AI inference startups that grew by renting compute from cloud providers and reselling it have begun purchasing their own Nvidia GPUs, with Fal, Together, and Baseten among them, because leased capacity has gotten harder to secure, according to The Information. Some clouds now favor customers with their own end users over resellers, and a person familiar said CoreWeave stopped selling compute to Baseten after launching a rival inference service in March, which someone close to Baseten disputed. Fal is buying chips for leased data centers with part of the $1.65B it raised this month and has told investors it wants to borrow for more. But Fireworks, the most valuable of the group at $17.5B, has told investors that owning hardware is a commodity business and has no plans to buy chips.
