Fireworks AI raises $1.5B Series D at $17.5B valuation; enterprise AI inference surpasses $1B ARR

Fireworks AI, a San Mateo-based platform enabling enterprises to fine-tune and serve open AI models on proprietary data, closed a $1.505 billion Series D on July 15-16 at a $17.5 billion valuation. Atreides Management, Index Ventures, and TCV led the round; NVIDIA, Lightspeed, Bessemer, Menlo, Insight Partners, Ontario Teachers' Pension Plan, and Lone Pine Capital also participated. Founded in 2022 by Lin Qiao and six former Meta PyTorch engineers, the company announced it has surpassed $1 billion in annualized revenue run rate—a 5x jump year-over-year—and now serves 40+ trillion tokens daily, up from 15 trillion in the same period.

Fireworks' pitch: as open models approach frontier-model performance, enterprises want to own and customize the AI powering their products rather than rent it. The company provides infrastructure to fine-tune open models on customers' data and serve them in production. Critically, 95% of tokens served through Fireworks are from specialized models—not off-the-shelf foundations. The platform offers 200+ models across text, image, and multimodal formats, supports new open releases within hours, and backs customers including Uber, Shopify, Revolut, Harvey, Cursor, Doximity, Elastic, GitLab, and MongoDB. Fireworks claims 5-10x cost advantage over equivalent closed models.

For practitioners building on open-source AI, Fireworks signals a sea change: specialized inference at enterprise scale is now a capital-backed business, not a research bet. NVIDIA's participation is instructive—the chip vendor is hedging by backing the platform reselling its compute. Fireworks plans to triple headcount from 200 to 600 by year-end, expand global compute capacity, and deepen partnerships with Microsoft and NVIDIA. The company last raised $250M at $4B in October 2025, so a 4.4x valuation jump in nine months reflects both scarcity of capital and genuine momentum in the open-model-as-enterprise-infrastructure thesis.