Atreides Management, Mubadala Capital, and Valor Equity Partners co-led the round, which received backing from Crusoe’s new and existing investors, including Founders Fund, Nvidia, and Qatar Investment Authority (QIA).
In the announcement, Crusoe said that it will use the new funding to support the development of its AI factories. However, rather than focusing on large-scale expansion, the company will transition its large-scale vertically integrated campuses to “modular Crusoe Spark units” that support the growth of its Cloud service.
The refocusing shifts what Crusoe is known for, as the developer behind projects including Oracle’s Abilene data-centre campus – which rents out computing capacity to OpenAI. This project boosted Crusoe’s credibility in the AI infrastructure space.
The company went on to expand the Abilene site for Oracle and OpenAI; however, it leased the buildings to Microsoft instead after the pair went with other developers. Chase Lochmiller, Crusoe’s CEO, estimates the site has around 2.1 GW of planned power capacity, including a 900 MW on-site gas plant for Microsoft.
To this end, Crusoe’s strategies thus far have been to develop and lease data centres and rent out its GPUs, and to run its customers’ AI models while charging for tokens.
As the company has historically operated this unorthodox approach to AI infrastructure, it is unsurprising that it would want to reinvent its strategy once again. By investing in its Spark units, Crusoe moves away from large-scale data centres and towards modular, smaller installations.
The Spark units can be built in Crusoe’s factories and delivered by heavy-duty transportation to areas with available power. According to Lochmiller, this will cut construction time down from years to weeks.
These portable units can then be used for inference, as they don’t need access to thousands of chips like training models often do. Lochmiller told The Wall Street Journal: “You don’t actually need an Abilene to do that.” He added that running inference on the same scale as training “can be a bit of overkill.”
Its manufacturing facility in Colorado has the potential to produce up to 1 GW of Spark capacity per year, and Spark units are already operational in Nevada.
Crusoe stated that the goal behind investing in modular units was to meet customers across the AI ecosystem as quickly as the market itself is moving. While large campuses meet the GPU, bandwidth, and power requirements of training models, smaller units can run functions such as inference without waiting years for data centre construction.
Moving in this direction could address several concerns of expanding large-scale AI infrastructure: access to large amounts of electricity, labour requirements, and environmental concerns. It could also help solve the bottlenecks caused by planning and insurance hold-ups.
Crusoe is not the only large firm investing in portable, modular data centres – it joins the ranks of Huawei, Schneider Electric, ABB, and more. With the public spotlight currently shining on large-scale data centres, perhaps the shift to smaller modular setups will alleviate worries concerning the rapid spread of large infrastructure.
RELATED STORIES
Capacity Europe 2026
The 25th anniversary edition of Capacity Europe 2025 will bring together 3,500+ decision-makers from the global connectivity and digital infrastructure community.
