Asset-Light GPU Marketplace Economics

The marketplace is best understood as a customer acquisition layer, not the economic core of the business. Fluidstack finds spare GPUs sitting in third party data centers, bundles them into usable clusters, and sells that capacity to startups that need machines fast, which keeps capital needs low because the company is not buying the hardware. The tradeoff is that a large share of revenue is passed through to infrastructure partners, which is why gross margin is far lower than in private cloud.1 2

• In 2024, the original platform marketplace ran at about 13% gross margin and roughly $340K ACV, versus about 85% gross margin and $100M+ ACV for Private Cloud. That spread shows why the marketplace can scale revenue quickly without becoming the main profit engine.1
• The workflow is closer to a broker than a landlord. Fluidstack brings demand, handles onboarding, and pre configures GPU clusters so AI teams can train and run models immediately, but the underlying capacity belongs to partner facilities. That keeps fixed assets light and speeds supply expansion.1 3
• This model resembles other GPU marketplaces like RunPod and Vast.ai, which aggregate third party supply and compete on price and availability for smaller customers. By contrast, Lambda and CoreWeave have moved upmarket toward dedicated multi year deployments where owning or tightly controlling the fleet supports much better unit economics.2 4 5

Going forward, the marketplace should matter less as a standalone business and more as the front door into larger contracts. As Fluidstack shifts more customers from hourly rentals into dedicated clusters with upfront commitments, the company moves from broker economics toward infrastructure utility economics, while still using the asset light layer to fill the funnel.1 2