Fluidstack pivoting to financed dedicated compute

Fluidstack is turning cheap spot GPU brokerage into a funnel for utility like infrastructure contracts. The marketplace side gets startups running fast by reselling third party capacity with small, flexible deals, then the best customers move into Private Cloud, where Fluidstack controls dedicated clusters, signs 2 to 3 year contracts, often collects 25% to 50% up front, and earns much higher gross margins on much larger commitments.

• The marketplace business is basically a matching layer. Fluidstack finds idle or newly deployed GPUs inside partner data centers, pre configures them for training and inference, and sells that capacity through smaller contracts averaging about $340K. That makes it an acquisition channel more than the long term profit engine.
  1 sacra 2 sacra
• Private Cloud is a different product and a different customer. Instead of brokering someone else’s GPUs, Fluidstack places its own GPU clusters in colocated facilities for large buyers like Meta and sovereign projects, driving about $112M of the $180M ARR base in 2024 with roughly 85% gross margins and $100M plus contract values.
  1 sacra 2 sacra 3 sacra
• This is the same broad playbook seen at Voltage Park and other GPU clouds, where on demand usage lowers the barrier to entry and dedicated capacity creates the real economics. What stands out is how far Fluidstack has already shifted mix toward dedicated infrastructure, and how that is reinforced by long dated hosting agreements with TeraWulf that start in 2026.
  2 sacra 4 sacra 5 terawulf

The next phase is a continued mix shift away from brokerage and toward financed, dedicated compute. As more revenue is tied to multi year infrastructure contracts, Fluidstack starts to look less like a marketplace and more like a regional AI utility, with higher margins, larger customers, and deeper dependence on financing, power, and long term site buildouts.