Fireworks, the San Mateo-based platform for building and serving customized AI models, has raised a $1.505 billion Series D at a $17.5 billion valuation, securing one of the largest private financing rounds in the AI infrastructure sector this year. The round was led by Atreides Management, Index Ventures, and TCV, with participation from existing investors NVIDIA, Lightspeed Venture Partners, and Evantic, according to details reported by Pulse 2.0 and corroborated by coverage in Quartz, Yahoo Finance, and HPCwire.
The financing lands as investors pour record sums into the picks-and-shovels layer of the generative AI boom, a wave this publication's breaking AI news desk has tracked across a string of oversized infrastructure deals. What sets Fireworks apart, the company argues, is not raw compute capacity but a bet that the future of enterprise AI belongs to specialized models tuned on each company's own data rather than rented from a handful of frontier labs.
From $4 Billion to $17.5 Billion in a Single Round
The valuation marks a steep step up. Fireworks previously raised a $250 million Series C at a $4 billion valuation, meaning its worth has roughly quadrupled over the course of one funding cycle. The jump is underpinned by explosive commercial momentum: the company said it has now surpassed $1 billion in annualized revenue run rate, a fivefold increase year-over-year.
Usage on the platform has scaled in tandem. Daily token volume processed through Fireworks has grown from roughly 15 trillion to more than 40 trillion tokens over the same period, according to the company's announcement. That kind of throughput reflects deep production deployment rather than experimentation, a distinction investors have increasingly demanded as they sort sustainable AI businesses from speculative ones.
A Bet on 'Specialized Intelligence'
Fireworks was founded by the team behind PyTorch, the open-source deep learning framework that underpins much of modern machine learning. Its platform lets enterprises take open-source AI models and customize them on proprietary data, workflows, and use cases, then serve those specialized models in production at scale.
The core thesis is a direct challenge to the prevailing assumption that the most capable general-purpose models from frontier labs will dominate. Fireworks argues that closed models, however powerful, cannot fully reflect the domain knowledge, customer context, and operational workflows unique to any given company and that customized open models can match or outperform them on the specific tasks a business cares about, while running faster and at lower cost.
The company points to adoption as proof: 95% of the tokens currently served through Fireworks come from models that have been customized rather than used off the shelf. Named customers include Uber, Shopify, Doximity, and Revolut. The platform offers more than 200 models spanning text, image, and multimodal formats, with support for major new open-source releases typically available within hours of launch.
Who Backed the Round
The Series D assembled a broad syndicate of growth and crossover investors. The round was co-led by Atreides Management, Index Ventures, and TCV. Returning backers NVIDIA, Lightspeed Venture Partners, and Evantic also participated, alongside new and additional checks from 20VC, Bessemer Venture Partners, Insight Partners, Lone Pine Capital, Menlo Ventures, and the Ontario Teachers' Pension Plan. Moomoo and other outlets highlighted NVIDIA's continued backing, a signal of alignment between the chipmaker and one of the fastest-growing consumers of inference compute.
Gavin Baker, Chief Investment Officer and Managing Partner at Atreides Management, framed the bet as a hedge on a multi-model future. "We believe both frontier and open models will increasingly be used together," Baker said in a statement cited by Pulse 2.0. "With a differentiated platform used and trusted by enterprises to train, customize, and serve open models in production, Fireworks is exceptionally well-positioned to unlock the value of specialized intelligence."
Fireworks itself framed the choice in starker terms, describing two diverging paths for the industry. In one, intelligence belongs to a few large labs and everyone else simply rents it; in the other, every company builds specialized intelligence of its own, shaped by the domain only it understands. "We are building towards the second," the company said.
A Crowded, Well-Funded Field
The round intensifies an already heated race among independent AI infrastructure providers. Rivals including Together AI, which recently raised $800 million at an $8.3 billion valuation, are chasing the same enterprise demand for faster, cheaper, and more controllable model serving. What is unusual about Fireworks' position is the combination of a nine-figure revenue run rate, NVIDIA's strategic backing, and a product wedge specifically open-model customization that sits adjacent to, rather than directly under, the frontier labs.
Proceeds from the Series D will fund expansion of engineering headcount, additional global compute capacity, and deeper partnerships with cloud providers including Microsoft and NVIDIA. The company plans to keep broadening its model catalog and tightening the gap between when a new open-source model is released and when enterprises can run a customized version of it in production.
For now, the financing reinforces a broader pattern shaping 2026's AI market: the biggest capital is flowing not to the companies that build the smartest single model, but to the infrastructure that lets every other company put those models to work on its own terms. Whether specialized intelligence proves durable as a category or compresses into features offered by the cloud giants will be one of the defining questions of the next funding cycle.
Stay Ahead of AI
Fireworks' mega-round is the latest signal that enterprise AI infrastructure remains the richest seam in the industry. For more on the deals, models, and policy moves reshaping artificial intelligence, follow our latest AI developments as they break.
Read more AI news ->
