
Chevron signed a 20-year contract on June 22 to build a 2.67-gigawatt gas-fired power plant in West Texas that will feed a Microsoft data center. The plant will never touch the public grid. That bypass, not the press-release "America's natural gas advantage" framing, is the part of the announcement that actually matters.
The project, which Chevron has internally code-named "Kilby," is being developed with Engine No. 1, the activist investment firm. Chevron's subsidiary Energy Forge One signed the 20-year power purchase agreement. Generation will be anchored by GE Vernova turbines, the GE power-equipment business that was spun off in 2024, with additional capacity from Solar Turbines, a Caterpillar subsidiary. The phased build-out targets 2.67 gigawatts.
A standard long-term power deal, the kind utilities sign with wind and solar developers every month, runs through the public grid. The generator sells electricity to the local utility, which then delivers it to end customers. That means the project has to clear an interconnection queue, get studied by the grid operator, pay for transmission upgrades, and get folded into long-term planning. If the utility's ratepayers are on the hook for any of those costs, they can show up at a hearing.
Kilby skips all of that. Because the plant is co-located—meaning it physically sits next to the data center and serves only that data center—the electrons never touch the public grid on Microsoft's side. Because the plant is co-located and serves only Microsoft, ERCOT—the operator of the Texas grid—may not need to approve the build under current rules, a point that regulators and legislators are still examining. Local ratepayers do not pay for it. And because there is no rate base, there is no rate case.
Chevron's announcement positions Kilby as "among the largest co-located natural gas power and data center developments in the United States." The press release frames the project as a way to monetize "America's natural gas advantage" and meet "the growing electricity needs of the AI economy." Those are corporate talking points, not independent technical findings. The phrasing also elides what the structure actually does, which is shift most of the planning risk and the externality cost out of the regulatory perimeter that a normal grid-connected plant would have to clear.
What the press release does not say matters more than what it does. The Final Investment Decision has not been made. The capital split between Chevron, Microsoft, and Engine No. 1 is not public. The offtake price is not public. There is no mention of on-site carbon capture, no retirement schedule for the gas turbines, and no commitment to replace generation with clean energy at any point over the 20-year tenor. A Hacker News thread on the announcement surfaced exactly that gap: readers quickly noted the absence of any carbon accounting, water-use disclosure, or grid-impact analysis, and asked how a multi-gigawatt gas commitment could be made with so little public scrutiny.
The co-location pattern is not unique to Chevron and Microsoft. Talen Energy signed a similar arrangement with Amazon Web Services at its Susquehanna nuclear plant in Pennsylvania. Meta, Google, and others have all pushed in the same direction. What Kilby adds to the pattern is scale. At 2.67 gigawatts, Kilby is roughly twice the capacity of a large nuclear reactor, built as a phased gas plant. The structural choice is also more aggressive. The Susquehanna deal uses existing nuclear generation. Kilby is a greenfield gas build designed from the start to never touch the public grid.
That is the part regulators, if they choose to, will eventually have to address. Co-located data center power deals are not illegal. They are, however, a category that lets the largest single new loads on the U.S. grid avoid the planning process that the rest of the grid has to follow. In Texas, where ERCOT is already struggling to keep up with industrial demand growth, the result is that hyperscaler AI power can land without showing up in any interconnection study, any transmission plan, or any rate case. Other generators, including renewables, do not get the same courtesy.
It is worth saying out loud what Kilby also is not. It is not a guarantee that the deal closes. Whether Engine No. 1 has previously developed a multi-gigawatt gas plant of this scale is not clear from available sources, and Chevron's Final Investment Decision is still ahead. It is not a public ratepayer win. Even though Texas's grid is benefiting economically from the data center boom, none of the cost recovery flows through residential or commercial rates on this contract. And it is not a clean-energy answer. The plant is gas, the fuel will largely come from the Permian Basin, and the 20-year tenor exceeds the typical economic life of a wind or solar project built today.
What to watch next: the Final Investment Decision, which Chevron has not yet scheduled. The capacity split between GE Vernova and Solar Turbines, which determines how much of the equipment bill flows to which manufacturer. Any disclosure from ERCOT about whether co-located deals of this size will be folded into the grid plan retroactively, or whether they will continue to be invisible. And any signal from the Texas legislature or the Public Utility Commission of Texas about whether behind-the-meter generation at this scale will eventually need a planning review, even if it does not need an interconnection study.
The Kilby deal is real, and the scale is real. What is still being negotiated is whether a power procurement model that large can keep running entirely outside the public planning process for twenty years. Chevron's announcement is the first move of that negotiation, not the last.