• 2 min read
Fireworks lands $1.5bn as custom AI demand surges
Fireworks raised a $1.5bn Series D at a $17.5bn valuation, betting companies want to build tailored AI instead of renting models.
Image: TNW
The AI market has largely been built on renting models from big labs. Fireworks just raised $1.5bn to push a different idea: companies will increasingly build their own intelligence instead.
The Series D values the San Mateo startup at $17.5bn, according to the company. The round was led by Atreides Management, Index Ventures, and TCV, with existing backer Nvidia also participating.
“Fireworks has assembled one of the most elite and technical teams in AI.”
Fireworks runs open models for other companies and helps them tune those systems on their own data. The company calls the result “specialized intelligence”: models shaped by the proprietary knowledge a single business holds.
That pitch appears to be gaining traction. Index Ventures said Fireworks' revenue has passed $1bn on an annualised basis, up fivefold in a year. The platform now serves more than 40 trillion tokens a day, and 95% of them run on specialized models.
Scale, cost, and competition
That token volume stands out. Based on CNBC’s numbers, Fireworks handles more developer requests each day than Google or OpenAI report serving, even though its revenue is only a fraction of theirs.
Recommended reading
Hassabis says STEM makes you 10x better at AI
The company is riding a broader shift in the market as open models narrow the gap with the best closed systems. That has pushed competition away from simply having the biggest model and toward offering the cheapest capable one. Fireworks says its systems run at a fifth to a tenth of the cost.
Chief executive Lin Qiao described the moment as a fork in the road.
“In one [path], intelligence belongs to a few big labs, and everyone else rents it. We are building towards the second.”
The view is shared by other tech leaders cited by the source, including Microsoft CEO Satya Nadella, who argues companies should use models without giving up the knowledge that makes them distinct, and Palantir CEO Alex Karp, who says customers want to “own the means of production.”
Founded in 2022 by former Meta engineers behind PyTorch, Fireworks competes with Together AI and Baseten in inference, and increasingly with neocloud rivals in training. One risk is customer concentration: Cursor once accounted for about half its revenue, though Qiao said the customer base is now broader. Fireworks plans to grow headcount from 200 to 600 by year-end.
“This is the year when we’ll really hit the gas.”
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via TNW
