Google Is Building Its Next AI Campus on West Texas Wind

Crusoe revealed Monday that it is the developer of Google's massive data center campus under construction outside Amarillo, Texas — a project engineered from the ground up around West Texas wind power, with 5,500 workers on site daily and a cooling system designed to leave the local water supply almost untouched.

Published September 29, 2026 · By Daniel Rao

Wind turbines over West Texas farmland. Google's new Armstrong County campus draws power directly from the nearby Goodnight wind farms. Image: Matthew T Rader / CC BY-SA 4.0.

The most important detail of Google's newest AI data center is not its size. It is where it plugs in. On Monday, Crusoe — the company that calls itself "the AI factory company" — announced that it is the developer of Google's data center campus under construction in Armstrong County, outside Amarillo, Texas. The campus, which broke ground in June 2025, is being built directly at the source of West Texas wind energy, with a power arrangement designed to run one of the world's most energy-hungry computing sites on one of the country's most abundant clean-energy resources.

This is the energy-first strategy that the entire AI industry is now converging on, made concrete. Rather than building a data center and hoping the grid delivers, Crusoe is building the data center where the electrons already live.

The project: Project Llano

The Armstrong County campus — known in development filings as Project Llano — sits on roughly 1,100 acres outside Amarillo. According to Crusoe's Monday announcement, more than 5,500 skilled laborers are on site daily, and the company says it is building "at the same swift pace we set in Abilene," referring to its earlier Texas project. The buildout is expected to reach six buildings, with construction-industry trackers estimating an investment north of $2 billion and completion targeted for the end of 2027.

For Google, the campus is one piece of a far larger Texas play. The project forms part of Google's reported $40 billion investment program across the state to build out its AI infrastructure, per construction-industry reporting. Texas has become the de facto capital of the American AI buildout — cheap land, cheap power, a permissive permitting culture, and an ERCOT grid that lets developers contract directly with generators.

"Building the future of AI requires a new kind of infrastructure, starting with energy," Crusoe CEO and co-founder Chase Lochmiller said in Monday's announcement. "Crusoe's strategy is to build AI factories directly at the source of abundant, low-cost energy — to both improve utilization of existing assets as well as catalyze new energy generation."

Wind as the power plant

The energy arrangement is the story. The campus has a direct connection to Serena's adjacent Goodnight 1 wind farm, a 265.5-megawatt facility that is already operational, and to Goodnight 2, another 265.5-megawatt wind farm currently under construction — a project Crusoe says was actually catalyzed by its development at the site. In other words, the data center is not just consuming existing wind power; it is the economic reason new wind power got built.

The surplus arrangement runs in reverse when the wind is strong: when generation exceeds the campus's demand, the excess flows back into the grid. Crusoe says that benefits local ratepayers and reduces curtailment — the wasteful practice of shutting down West Texas turbines when the grid cannot absorb their output. West Texas wind is famously abundant and famously stranded; Armstrong County is the model for unstranding it with AI.

This is the mirror image of the rest of the industry's power strategy. While KKR's Helix Digital is assembling billions to finance new baseload generation, and Meta is pouring hundreds of billions into domestic infrastructure, Crusoe's pitch is locational: put the data center where the clean megawatts already spin, and the economics work without new gas turbines.

The water question — answered in advance

The other local flashpoint for data centers is water, and Crusoe moved early to neutralize it. The campus will use a closed-loop, non-evaporative liquid cooling system, which the company says is designed to limit the facility's water footprint to primarily domestic uses — "such as kitchens," in the announcement's striking phrase.

That is a deliberate contrast with the industry's water problems. Data centers in drought-prone regions have drawn intense local opposition over evaporative cooling's water consumption, and several communities have pushed back on projects that threatened municipal supplies. A closed-loop system recirculates the same coolant instead of evaporating fresh water into the air — more expensive to build, far cheaper politically. In West Texas, where water is as scarce as wind is abundant, it was likely non-negotiable.

Related: Oracle's $18 Billion New Mexico AI Data Center Hits the Wall — our report on what happens when an AI campus goes wrong: force majeure, an 89-cent bond, and a WSJ investigation.

Why Texas, why now

Armstrong County did not get a Google campus by accident. West Texas combines three things the AI buildout desperately needs: world-class wind and solar resources, vast tracts of available land, and an electricity market — ERCOT — that operates independently of the federal interstate grid and lets large buyers contract directly with generators. The result is the fastest path in America from a signed deal to flowing electrons.

The competitive dynamics matter. Oracle's $18 billion New Mexico project — just across the state line — has collapsed into a force-majeure dispute with its bondholders trading at 89 cents on the dollar. xAI and others are racing to secure Texas sites. Microsoft, Amazon, and Meta are all deep into Texas expansions. In that market, the companies that can show up with power already contracted win, and Crusoe's model — developer, power broker, and builder in one — is designed to win exactly that race.

The local economics are the other half of the story. A project of this scale is expected to generate roughly $28 million a year in local tax revenue once complete, per construction-industry estimates — transformative money for a county whose entire population is smaller than a single shift on the construction site. The campus is expected to create between 50 and 70 permanent operating jobs, modest for the investment but significant for the Panhandle. The politics of the AI buildout in rural America run on these numbers: the turbines are invisible to voters; the tax revenue is not.

Our analysis: the two futures of AI infrastructure

What follows is analysis. The AI infrastructure story of 2026 is really two competing theories. Theory one, the Helix model: assemble enough capital to finance new power generation and build it alongside the data centers. Theory two, the Crusoe model: find where clean power already exists — or is already planned — and build the data center next to it, letting the campus itself catalyze the next wind farm.

Armstrong County is the strongest evidence yet that theory two works at hyperscale. Goodnight 2 would not exist without Crusoe's campus — that is catalyzed generation, not contracted generation, and it is the distinction the industry's critics keep asking for. When a data center causes new clean energy to be built rather than diverting existing supply from other users, the "AI is straining the grid" objection loses its teeth.

The caution is in the framing. Crusoe's announcement is a company press release, and the figures — 5,500 workers, minimal water use, ratepayer benefits — are Crusoe's. The closed-loop cooling claim is verifiable in engineering terms but not yet in operating terms; the campus does not open until 2027. And the Goodnight 2 catalysis narrative, while plausible, runs through the same developer telling the story.

Still, the direction is unmistakable. The industry has spent two years learning that the AI race is an energy race, and Monday's announcement shows the learning is now structural: data centers designed around the wind farm, not the grid connection. In the contest between theory one and theory two, West Texas just placed a very large bet.

Sources: Crusoe press release via GlobeNewswire (September 28, 2026); Unite.AI (September 28, 2026); Texas Construction Hub via LinkedIn (Project Llano details — acreage, investment estimate, timeline, tax revenue — attributed as industry estimates); construction-sector reporting on Google's Texas investment program.