Exclusive | Positron Valued at $5 Billion in New Funding as CPU Demand Surges

Positron, an artificial-intelligence chip startup focused on designing hardware that helps AI models run more quickly, said it has raised $875 million at a valuation of $5 billion, marking a significant increase in the company’s worth over the last six months.

The round was co-led by venture-capital firm NEA, Atreides Management, Valor Equity Partners, Andra Capital, analyst Dylan Patel’s SemiAnalysis Capital and Netscape co-founder Jim Clark. Positron raised $230 million in its most recent funding round, in February of this year, valuing the company at $1 billion.

Positron, based in Reno, Nev., is led by a team of executives that previously worked at cloud-computing firm Lambda and chip designer Groq. The startup is completing work on its flagship product, an AI processor called Asimov, which is integrated into a server called Titan. The chip is designed with a “memory-first architecture,” said Positron Chief Executive Mitesh Agrawal, a Stanford-educated former financial analyst.That means that each Asimov chip is loaded with up to 2.3 terabytes of memory, making it especially suited to fast processing of AI queries, known as inference. How fast an AI model can run typically depends largely on how much memory each processor has access to, whereas training a model relies more on computing muscle.

Positron has already shipped approximately 50 of its previous generation of server racks, called Atlas, to Oracle, and counts finance firm Jump Trading and AI firm Parasail as customers, the company said.

The rise of AI agents, or autonomous bots that perform tasks on behalf of humans or companies, has prompted an immense wave of demand for computing hardware, especially chips customized for inference.

Large chip companies including market leader Nvidia and its top competitors, such as Advanced Micro Devices and Google, have invested heavily in improving their inference capabilities in the past year. Meanwhile, a crop of startups and newer entrants to the market, including Cerebras and Etched, as well as the custom silicon arms of big AI labs, are all rolling out new chip designs meant to increase the speed and efficiency of inference processing.

Positron is seeking an edge in the crowded chips business by bringing its chips to market as quickly as possible.

“You have inference really, really ramping up right now, but I believe that demand is just going to keep on getting bigger,” Agrawal said. “Speed is key.”