Microsoft West Texas datacenter expansion (Pecos + Project Kilby gas deal)

The most important word in the Pecos announcement is not "gigawatt" — it's "co-located." Project Kilby sits on the same ground as the data center it feeds, wired straight into the racks rather than into ERCOT, the public grid that serves the rest of Texas. Microsoft's own framing is blunt: this demand does not take from the current grid ^[1]. That is the entire point. A behind-the-meter arrangement means the gas plant and the data center are effectively one private power island; the utility never sells the electrons, the regulator never balances them against household demand, and the multi-year interconnection queue that has become the chokepoint for grid-connected AI buildouts simply doesn't apply.

Why this matters: the binding constraint on AI right now is not chips or capital, it is megawatts you can actually energize on a timeline. By pairing roughly 2 GW of compute with a dedicated plant ramping to 2.67 GW ^[2], Microsoft converts a queuing problem into a construction problem — slower physics, but a problem money and turbines can solve. The trade-off, which critics flag, is oversight: siting generation behind the meter lets tech companies bypass the public grid and may face less regulatory scrutiny than a utility-scale plant would ^[9]. The grid stays untouched, but so do some of the rules that normally come with building a power station this size.

For Chevron, Project Kilby is a category change, not just another project. An oil major is becoming, in effect, an AI power utility — taking its own Permian Basin gas and its own infrastructure and converting them into a 20-year contracted cash-flow stream that is independent of the oil price cycle ^[4]. Chevron New Energies president Jeff Gustavson is explicit that this is a land grab among peers: "In our peer group a lot of others are talking about doing things like this. We're now actually doing it," he said, leaning on the basin itself as the moat — "This is the most abundant gas basin in the country, maybe the world" ^[2]. The structure underlines the seriousness: an estimated $7 billion plant spanning more than 2,000 acres ^[2][3], with investment fund Engine No. 1 holding an option to buy a 50% stake ^[2], GE Vernova supplying the bulk of the turbines and Solar Turbines filling in the rest ^[4].

It fits a 2025-2026 pattern of fossil incumbents pivoting toward AI load. ExxonMobil had already entered advanced talks to power data centers with gas plus carbon capture ^[5], and a wave of on-site gas plants — Meta's Hyperion alone reached 7.46 GW — was already underway by spring ^[6]. Chevron's twist is vertical integration: it owns the molecule, the pipe, and now the power contract. The catch is execution. Analysts note that if Chevron misses the 2026 final investment decision or the 2028 first-power date, or if costs overrun, the promised mid-teen returns could collapse into a low-return infrastructure project ^[7]— turning a strategic pivot into an expensive distraction.

Microsoft built its messaging around water, and the line it wants quoted is striking: the total lifecycle water use of this datacenter is only a fraction of that consumed annually by a typical fast-food restaurant ^[1], thanks to closed-loop cooling that recirculates water rather than evaporating it away. That claim lands in a charged context. Texas data centers are projected to consume up to 399 billion gallons of water by 2030, up from 49 billion in 2025 ^[3][8]— so a genuinely low-water design is a real differentiator in a region where the resource is already strained.

The harder number is carbon. Burning gas at this scale is, by definition, a new emissions source, and Kilby relies on Selective Catalytic Reduction controls rather than the carbon capture some peers tout ^[1]. Sector-wide, eleven gas-fired AI projects could emit over 129 million tons of greenhouse gases annually if run at full capacity ^[6]. Microsoft offsets the optics by noting it has contracted 4.7 GW of renewable electricity for its Texas operations ^[1], but that power flows to the grid, not into the gas-fed Pecos island — a clean-energy story running on a parallel track to the dirty-power reality of this specific campus. Community reception captured this split screen: pragmatists, including some in energy-focused forums, treat co-located gas as the responsible way to add load without taxing the grid, while environmental and quality-of-life critics push back on emissions, water, and the prospect of on-site turbines as a permanent new noise source.

Geography is doing a lot of work here. Reeves County offers cheap, abundant gas at the wellhead, vast open acreage for a 2,000-plus-acre plant, and a state that leads the nation with roughly 33 GW of planned data-center projects against a US total projected to double to 77 GW by 2030 ^[2]. The timing is a direct response to AI/cloud electricity demand outpacing grid supply: behind-the-meter gas is how Microsoft adds capacity at the speed AI roadmaps demand rather than the speed utility interconnection allows ^[1][2].

The local case is built on dollars and jobs. The data center is pitched as bringing over 6,000 construction jobs at peak and hundreds of permanent Microsoft roles ^[1], while the Kilby plant alone is projected to generate more than $10 billion in state and local tax revenue and support nearly 2,000 jobs ^[3][4]. That is why Reeves County Judge Leo Hung's "We are excited to welcome Microsoft to Pecos" ^[1]reads less as boilerplate and more as a small Permian community betting that the AI buildout, and the gas that powers it, will outlast the next oil downturn.