About FluidStack

FluidStack aggregates GPU capacity across partner data centers worldwide and packages it as coherent clusters. Where a marketplace sells you one machine at a time, FluidStack’s business is assembling several hundred or several thousand GPUs with the interconnect and storage to train a frontier model on them.

The customer profile is accordingly narrow: AI labs, well-funded startups, and research institutions with a training run that needs more hardware than they can get on demand from anyone else. Several notable labs have run large jobs on their capacity.

Pricing

On-demand rates for smaller allocations are competitive, with A100 80GB around $1.60/hr and H100 SXM around $2.40/hr. The real product is reserved capacity, priced by contract against term and cluster size, and typically well below on-demand.

Lead times for large clusters are measured in days or weeks rather than minutes. This is procurement, not self-service.

Service Details

Clusters ship with InfiniBand between nodes, high-performance shared storage, and Slurm or Kubernetes for scheduling, depending on what your team uses. Operating systems are Linux throughout, and the environment is configured for distributed training rather than left generic.

Because capacity comes from multiple underlying operators, specifics such as exact location, network topology, and storage backend vary by allocation and are worth pinning down in writing before signing.

For a single GPU this month, look at RunPod or DataCrunch instead. For a thousand H100s for a quarter, this is the category of provider you talk to.

Available in: United States, United Kingdom, France, Iceland, and partner facilities globally

Learn More

For more information about FluidStack, visit their website at https://www.fluidstack.io/

Plan	Price/mo	CPU	RAM	Notes
A100 80GB	—	30 vCPU	240GB	Around $1.60/hr
H100 SXM	—	28 vCPU	250GB	Around $2.40/hr
Reserved cluster	—	—	—	Hundreds to thousands of GPUs, contract pricing