Fluidstack's GPU cloud pipeline
Fluidstack at $180M ARR
Fluidstack is building a farm system for the GPU cloud market. The important point is not just that startups can begin there, it is that Fluidstack uses small, fast rentals to get into a lab early, then moves that customer onto dedicated clusters once training runs become continuous and spend jumps from hundreds of thousands to nine figures. That turns a commodity resale business into a pipeline for much richer infrastructure contracts.
-
The workflow is concrete. A startup can start by renting available GPUs through Fluidstack’s marketplace style platform, where clusters are preconfigured and ready for training or inference. If usage becomes steady, Fluidstack can place its own GPUs in a colocated facility and dedicate that cluster to one customer under a 2 to 3 year contract with 25% to 50% paid up front.
-
CoreWeave and Crusoe sit further up the curve because they are built around much larger, longer commitments. CoreWeave scaled to about $2B of 2024 revenue on large reserved contracts and used those contracts to finance more GPUs. Crusoe similarly leans on owned infrastructure and cheap power, with a more capital heavy model designed for customers whose demand is large enough to justify custom capacity.
-
That is why well funded labs such as Character.ai, Poolside, and Mistral fit the middle lane. They are too compute hungry to stay forever on ad hoc rentals, but may still want speed, flexibility, and easier onboarding before locking into the scale, procurement process, and infrastructure footprint of a CoreWeave or Crusoe relationship.
Going forward, the winners in GPU cloud will increasingly be defined by which rung of the customer ladder they own. Fluidstack is strongest when it captures teams at the moment demand becomes repeatable, then upgrades them into private clusters before hyperscalers and large neoclouds absorb the account at full industrial scale.