Positron Raises $875 Million for Inference Chips; Founders Unable to Infer Where Most of It Will Go by Q3
The Series C values the company at $5 billion, and its burn rate at "a number we're modeling."
RENO, Nev. — Positron, a startup building chips optimized for AI inference, closed an $875 million Series C this week at a $5 billion valuation, and its founders said they are "extremely confident" in their technology's ability to infer things and "working on" its ability to infer where the money will be in nine months.
"Inference is the future," said the company's CEO. "Training is a one-time cost. Inference is forever. Every query, every token, every time someone asks a chatbot to make a poem shorter. We are building the silicon for that, and we now have the capital to do it, for a while."
The funding will support a tape-out of the company's second-generation chip, a process that costs tens of millions of dollars per attempt and that the CEO described as "like a rocket launch, except if it fails you don't get a cool video."
Investors said the round reflects strong demand for alternatives to Nvidia, whose dominance of the accelerator market has made every chip startup's pitch deck begin with the same slide. "Slide one is always the Nvidia margin," said one investor. "Slide two is always 'we can do it for less.' Slide three is where it gets interesting, and Positron's slide three was very good."
The company's projected timeline calls for production silicon within 18 months, customer deployments within 24, and profitability at a date the company's financial model lists as "TBD (inference)."
Positron said it would use the round to grow its team from 200 to 500, primarily engineers, and to purchase Nvidia GPUs to test its chips against, which the CEO acknowledged was "a little ironic" but "unavoidable, and also they're very good."