Fluidstack shifts revenue to private cloud

Fluidstack is becoming less like a GPU broker and more like a compute owner with contracted revenue. The marketplace business still helps small AI teams find spare capacity fast, but the economic center has moved to dedicated clusters that Fluidstack owns inside colocation sites, where customers sign 2 to 3 year deals, pay part of the contract up front, and generate much higher gross margins and contract sizes.1 2 3

• The two businesses look very different in practice. Marketplace customers rent third party GPUs for shorter term jobs, while Private Cloud customers get dedicated machines reserved for them, which turned a roughly 13% gross margin, $340K ACV business into an roughly 85% gross margin, $100M plus ACV business.1
• This also changes what the marketplace is for. Instead of being the main event, it works as a feeder channel where startups begin with flexible rentals, then move up to dedicated capacity as workloads grow. That is the same path described for customers ranging from smaller AI startups to larger platforms like Meta.1 2
• The mix shift mirrors a broader neocloud pattern. Lambda also moved away from lower commitment products and toward multi year supercluster contracts with anchor customers, showing that scarce GPU supply rewards providers that lock hardware into large predictable deployments rather than reselling it job by job.3

From here, more of Fluidstack's revenue should come from large reserved deployments, not spot style rentals. That points to a business with fewer customers, bigger contracts, more financing tied to owned GPUs, and a clearer path to competing with CoreWeave, Crusoe, and hyperscalers for national and enterprise scale AI infrastructure spend.2 3