Fireworks AI

Fireworks AI, founded in Redwood City by a team of former Meta engineers who built PyTorch, runs an inference cloud optimized for fast, low-cost serving of open-source and custom models. Led by co-founder and CEO Lin Qiao, the company processes large volumes of tokens daily for thousands of customers including Uber and Shopify. It raised a $250 million Series C in October 2025 at a $4 billion valuation, co-led by Lightspeed, Index Ventures, and Evantic Capital, with NVIDIA and AMD as strategic investors, and reported roughly $280 million in ARR at the close. By early 2026 the company was reported to be in talks for a new round at a ~$15 billion valuation.

It is a leading independent inference platform for open models, with PyTorch-pedigree engineering and one of the steepest revenue ramps in AI infrastructure.

fireworks.ai AI / GPU Cloud

United Statesprivate

At a glance

Total raised

$327M

last round series c

Headcount

51-200

approximate · public range

Headquarters

Redwood City

United States

Founded

2022

4 years active

Status

Private

AI / GPU Cloud

Latest valuation

For subscribers

Subscribe to see the full profile

Including Funding rounds, Bull / Bear thesis, Stock + earnings, Roster changes, Patents, News, and Open roles.

Already subscribed? Sign in →

Lin Qiao

Co-founder & CEO

Lin Qiao earned bachelor's and master's degrees in computer science from Fudan University and a PhD in computer science from UC Santa Barbara. She began her career as a researcher at IBM working on data infrastructure and databases, later worked at LinkedIn, and spent roughly seven years at Meta as a Senior Director of Engineering, where she led the development and deployment of PyTorch across Meta's data centers and devices. In 2022 she co-founded Fireworks AI in Redwood City with six colleagues — many from the PyTorch team — naming the company to honor those roots. She serves as CEO of the inference platform.

Chenyu Zhao

Co-founder

Chenyu Zhao is a co-founder of Fireworks AI, the inference cloud started in 2022 by Lin Qiao and a team largely drawn from Meta's PyTorch engineering organization. The founding team built Fireworks around fast, low-cost serving of open-source and custom models. Additional verifiable detail about Zhao's earlier career could not be confirmed from public sources.

Dmytro Ivchenko

Co-founder

Dmytro Ivchenko is a co-founder of Fireworks AI, founded in 2022 by a group of engineers including former members of Meta's PyTorch team. He is part of the technical founding team behind the company's high-throughput inference engine. Additional verifiable detail about his earlier career could not be confirmed from public sources.

James Reed

Co-founder

James Reed is a co-founder of Fireworks AI, the inference platform launched in 2022 by Lin Qiao and a team with deep PyTorch roots. He is part of the founding engineering team behind the company's model-serving technology. Additional verifiable detail about his earlier career could not be confirmed from public sources.

Pawel Garbacki

Co-founder

Pawel Garbacki is a co-founder of Fireworks AI, which was founded in 2022 by Lin Qiao alongside several engineers from Meta's PyTorch organization. He is part of the technical founding team building the company's inference infrastructure. Additional verifiable detail about his earlier career could not be confirmed from public sources.

Benny Chen

Co-founder

Benny Chen is a co-founder of Fireworks AI, the open-model inference cloud started in 2022 by Lin Qiao and a team largely from Meta's PyTorch group. He is part of the founding team behind the company's serving platform. Additional verifiable detail about his earlier career could not be confirmed from public sources.

AI bull / bear

Bull case

Fireworks AI is running one of the steepest revenue ramps in infrastructure history: roughly $280M ARR at its October 2025 Series C — $250M at a $4B valuation co-led by Lightspeed, Index, and Evantic, with NVIDIA and AMD as strategic investors — climbing to a reported $315M by February 2026, up 416% year over year, on a platform processing north of ten trillion tokens a day. The pedigree explains the throughput: Lin Qiao ran the Meta team that built PyTorch, and the co-founding bench is the engineering core of the framework the entire industry trains on, now applied to a proprietary serving engine that wins on latency and cost at production scale. The customer list is the AI-native elite — Cursor, Perplexity, Notion, Uber, DoorDash, Shopify — the workloads growing fastest in all of software, and the fine-tune-to-endpoint loop makes those relationships stickier than raw token serving: a customer's tuned model lives on Fireworks. Bloomberg's May 2026 report of funding talks at $15B, with Index set to co-lead again, signals insiders doubling down rather than rotating out. As open models close the gap with frontier labs, the company that serves them best captures the arbitrage — and Fireworks is currently that company.

Bear case

A $15B valuation on roughly $315M of ARR is ~48x revenue for a business whose gross margins are structurally suspect: Fireworks rents GPU capacity it does not own and sells serving of models it does not make, the classic mid-stack position squeezed between neocloud suppliers and price-warring peers. Its serving-engine edge decays on a schedule — vLLM and SGLang keep absorbing the same optimizations in open source, and every rival ships them a quarter later at lower price, with DeepInfra undercutting from owned hardware and Baseten freshly armed with $1.5B. The celebrated customer list is the concentration risk: AI-native companies burning venture money, with unproven unit economics of their own, and both the incentive and the talent to vertically integrate — Cursor already trains proprietary models, and any customer at Fireworks' top decile can take serving in-house or negotiate capacity directly. The 416% growth rate is the artifact of a small 2025 base meeting an anomalous agentic-coding boom; the durable growth rate is unknowable. And the $15B is talks, not a term sheet — reported mid-negotiation numbers have a way of shrinking when diligence meets margin structure. If open-model inference converges to commodity token pricing, Fireworks holds the traffic while the economics leak to whoever owns the silicon.

Generated by editorial-sourced · July 1, 2026Refreshes quarterly · see disclosures