Fireworks Raises $1.505B Series D at a $17.5B Valuation

Fireworks Raises $1.505B Series D at a $17.5B Valuation

In This Article

  1. The round, in one paragraph
  2. The numbers the company disclosed
  3. What Fireworks actually sells
  4. From $552M to $17.5B in two years
  5. Why it matters
  6. Common questions

Key Takeaways

On July 16, 2026, Fireworks announced a $1.505 billion Series D at a $17.5 billion post-money valuation, co-led by Atreides Management, Index Ventures and TCV, according to the company's own announcement and the accompanying BusinessWire release. This is our read on what the round signals for teams deciding whether to rent intelligence from a frontier API or serve their own.

The round, in one paragraph

The headline figure is usually rounded to $1.5 billion; the precise number in the company's materials is $1.505 billion, which deal counsel Orrick called "over $1.5 billion." Beyond the co-leads, the release names 20VC, Bessemer Venture Partners, Evantic Capital, Insight Partners, Lone Pine Capital, Lightspeed Venture Partners, Menlo Ventures, NVIDIA, Operator Collective, Ontario Teachers' Pension Plan, Original Capital, Prysm Capital, Quantum Capital and TIME Ventures — with Evantic, Lightspeed and NVIDIA returning from earlier rounds. Ontario Teachers' confirmed its participation on its own site. Per SiliconANGLE, the proceeds go to expanding compute infrastructure and hiring engineers.

A sourcing note: the release carries a San Mateo, California dateline, while company databases and earlier coverage list Redwood City. We could not resolve which is current, so we note both.

The numbers the company disclosed

Fireworks paired the raise with operating metrics. All come from the company and are unaudited, so read them as disclosures rather than established facts: it says it has surpassed $1 billion in annualized revenue run rate, a 5x increase year over year; that daily token volume has gone from 15 trillion to more than 40 trillion tokens per day; that it offers 200-plus models across text, image and multimodal; and that more than 95% of the tokens it serves come from models specialized on customers' proprietary data. The release names Uber, Shopify, Doximity and Revolut as customers; SiliconANGLE separately names Samsung Electronics and GitLab, and the company's blog mentions Cursor and Harvey.

95%+
Share of tokens Fireworks says it serves from models specialized on a customer's own data, rather than general models used as-is.
Company-stated and unaudited. Alongside a claimed $1B annualized run rate and 40T+ tokens served per day.

Chief executive and co-founder Lin Qiao framed the thesis directly in the release: "There are two paths forward for AI. In one, intelligence belongs to a few big labs, and everyone else rents it. In the other, every company in the world builds specialized intelligence of its own, shaped by the domain only it understands." Gavin Baker, CIO and managing partner at lead investor Atreides Management, said in the same release that "Fireworks has assembled one of the most elite and technical teams in AI, paired with technology that consistently sets the pace for the industry and commercial momentum that very few companies have ever achieved at this scale."

What Fireworks actually sells

Stripped of the positioning, the product is inference and fine-tuning as a managed service for open-weight models. SiliconANGLE describes two serving modes — a serverless option needing minimal configuration, and dedicated GPU deployments for more performance and control — plus a tuning agent that automates training workflows and hyperparameter search, four parallelization techniques matched to different model types, quantization to cut the hardware bill, and demand-tracking autoscaling.

The implied workflow is concrete: take a capable open-weight model, tune it on your own data, serve it behind an endpoint, and keep the resulting weights. Whether that beats a closed frontier model on cost depends on your volume and quality bar — we verified no price comparison and will not supply one. Our guide to open-weight models versus frontier APIs covers how to run that comparison, and the LLM price calculator handles the arithmetic.

From $552M to $17.5B in two years

The step-ups are steep enough to be worth laying out plainly.

Fireworks funding history, as reported

RoundAmount / valuationDate & leads
Series B$52M / $552MJuly 2024, led by Sequoia, per Bloomberg
Series C$250M / $4BOct. 28, 2025, co-led by Lightspeed, Index, Evantic
Series D$1.505B / $17.5BJuly 16, 2026, co-led by Atreides, Index, TCV

On the Series B valuation, sources conflict: Bloomberg reported $552 million in July 2024, while several funding trackers list $522 million. We give both. The Series C is firmer: $250 million at $4 billion, announced October 28, 2025 and co-led by Lightspeed, Index and Evantic with Sequoia participating, per the company's release and Orrick. We omit a lifetime-total-raised figure: the sources we checked disagree, and a number we cannot reconcile is not worth printing.

One more piece of context. On May 27, 2026, Bloomberg reported that Fireworks was in talks at a $15 billion valuation. Seven weeks later the round closed at $17.5 billion — above the figure under discussion, and roughly 4.4x the Series C mark set nine months earlier. The company was founded in 2022 by seven former Meta engineers, five of them core PyTorch contributors; Qiao previously led the PyTorch team there.

Why it matters

The section above is reported. What follows is our analysis.

The signal here is not the dollar amount. It is the 95% figure. A serving platform reporting that nearly all its traffic runs through customer-specialized models is describing a market where the differentiator has moved from the model to the data and workflow wrapped around it — a familiar pattern as infrastructure categories mature.

A big round is not a verdict on the technology

Private valuations are a claim about the future priced by a handful of buyers, not a measurement of product quality. A $17.5 billion mark tells you what investors believe about specialized inference; it tells you nothing about whether fine-tuning an open-weight model beats a frontier API on your workload. Only your own evaluation answers that.

The practitioner takeaway is narrower than the headline. Capital at this scale usually buys capacity and price competition, which favors buyers near term. It also raises the cost of an architectural mistake: committing a production system to a serving platform is a real dependency, and what protects you is portability of the artifacts you own — your data, your evaluation suite, your tuned weights. Before moving a workload, decide what you would need to take with you. Our framework for build versus buy and our guide to how teams actually test are the homework worth doing first; why AI pilots never reach production covers what goes wrong afterward.

Run the comparison on your own numbers

Open-weight and fine-tuned, or a frontier API? The answer depends on your token volume and quality bar — not on anyone's valuation. Our free calculator lays out the real cost side by side.

Open the price calculator

Sources: Fireworks — Series D announcement; BusinessWire (July 16, 2026); Ontario Teachers' Pension Plan; SiliconANGLE; AIwire; Orrick (Series D); BusinessWire (Series C, Oct. 2025); Bloomberg (May 2026); Bloomberg (July 2024). Analysis and framing by Precision AI Academy.

Common questions

Is the $1 billion revenue figure audited? No. It is an annualized run rate the company disclosed in its own announcement. Treat it, and the token-volume figures, as company statements rather than verified results.

Did Sequoia participate in the Series D? Sequoia backed earlier rounds, including the Series B it led and the Series C, but it is not on the Series D investor list in the release. We found no confirmation either way, and will not imply one.

Why do some outlets say $1.5 billion and others $1.505 billion? They are the same round. $1.505 billion is the precise figure; $1.5 billion is the rounding used in headlines, and Orrick's phrasing is "over $1.5 billion."

Does this change anything for my stack today? No. A funding round does not change an API, a price sheet, or a model's quality. It signals where investors think serving demand is heading — worth knowing, not worth acting on by itself.

About Precision AI Academy

Precision AI Academy publishes practical AI news, plain-language analysis, and 137 free courses for builders and working professionals. It is a sister site of Precision Federal, a federal software and AI firm. We verify the numbers, cite the primary sources, and skip the hype.