Fluidstack's Shift to Compute Utility

The real shift is from reselling spare GPUs one booking at a time to owning long lived compute capacity that customers treat like power from a grid. In the marketplace model, Fluidstack matched startups to third party data centers and kept a thin take rate. In Private Cloud, it controls the GPUs, signs 2 to 3 year contracts, collects 25 to 50% up front, and sells dedicated clusters to buyers like Meta and governments, which lifts gross margin from 13% to about 85%.

• The broker business still matters, but mostly as an onramp. Small teams like Character.ai or Poolside can start on short term rentals, then move onto dedicated clusters once their workloads become predictable and large enough to justify $100M plus contracts.
  1 sacra 2 sacra
• This is also a customer quality shift. Startups buy bursty capacity and compare hourly prices. Meta, Mistral, and nation states buy reserved capacity, power, networking, and compliance together, which makes the product look less like cloud spot inventory and more like critical infrastructure.
  2 sacra 3 sacra
• The margin mix shows the destination. Private Cloud was about 62% of revenue, or $112M ARR, by the end of 2024, and is projected to reach 93% by 2027, while the marketplace falls to a minority of revenue. That is the economic signature of a utility model.
  1 sacra 2 sacra 4 sacra

Going forward, the winners in European AI infrastructure are likely to look less like marketplaces and more like financed operators of sovereign compute capacity. If Fluidstack keeps converting developer demand into multi year dedicated deployments, it can become the default utility layer underneath Europe’s model labs, enterprises, and national AI programs.