Crusoe Spark Liquid is a prefabricated, liquid-cooled modular data center for Crusoe Cloud customers running NVIDIA Vera Rubin NVL72 and GB300 NVL72 racks. It is the sixth generation of Crusoe Spark, built at the company's factory in Brighton, Colorado and shipped ready to deploy, which Crusoe says cuts on-site construction from years to months. Each module is rated for up to 3.5 MW of compute, five times the density of the air-cooled Spark, using closed-loop direct liquid cooling that does not consume or evaporate water after the system is filled. Production of the liquid modules is planned for 2027, as Vera Rubin rack-scale systems (72 Rubin GPUs and 36 Vera CPUs per rack) move into deployment. The design is part of NVIDIA's DSX POD ecosystem, a reference architecture for modular Vera Rubin sites. Founded in Denver in 2018 by Chase Lochmiller and Cully Cavness, Crusoe manufactures the modules, builds data centers, and offers GPU capacity on Crusoe Cloud. Spark Liquid is for adding capacity at new or existing sites, for inference placed closer to users, and for compute in countries that require data to stay local. Air-cooled Spark modules are already in production use. Ollama runs dedicated workloads on them in the United States, including modules powered by solar and second-life battery storage, to host selected open models. The same launch opened private preview for two Crusoe Cloud services, Managed RL and Managed MLflow, next to managed inference and serverless fine-tuning that are already generally available. Managed RL lets a team adapt an open model with its own data, reward function, and Python training loop, including a Tinker-compatible API, while Crusoe runs the GPUs, sharding, and serving. Managed MLflow hosts the open-source MLflow tools beside those GPUs for experiment tracking, agent traces, evaluations, and routing across model versions, so teams do not have to run their own tracking server or database.