Fluidstack shifts to dedicated GPUs

Private Cloud marks the point where Fluidstack stopped acting mainly like a GPU broker and started acting like a compute utility for very large buyers. Instead of helping startups find spare capacity, it buys and operates clusters itself, places them in colocation facilities, and signs 2 to 3 year contracts with 25% to 50% paid up front. That shifts revenue toward $100M plus deals, much higher gross margins, and customers that need reserved capacity for long training runs.

• The split is already visible in the numbers. Private Cloud was about 62% of revenue, or $112M ARR, by the end of 2024, versus 38% for the marketplace business, and carried roughly 85% gross margins versus 13% for the marketplace.
  1 sacra 2 sacra
• The product difference is concrete. Marketplace customers rent GPUs through a console or API from third party capacity. Private Cloud customers get Fluidstack owned clusters, preconfigured software environments, and dedicated hardware that is not shared with other tenants.
  2 sacra 3 sacra
• This is the same ladder used across the GPU cloud market. Fluidstack and Together AI start with speed and developer friendliness for startups, while CoreWeave and Crusoe are built around long term capacity commitments for the biggest AI labs and platforms. Private Cloud lets Fluidstack move up that ladder without losing early customers.
  2 sacra 4 sacra

The next step is more of this model, not less. As Fluidstack adds financing against its GPU base and turns dedicated clusters into national scale infrastructure in Europe, Private Cloud becomes the bridge from startup rentals to sovereign and hyperscale compute contracts, which is where the biggest and stickiest spend in AI infrastructure is heading.