Fireworks Hits $17.5B Valuation as AI Firms Cut Costs

A Fast-Moving Repricing of AI Infrastructure

Fireworks, the San Mateo-based AI infrastructure company founded by former Meta engineers who helped build PyTorch, has closed a $1.505 billion Series D round that values the company at $17.5 billion 1011. The round was led by Atreides Management, Index Ventures and TCV, with participation from Nvidia, Lightspeed Venture Partners, Evantic Capital, Bessemer Venture Partners, Menlo Ventures, Insight Partners, Lone Pine Capital, Ontario Teachers' Pension Plan, 20VC, Operator Collective, Original Capital, Prysm Capital, Quantum Capital and TIME Ventures 111619. The size of the check and the speed of the valuation jump mark one of the largest recent bets on the infrastructure layer that sits between AI model developers and the enterprises that actually run those models in production 114.

The jump is striking mainly because of how recently Fireworks was valued far lower. The company raised a $250 million Series C in October 2025 at just a $4 billion valuation, led by Lightspeed Venture Partners, Index Ventures and Evantic, with Sequoia Capital also participating 1213. Reports in May 2026 indicated Fireworks was in talks to raise at around $15 billion, nearly four times its Series C price, before the round ultimately closed even higher, at $17.5 billion 2019. Some tracking sites list the post-money figure slightly differently, at roughly $15.5 billion, reflecting the kind of discrepancies that often appear across financing databases when secondary transactions and rounding are involved 17.

What Fireworks Actually Sells

Fireworks operates a cloud platform that lets developers fine-tune, customize and deploy AI models, with a particular focus on open-weight systems rather than proprietary frontier models 1411. Its offering includes managed GPU clusters priced on usage, an automated agent that handles the technical work of fine-tuning a model once a customer describes a task and uploads training data, and multiple parallelization techniques tuned to different model architectures 14. Once a model is customized, customers can run it through either serverless hosting or dedicated GPU deployments that Fireworks says deliver stronger performance 14.

The pitch, as described by CEO and co-founder Lin Qiao, is that companies do not need to rent access to a handful of expensive general-purpose models for every task. Instead, they can adapt open models to their own data, workflows and definitions of quality, producing systems that match or beat closed-model performance while running faster and far more cheaply 1114. Qiao told CNBC that Fireworks' cost advantage over an equivalent-quality closed model can run five to 10 times cheaper, a claim central to the company's growth story though not independently verified 15.

The Numbers Behind the Valuation

Fireworks says it has crossed $1 billion in annualized revenue run rate, a fivefold increase year-over-year, while daily token volume — the basic units of text an AI model processes — nearly tripled from 15 trillion to more than 40 trillion 1116. Third-party estimates suggest the ramp was even steeper in recent months, with annualized revenue climbing from roughly $305 million at the end of 2025 to about $800 million by May 2026 before surpassing $1 billion at the time of the Series D 1819. Fireworks also says more than 95% of tokens served on its platform now come from models that have been specialized on customer data, a figure the company points to as evidence that enterprise AI usage is shifting away from generic frontier access 1011.

Disclosed or reported customers include Uber, Shopify, Doximity, Samsung, GitLab and Revolut 111416. CNBC has noted that coding startup Cursor once accounted for more than half of Fireworks' revenue, a concentration the company has since diversified away from as more businesses adopt lower-cost AI options 1. At more than $1 billion in run-rate revenue, the new valuation implies roughly a 17.5-times revenue multiple, which analysts note is aggressive relative to rival Together AI's roughly 7-times multiple on its $8.3 billion valuation, though still below the multiples some frontier model labs command 1819.

Where the Money Goes

Fireworks says it will use the new capital to expand global compute capacity, hire additional engineers, build out its sales organization and deepen partnerships with Microsoft and Nvidia 1116. The company draws on computing capacity from more than 20 suppliers and struck a partnership with Microsoft in March that lets Microsoft customers access models through the Fireworks platform 15. Nvidia's role as both investor and chip supplier underscores how the GPU maker is positioning itself across the AI value chain — not only selling the hardware that trains and runs these models, but also backing the companies building businesses on top of it 101516.

Competing for the Inference Layer

Fireworks competes most directly with Together AI, valued at roughly $8.3 billion, and Baseten, valued around $1.3 billion, both of which occupy similar ground in helping enterprises run and customize AI models outside of the major cloud providers 19. It also competes indirectly with Amazon, Google and Microsoft, which can bundle model hosting into their existing cloud relationships 1619. Gavin Baker of Atreides Management framed the investment thesis as model-agnostic, saying frontier and open models will increasingly be used together rather than one replacing the other 1116.

That framing places Fireworks' round within a broader pattern of 2026 AI funding, where investors have shown willingness to write very large checks into infrastructure and inference companies even as overall deal counts in adjacent sectors like fintech have fallen 7. Crunchbase's weekly tracking of megadeals has repeatedly highlighted AI and energy as the dominant categories for nine-figure-plus rounds this year 9, while broader coverage of the AI funding landscape shows capital concentrating in fewer, larger transactions across frontier labs, infrastructure providers and specialized vertical applications alike 68.

A Broader Venture Landscape

Fireworks' raise arrives amid continued volatility in how venture investors are pricing AI bets more broadly. Anthropic's transformation of Menlo Ventures' roughly $1 billion investment into an estimated $14 billion in value has been cited as evidence that outsized AI returns remain possible, even as analysts caution that identifying the next breakout company is only getting harder 4. Elsewhere in AI-adjacent funding, identity security startup Oak emerged from stealth with $60 million 2, liquid-data startup Apoha raised $36 million to target pharmaceutical and food applications 5, compliance-communications platform LeapXpert secured $180 million led by Goldman Sachs Growth Equity 6, and Ant Group backed a $73.6 million round for humanoid robotics startup Zeroth Robotics as part of China's growing robotics push 3. These deals illustrate how capital is spreading across very different corners of the AI ecosystem even as headline-grabbing infrastructure rounds like Fireworks' dominate attention.

The Risks Ahead

Despite the enthusiasm, Fireworks faces meaningful structural risks. Open-source serving frameworks such as vLLM and SGLang continue to improve, which could erode the proprietary performance edge Fireworks relies on for pricing power 18. Annualized revenue run rate is not the same as profit, and GPU costs remain a persistent pressure on margins, particularly as the company competes partly on being cheaper than closed alternatives 18. Hyperscalers with their own model-hosting products and existing enterprise relationships pose a long-term competitive threat, even as Fireworks partners with Microsoft in the near term 16. And while the company says it has diversified beyond its early reliance on Cursor, it has not published a detailed customer-revenue breakdown, leaving open the question of how concentrated its growth remains among a handful of large accounts 1.

Still, the scale of investor commitment — from Nvidia's strategic backing to the participation of institutions like Ontario Teachers' Pension Plan and Insight Partners — signals confidence that the shift toward specialized, cost-efficient AI deployment is more than a passing trend. Whether Fireworks can convert its rapid revenue growth into durable margins, in a market where inference technology is advancing quickly on multiple fronts, will determine whether its $17.5 billion price tag proves prescient or premature.