Key facts

Founded
—
HQ
Redwood City, United States
Employees
—
Funding stage
Series D · US$1.505bn
Further facts
  • Funding: Atreides, Index Ventures and TCV led a US$1.505bn Series D at a US$17.5bn valuation in July 2026, with Evantic, Lightspeed and NVIDIA joining.
  • Scale: Fireworks passed US$1bn in annual recurring revenue by July 2026 and serves more than 40 trillion tokens a day.
  • Capital: The Series C in October 2025 raised US$250m at a US$4bn valuation, led by Lightspeed, Index Ventures and Evantic.

Focus areas

LLM inference AI infrastructure Open-weight models Reinforcement learning Model training

About

Fireworks AI runs an inference cloud where companies serve, fine-tune and scale open-weight generative AI models in production. It was founded by engineers from the team that built PyTorch at Meta, led by chief executive Lin Qiao, and is based in Redwood City, California. Its platform hosts more than 200 models across text, image, audio and multimodal formats. Customers named by the company include Uber, Shopify, Doximity and Revolut, and it partners with Microsoft and NVIDIA. In July 2026 it raised a US$1.505bn Series D at a US$17.5bn valuation after annual recurring revenue passed US$1bn.

How they use AI

Fireworks optimises the serving stack for open models so that customers get low-latency inference without running their own GPU fleets. Customers tune models on their own application data with supervised fine-tuning or reinforcement learning, and its Eval Protocol tooling measures results before release. By July 2026, 95% of the tokens it served came from customer-specialised models rather than base models, and daily volume had grown from 15 trillion to more than 40 trillion tokens. The company has SOC 2 Type 2 and ISO 27001, 27701 and 42001 certifications.

Products and services

  • Serverless inference — Pay-per-token access to 200+ open-weight models across text, image, audio and multimodal
  • On-demand deployments — Dedicated GPU deployments for production workloads
  • Fine-tuning — Supervised and reinforcement fine-tuning on customer data
  • Eval Protocol — Open tooling for evaluating model quality during tuning

Notable customers / partners

Key people

  • Lin Qiao — Co-founder and CEO
  • Benny Chen — Co-founder
  • Dmytro Dzhulgakov — Co-founder
  • Chenyu Zhao — Co-founder
  • James Reed — Co-founder

Investors