Fireworks raises $1.5B at $17.5B valuation

Fireworks AI closed a Series D round of $1.505 billion, putting its valuation at $17.5 billion. Atreides Management, Index Ventures and TCV led the round; Bessemer, Menlo Ventures, Lightspeed, NVIDIA and 20VC also put in money. The company says it has crossed $1 billion in annualized revenue run rate, up roughly 5x year over year, and now serves more than 40 trillion tokens a day.

Fireworks, founded by engineers who built parts of PyTorch, sells infrastructure for running open and custom models rather than a single house model. The company says more than 95% of the tokens it serves now come from models specialized on a customer’s own data and tuned for a specific job, not general-purpose calls to a stock frontier model. Customers named in the announcement include Doximity, Revolut and Shopify, each running models trained or tuned on their own workflows and data.

What it means for operators

A vendor selling infrastructure, not a model, just posted the kind of revenue growth normally associated with a frontier lab, and its own numbers say the growth is coming from specialization rather than general chat traffic. That is a useful data point if your team has been treating “fine-tune a smaller model for this one workflow” as a future optimization rather than something worth doing now: at this point it is where a meaningful share of production AI spend is already going. It also adds another well-funded, model-agnostic option alongside the Grok on Amazon Bedrock and Amazon Bedrock Managed Knowledge Base integrations if you’re comparing where to run specialized workloads instead of defaulting to a single provider’s stack.