Microsoft's 2GW AI Campus Comes With Its Own Gas Plant

Microsoft is building a 2GW AI campus in Pecos, Texas — so power-hungry it's constructing its own Chevron-turbine gas plant rather than rely on the public grid.

4 min readEAEvgenii ArsentevEvgenii Arsentev · PhD

Microsoft is building a roughly 2-gigawatt data center campus in Pecos, Texas — and because the public power grid can't supply that much electricity reliably, the company is funding its own on-site gas turbine plant using Chevron hardware. The campus is described as one of the largest single capacity additions in Microsoft's history, and it's a clear signal of what frontier AI infrastructure demands are doing to energy planning.

Construction will take five to seven years, putting the target operational date around 2028. At peak activity, the project will create more than 6,000 construction jobs, with hundreds of permanent positions once the facility is running. The total investment hasn't been disclosed beyond 'multibillion-dollar.'

Why Microsoft stopped waiting for the grid

The context matters here. Dozens of US data center projects were cancelled in 2026 as communities pushed back over rising electricity bills and water consumption concerns. To get ahead of that opposition, Microsoft published an open letter to Pecos residents before breaking ground. It promises the campus won't raise local electricity prices, will use closed-loop cooling that the company claims consumes 'only a fraction' of the water used annually by a typical fast-food restaurant, and will return more water than it takes from local sources.

The private power generation strategy is a direct answer to grid constraints. Texas's ERCOT grid handles the country's highest electrical demand and has a history of capacity crunches; building generation on-site removes that bottleneck entirely. This approach is spreading across the industry: as AI model training and inference demand explodes, tech companies are increasingly treating power the same way they treat servers — building it custom, in-house, and at scale, rather than waiting for utilities to catch up.

What I'd actually do

For anyone building AI products, this is worth watching for one practical reason: massive compute buildouts like this typically translate into lower costs for running AI two to three years after capacity comes online. The enormous bet on AI infrastructure today is, in effect, a bet that AI gets cheaper for everyone building with it tomorrow. That's relevant if you're pricing a product or deciding which AI features are viable right now.

#microsoft#ai-infrastructure#data-center#energy

Related guides

EAEvgenii Arsentev

Author

Evgenii Arsentev

PhD · Chief Product Officer at a tech company

Want to actually build this?

Guides explain. The free course transforms — personalized, gamified, and built to get you shipping fast.

◉ Start the free course

Source: the-decoder.com