Fireworks AI: Inference, Open Models and US$1bn ARR
Fireworks AI Co-founder and CEO Lin Qiao is building an inference cloud where enterprises own specialised open AI models, serving 40tn tokens every day
TL;DR
- Fireworks AI serves over 40 trillion tokens daily, with 95% originating from customer-specialized open AI models.
- The company achieved an annualized revenue run rate exceeding US$1 billion and a US$17.5 billion valuation in July 2026.
- Fireworks AI operates in the inference layer, offering tools for fine-tuning and training specialized AI models for enterprises.
- Customers select open models, fine-tune them with proprietary data, and serve them at scale, paying for performance.
- Notable customers include Uber, Shopify, Notion, and Cursor, with documented improvements in latency and speed.
- The company raised US$1.505 billion in a Series D round led by Atreides Management, Index Ventures, and TCV.
- Funding will expand global compute capacity and deepen relationships with strategic partners NVIDIA and Microsoft.
- Fireworks AI supports over 200 text, image, and multimodal models and quickly integrates new open releases.
- The platform aims to provide core infrastructure for enterprises building their own specialized AI intelligence.