Neocloud

FluidStack

Aggregated datacenter capacity brokered into large training clusters at sharp prices.

Aggregates underutilized datacenter GPU supply and packages it into clusters, a broker with an operations layer. Known for landing very large deals (including sovereign AI projects) at prices traditional clouds won't match.

Visit FluidStack ↗

GPU pricing on FluidStack

On-demand list per GPU-hour, last reviewed 2026-09. Methodology.

GPU $/GPU-hr vs cheapest Notes
H100 PCIe $1.85 cheapest
H100 SXM $2.10 1.1×
Gaudi 3 $2.30 1.2× Via IBM Cloud capacity

Strengths

  • Sharp pricing on large clusters
  • Fast large-scale procurement

Watch for

  • Capacity is brokered, diligence the underlying DC
  • Less self-serve than peers

Alternatives

CoreWeave

The flagship neocloud: Kubernetes-native GPU fleets built for training clusters, now public company.

Lambda

The ML researcher's default: one-click clusters, sane pricing, no quota theater.

RunPod

Community and secure cloud tiers, per-second billing, and serverless GPU endpoints.