What Fluidstack does
Fluidstack is a neocloud-style AI infrastructure company that builds, owns, and operates large-scale compute capacity for frontier model training and inference. The company positions itself around speed of deployment—aiming to deliver gigawatt-scale compute in about 6 months rather than the industry’s 18–24 months—by acquiring power, designing and building data centers, and operating them end-to-end. Fluidstack describes its main customers as leading AI labs, along with governments and enterprises that need reliable, secure, high-scale GPU compute delivered quickly.
On the software side, Fluidstack’s platform includes managed cluster orchestration layers for both Kubernetes (for training and inference) and Slurm (for training/batch workloads). In its documentation, Fluidstack states that its managed Kubernetes is a bare-metal container orchestration platform designed for frontier model training and inference, and that it supports operational features such as GPU and network operators as well as health checks and observability. It similarly describes “managed Slurm” as a bare-metal batch orchestration platform designed for frontier model training, including user management, Pyxis/Enroot support, topology-aware scheduling, and health checks/observability.
Strategically, Fluidstack has emphasized large committed compute buildouts and marquee partnerships. In late 2025, Fluidstack announced a $50B U.S. computing infrastructure deal with Anthropic to build custom data centers in New York and Texas, describing associated job creation and stating Anthropic selected Fluidstack for “exceptional agility” and rapid delivery of gigawatts of power. The company also announced the relocation/establishment of its global headquarters in New York City in support of U.S. scaling. In parallel, Fluidstack has also highlighted technology and ecosystem partnerships (e.g., with NVIDIA, Borealis Data Center, Dell Technologies; and an alliance with DDN and Mistral AI) as well as an industry ranking/award for GPU cloud performance and operations via SemiAnalysis’s ClusterMAX.