Crusoe has introduced Crusoe Spark Liquid, a modular “AI factory” data center platform designed around high-density, rack-scale AI systems, including NVIDIA GB300 NVL72 and the NVIDIA Vera Rubin NVL72. Crusoe says the goal is to add Crusoe Cloud capacity faster and place it closer to users for lower-latency AI inference.
Crusoe Spark Liquid uses closed-loop, non-evaporative direct liquid cooling (DLC) and is rated to support up to 3.5 MW of compute per modular unit. Crusoe says the units are designed to be deployed in clusters, scaling out for AI training and inference.
The company describes Crusoe Spark as a compact modular data center built in Crusoe’s factory and shipped ready to deploy. For operators, the practical takeaway is the schedule: Crusoe says Spark is intended to compress field build timelines “from years to months,” which can be as important as raw kW capacity when you’re trying to stand up GPU infrastructure on a fixed silicon delivery window.
On cooling and water, Crusoe says the closed-loop, non-evaporative DLC architecture delivers liquid-cooling thermal benefits “without consuming or evaporating water,” with the intent of supporting deployments in water-stressed regions. Crusoe also characterizes Spark Liquid as its sixth generation of modular data centers and says it represents a 5x increase in compute density over previous generations.
Crusoe says Crusoe Spark Liquid is already deployed for Crusoe Cloud customers in production, including Ollama. Crusoe describes Spark Liquid as a vertically integrated system that it builds and operates end to end, tying modular capacity directly to its cloud platform.
Manufacturing is planned in Colorado: Crusoe says Crusoe Spark Liquid will join existing Crusoe Spark production at the company’s Brighton factory in 2027.
Alongside the infrastructure platform, Crusoe also introduced two managed software offerings for its cloud: Crusoe Managed RL (reinforcement learning) and Crusoe Managed MLflow, both in private preview.
“Spark Liquid is Crusoe’s sixth generation of modular data centers, taking the best of everything we’ve learned across hundreds of past modular deployments to define a new standard of excellence for liquid-cooled modular computing infrastructure,” said Cully Cavness, co-founder, president, and chief strategy officer at Crusoe.
For the Ollama deployment, Crusoe says Ollama uses Crusoe Spark infrastructure in the US to host select models on Ollama’s cloud, and that Ollama also runs dedicated workloads on Crusoe Spark modules “powered by renewable solar energy and second-life battery storage.” “They own the full stack, from power to GPUs, across data centers in the U.S. and Europe,” said Michael Meltzer, chief business officer at Ollama.
Source: Crusoe
