Crusoe sells troubled Wyoming AI data center project to Google

Crusoe, the Denver-based AI data center developer, has sold its unfinished data center project in southeastern Wyoming to Google, the company confirmed on Friday. Google did not disclose how much it paid, but it announced plans to invest $17 billion to develop the campus. The project was once planned to reach 10 gigawatts of power capacity.

The sale moves one of the largest announced AI compute sites from an independent developer to one of the biggest technology companies. TechCrunch describes the project as troubled. Google, which Crusoe had tried and failed to sign as a customer for the site, is taking over development after reportedly raising concerns about the project’s cost and construction timeline under Crusoe’s management. Market news service LiveSquawk summed up the deal the same way: Google buys Crusoe’s Wyoming data center site, with no terms disclosed.

What Google is buying from Crusoe in Wyoming

The asset is an unfinished data center project, not a running facility. Crusoe announced it last year with Tallgrass, an infrastructure company backed by Blackstone, and planned to scale it over time to 10 gigawatts of power capacity. Power capacity measures how much electricity a site can draw. For AI data centers it is the usual way to describe size, because power, more than floor space, limits how many GPUs (the graphics processors used to train and run AI models) a site can hold.

Google’s own figure is the $17 billion it says it will invest in developing the campus. The purchase price paid to Crusoe stays private.

Detail	What the sources say
Seller	Crusoe, an eight-year-old, Denver-based data center developer
Buyer	Google
Location	Southeastern Wyoming
Status	Unfinished project
Original partner	Tallgrass, backed by Blackstone
Planned scale	Up to 10 gigawatts of power capacity
Google’s planned investment	$17 billion to develop the campus
Purchase price	Not disclosed

Why the Wyoming project ran into trouble

Neither company has given a detailed public account of what went wrong. The clearest description comes from TechCrunch, which reports that Google raised concerns about the project’s cost and construction timeline under Crusoe’s management. That concern is attributed to earlier reporting, not to a statement from either company. Bloomberg first reported the planned sale in June.

The shape of the deal is unusual. Google started as a prospective customer that Crusoe could not sign for this site. It ends as the owner and developer. Under the common model, a developer like Crusoe builds and runs the facility and a large tenant leases the capacity. In Wyoming, the would-be tenant has instead taken the project in-house.

TechCrunch frames the deal as a sign of a wider problem. Independent data center developers can find it hard to deliver very large projects on time and within budget, given the difficulty of securing equipment and power. Growing opposition from local communities can complicate development further. That last point ties into a broader pattern of local scrutiny of AI data centers, covered in Unicorn API’s report that OpenAI drops secrecy deals with local officials for its data centers.

Google and Crusoe are still working together in Texas

The sale does not end the relationship. Despite taking over the Wyoming facility, Google has not walked away from Crusoe entirely: Crusoe is developing another data center for Google near Amarillo, Texas. The sources do not give the size or timeline of the Amarillo project.

For Google, the Wyoming campus adds to an infrastructure push behind a company that, in TechCrunch’s description, has repositioned itself as AI-focused across Search, Android, YouTube and its internal structure, while rolling out its Gemini model in more countries and use cases. The same coverage has tracked Google’s moves deeper into its own data center hardware, including Axion, which Google announced in 2024 as its first custom Arm-based data center processor. Building or owning sites directly gives a company of Google’s size more control over when capacity arrives, though the sources do not say this was Google’s stated reason for the purchase.

Crusoe’s business after the Wyoming setback

The Wyoming problems did not stop Crusoe from raising money. The company raised a $3.9 billion round in September and was recently valued at $30.9 billion. Alongside that round it added three board members, including JB Straubel, the Redwood Materials founder and Tesla board member who invested in Crusoe in 2021. Crusoe later became the first customer of Straubel’s energy-storage business.

Crusoe earns money in three ways, according to TechCrunch:

• Colocation: leasing data center space to customers that bring their own GPUs.
• GPU rental: renting out its own GPUs.
• Inference: selling compute used to run AI models, as opposed to training them.

Its largest projects serve some of the best-known names in AI. The campus in Abilene, Texas, is being built in phases for Oracle and OpenAI. Crusoe is also building a separate 900-megawatt campus next door for Microsoft, with its first building expected to come online in mid-2027. Meta is a customer too. Bloomberg reported that trading firm Jane Street signed a $13 billion, five-year cloud deal with Crusoe.

The scale comparison shows how ambitious the Wyoming plan was. The Microsoft campus in Abilene is 900 megawatts. The Wyoming target of 10 gigawatts is 10,000 megawatts, more than ten times that figure.

Crusoe was founded by Chase Lochmiller, now its CEO, and Cully Cavness. It began as a cryptocurrency-mining business powered by natural gas that would otherwise have been burned off. That origin in building compute near cheap or stranded energy still shapes the business. The Wyoming sale shows the limits of that model at the largest scale.

What the sale means for teams building on AI models

Most developers never deal with a data center developer directly. They reach models through APIs and cloud providers. But the capacity behind those APIs comes from projects like this one, and delays at the multi-gigawatt scale affect how quickly new compute comes online for everyone downstream. Several points from the reporting matter for builders:

• Announced capacity is not delivered capacity. The Wyoming project was announced with a 10-gigawatt goal, yet it changed hands unfinished after reported cost and timeline concerns. Headline gigawatt figures are plans, and plans can slip.
• Power and equipment are the bottlenecks. TechCrunch names securing equipment and power as the main difficulties for independent developers. Those same constraints shape GPU availability and inference capacity at the providers developers rely on.
• Local politics is now part of the build. Community opposition can complicate development, adding a non-technical risk to compute roadmaps.
• Independent providers still matter. Crusoe still sells inference and GPU rental, and it remains a supplier to Google, Microsoft, Oracle, OpenAI and Meta. Losing one project has not removed it from the market.

The deal also shows concentration at work. When an independent developer struggles with a site this large, one of the biggest buyers of compute can step in and take it over. Over time, that pattern could tilt very large AI infrastructure toward the largest tech companies. The sources raise the difficulty for independents but do not predict that outcome.

Open questions about the Wyoming campus

Much about the deal is still unknown from the available reporting:

• Price. Google has not disclosed what it paid Crusoe.
• Scale under Google. The 10-gigawatt figure was Crusoe and Tallgrass’s plan. The sources do not say whether Google will keep that target.
• Tallgrass’s role. Tallgrass co-announced the project, but the sources do not say what role, if any, it keeps after the sale.
• Power supply. A reply to LiveSquawk’s post on X, from an investment account, claims the campus pairs the site with a planned power system of about 702 megawatts using fuel cells and carbon capture, involving Bloom, AEP and Tallgrass. That is an unverified social media claim. Neither the companies nor TechCrunch confirm it in the available sources.
• Timeline. No completion date for the Wyoming campus appears in the sources.

What to watch next

The first thing to watch is whether Google gives more detail on its $17 billion plan: target power capacity, construction phases and when the first buildings come online. On Crusoe’s side, the near-term tests are its other large builds, especially the 900-megawatt Microsoft campus in Abilene, where the first building is expected to come online in mid-2027, and the Amarillo project for Google. Delivering those on schedule would go a long way toward answering the questions the Wyoming sale raises about independent developers at gigawatt scale.

More broadly, the reporting points to equipment, power and local opposition as the constraints on AI compute. How quickly those constraints ease, and who ends up owning the largest sites, will shape how much capacity reaches model providers, and through them, the developers building on AI APIs.