Key Points
- Fireworks AI raised a $1.505 billion Series D at a $17.5 billion valuation this week.
- Atreides Management, Index Ventures, and TCV led; Nvidia and Lightspeed also joined.
- The company passed a $1 billion annualized revenue run rate, up roughly fivefold in a year.
- Fireworks serves 40-plus trillion tokens a day, 95% from customer-specialized models.
Fireworks AI has closed a $1.505 billion Series D at a $17.5 billion valuation, the AI infrastructure company announced this week. Atreides Management, Index Ventures, and TCV led the round, with participation from Nvidia, Lightspeed Venture Partners, Evantic Capital, Bessemer Venture Partners, Menlo Ventures, and 20VC. It is one of the largest single financings of the year for a company that sits between the frontier labs and the enterprises trying to build on top of them.
The number is a marker of how fast this layer of the market is repricing. Two months ago, Fireworks was reported to be in talks to raise at a $15 billion valuation; the round it actually closed lands $2.5 billion higher. The company has now raised at least $1.8 billion in total.
Fireworks builds the infrastructure enterprises use to train and serve specialized models on open-source foundations and their own proprietary data, rather than calling a general-purpose model through an API. That focus shows up in the usage: daily token volume climbed from 15 trillion to more than 40 trillion year over year, and the company says more than 95% of those tokens now come from models tuned to a specific customer and job. Its annualized revenue run rate crossed $1 billion, a figure that has grown roughly fivefold in a year. Customers include Uber, Shopify, Doximity, Elastic, GitLab, and MongoDB, along with the legal-AI company Harvey and the coding tool Cursor.
The read: this is a wager on a specific shape the AI market is taking. Fireworks frames its pitch as "specialized intelligence" — the argument that companies will stop renting one general model and instead own tuned versions of open-source models trained on their own data. If that thesis holds, the value accrues not only to the labs that build frontier models but to the plumbing that lets everyone else run their own. That is the same logic behind treating AI infrastructure, not any single model, as the durable position — the case Uristocrat laid out in how the AI infrastructure boom actually gets owned. Nvidia writing a check into a company that moves 40 trillion tokens a day on its chips is the tell: the buildout keeps funding the tools that consume the buildout. The AI funding run continued as Meshy raised nearly $400 million at a $1.5 billion valuation.
Source: Fireworks AI, TechStartups.
Comments