The inference startup founded by former Meta PyTorch lead Lin Qiao has crossed $1 billion in annualized revenue and now processes more than 40 trillion tokens a day, a sign that enterprise AI is moving toward cheaper specialized models instead of paying frontier API prices for every task.
Nine months is a long time in AI infrastructure. In October 2025, Fireworks AI closed a $250 million Series C that valued the company at $4 billion. On July 15, the company announced a $1.505 billion Series D at a $17.5 billion valuation, led by Atreides Management, Index Ventures, and TCV. Nvidia, Lightspeed Venture Partners, Evantic Capital, Bessemer Venture Partners, Menlo Ventures, 20VC and others also participated. That's a more than fourfold valuation jump in under a year, but this isn't just a price-tag story. Fireworks says it has crossed $1 billion in annualized revenue run rate, up from more than $280 million at the Series C. The numbers are real.
Lin Qiao built Fireworks on a straightforward premise: general-purpose frontier models are expensive to run at scale, and most enterprise use cases don't need the full capability of OpenAI or Anthropic's biggest systems. You don't need a flagship model for everything. For a lot of work, you need a fine-tuned, cheaper, lower-latency version of a capable open model that knows your own data. Fireworks handles that infrastructure. Notion gives you the practical version. In a Fireworks case study published last year, Notion said fine-tuning with Fireworks cut latency from about two seconds to 350 milliseconds. You feel that on every keystroke.
Who's actually using it
The customer list tells you where the demand is coming from. Fireworks has named customers including Cursor, Notion, Uber, DoorDash, Shopify, Upwork and Harvey, and Sacra's company profile says the base had grown to more than 10,000 companies by October 2025. These aren't companies running AI demos for a board deck. They're companies with high-volume, latency-sensitive applications where cost per token actually shows up in the margin structure. At more than 40 trillion tokens processed daily, even fractional price efficiency compounds into something real.
The broader dynamic is one the industry has been circling for a while. Frontier model providers, especially OpenAI and Anthropic, have built a business renting access to the most capable models at prices that reflect the cost of training them. That's fine for low-volume tasks: a legal team drafting an occasional memo, or a startup prototyping a feature. It doesn't work when you're routing millions of requests through a production product every day. The maths gets uncomfortable fast. Fireworks is where companies go when they've done it.
What the valuation actually says
What's notable about the $17.5 billion valuation isn't only the size of it. It's the revenue multiple underneath. Sacra's company data says Fireworks had reached more than $1 billion in annualized revenue by July 2026, which puts the Series D at roughly 17 times forward revenue. That's rich, but it isn't floating in the way some AI valuations are. Investors are paying for current volume and a growth rate that has already shown up in revenue - not only for a future market on a slide.
Qiao's background matters here. Fireworks' own team page describes her as previously head of PyTorch at Meta, and The Org lists her Meta role as senior director of engineering from 2015 to 2022, leading AI frameworks and platforms including Caffe2 and PyTorch. Fireworks isn't a thin reseller of someone else's model endpoint. It has built its own inference stack, and its current pitch is that companies can specialize models on proprietary data, then serve them with lower latency and better cost control than a generic hosted model call.
Frankly, the raise is as much a data point on where AI infrastructure economics are heading as it is on Fireworks itself. The story of the past two years in enterprise AI has been companies discovering that buying frontier model access is the beginning of the infrastructure journey, not the end. Running models efficiently in production, at the latency and cost your application actually requires, is a separate problem. Frontier labs aren't built to solve every version of that problem for you.
There's a lot of market left to fight over. CB Insights lists Fireworks' total funding at about $1.812 billion after the Series D, and CNBC's interview with Qiao, summarized by The Next Web and Gate, put current headcount near 200 with a year-end target of 600. Hiring that fast brings its own pressure. Fireworks' next question isn't whether demand exists. It's whether a $17.5 billion independent inference business can stay ahead as frontier labs and rival startups - and the hyperscalers - all push harder into the same enterprise budget.
Also read: China signs 29 nations into a rival AI bloc while open-source models do the real work • BloombergNEF just revised its US data center power forecast 83% higher in seven months • Claude Fable 5 helped crack the Jacobian Conjecture after 87 years of failure
