Fireworks, a San Mateo cloud startup that hosts open-source AI models for developers (inference cloud; also expanding into training GPUs), raised 17.5 billion valuation, announced Thursday, July 16, 2026. The Nvidia-backed company said it exceeded $1 billion in annualized revenue — five times last year’s figure.
Round details: Series D jointly led by Atreides Management, Index Ventures, and TCV. Participants include Nvidia, Evantic, and Lightspeed Venture Partners.
Business:
- Competes with Amazon and Google for hosting open models developers embed in apps; also partners with Microsoft Foundry (customers can draw models through Fireworks; compute from 20+ suppliers including Microsoft).
- Hosts Chinese open models (DeepSeek, MiniMax, Z.ai) and OpenAI open-weight releases; pitch is “specialized intelligence” via customer data fine-tuning vs frontier labs’ “generalized intelligence.”
- CEO Lin Qiao (former Meta director; co-founded 2022 with six others): “super-linear demand”; cost vs equivalent closed models “five to 10 times cheaper.”
- Handles ~40 trillion AI tokens/day (company claim). Comparison: Google ~27T/day (May disclosure), OpenAI ~22T/day (March) for developer tools.
- ~200 employees; target 600 by end of 2026. Hired former Salesforce exec George Hu as president in April; building sales team after self-serve era. Proceeds for more GPUs and hiring.
- Clients: Elastic, GitLab, MongoDB. As of last year ~half of revenue from Cursor (Composer custom model); now “much more diversified.” Note: SpaceX agreed in June to acquire Cursor in a $60B stock deal closing this quarter.
Market framing: Evidence that companies want open-source alternatives amid rising closed-model costs; inference-cloud peers include Baseten and Together AI; training/neocloud peers include CoreWeave, Lambda, Nebius.