Fireworks closed a $1.505 billion Series D on July 16 at a $17.5 billion valuation, led by Atreides Management, Index Ventures, and TCV, with NVIDIA returning. The San Mateo inference platform, founded by ex-Meta engineers who built PyTorch, is now serving more than 40 trillion tokens a day, up 5x year-over-year and nearly triple its previous daily volume. Annualized revenue has crossed $1 billion. TNW notes the developer traffic exceeds what Google or OpenAI report serving.
The interesting number isn’t the valuation. It’s 95%. More than 95% of the tokens Fireworks serves come from models specialized on customers’ proprietary data, not off-the-shelf frontier models. That figure is the entire investment thesis compressed into a percentage.
CEO Lin Qiao makes the argument directly. “There are two paths forward for AI. In one, intelligence belongs to a few big labs, and everyone else rents it. In the other, every company in the world builds specialized intelligence of its own, shaped by the domain only it understands. We are building towards the second.”
It’s a positioning move as much as a technical one. The last two years of AI discourse have treated frontier-model access as the ballgame, with hyperscalers as landlords and everyone else paying rent. Fireworks is betting the rental model peaks early. Its pricing claim (a fifth to a tenth of closed-frontier costs) is the wedge; specialization is the lock-in.
The customer roster reads like validation of the thesis at both ends of the market. Cursor and Harvey run tuned models on the platform; Doximity, Revolut, Uber, and Shopify sit further up the enterprise stack. Downstream, a layer of specialization platforms including Glean, Dust, and LemonLime is packaging the same idea for teams that don’t want to touch inference infrastructure themselves.
Fireworks plans to grow from 200 to 600 employees by year-end. That’s the tell about how the company reads its own moment. You don’t triple headcount inside six months if you think the specialized-inference window is wide. You do it if you think the window closes fast and the incumbents are still pricing yesterday’s assumption that everyone rents.