Crusoe, an AI data-center developer known for building one of the largest artificial intelligence supercomputers globally, is shifting its strategy toward smaller, more modular data centers designed for AI inference workloads. After completing a major facility in Abilene, Texas, which services OpenAI through Oracle, the company is now focusing on scalable, factory-built data centers that can be quickly deployed to locations with available power.
The company’s pivot follows a $3.9 billion funding round that values Crusoe at approximately $30.9 billion. The investment was co-led by Atreides Management, Valor Equity Partners, and Mubadala Investment, with participation from Founders Fund, TPG, and sovereign wealth funds including the Qatar Investment Authority, according to Chief Executive Chase Lochmiller.
Crusoe’s approach reflects evolving needs within the AI sector. While large data centers remain essential for training sophisticated models, inference—the process of delivering AI-powered results to users—often requires substantially less computing capacity. “You don’t actually need an Abilene to do that,” Lochmiller said in reference to the large Texas facility. He suggested smaller units provide a more cost-effective and quicker solution for inference workloads.
The company’s new product, branded as Spark, comprises compact centers manufactured at Crusoe’s own facilities near Denver and then transported by truck to sites with sufficient power availability. By avoiding traditional, lengthy construction projects that depend on local labor and regulatory approvals, Crusoe aims to reduce lead times and operational expenses. Some Spark centers already operate in Reno, Nevada, using a combination of repurposed electric vehicle batteries and solar energy.
Founded in 2018, Crusoe initially installed computer hardware at oil and gas sites to harness waste energy, mainly serving cryptocurrency mining customers before shifting focus toward Nvidia GPU-based AI computing. The Abilene campus features roughly 2.1 gigawatts of power capacity, including a 900-megawatt gas plant dedicated to Microsoft’s share after Oracle and OpenAI redirected expansion plans to other developers.
Unlike many AI cloud providers that concentrate primarily on chip rental, Crusoe employs a diversified revenue model encompassing data-center sales, GPU rentals, and charging customers for AI output—referred to as “tokens.” This multi-tiered approach has sparked internal debate, with some board members urging a narrower focus, but Lochmiller maintains it offers insights into shifting demand and helps improve infrastructure design.
The company’s cloud customers include AI startup Perplexity, while its managed inference business has grown rapidly, reaching an annualized revenue run rate of over $100 million by mid-2026. Crusoe anticipates that the majority of this cloud inference workload will be powered by its smaller Spark facilities as production scales toward a potential annual output of one gigawatt of capacity.
Crusoe continues to develop large-scale data-center projects, including a site in Armstrong County, Texas, in partnership with Google, and additional developments in Texas and Missouri. Overall, the company reports having over six gigawatts of contracted data-center capacity.
Amid rising local concerns about the impact of AI data centers, Crusoe has engaged in economic and social initiatives. It estimates its contributions to the Abilene economy exceeded $20 million last year, with tax payments to the local school district expected to surpass the school’s annual budget. In Missouri, Crusoe is sponsoring community events like free admission to high school football games in Warrenton, where it is developing a new data center.