Volantis raises an $88m Series A to pool AI memory chips with light, and says its first system ships in 2027

Volantis, a San Francisco chip startup, announced an $88m Series A on 1 October. Lachy Groom and Abstract Ventures co-led the round, with John Doerr, VXI Capital, Triatomic and Susa Ventures taking part. Dwarkesh Patel, Naveen Rao and Sholto Douglas invested as angels.

The company is building an optical fabric that “connects large numbers of memory chips into a unified pool”, adding their bandwidth together as memory is added. The target is the trade-off in AI inference between how much memory a system can hold, which limits model size, and how fast it can read that memory, which limits response speed.

Volantis claims its first system, A-1, will run models of more than 20 trillion parameters at up to 10,000 tokens per second per user, and reduce inference cost per token. It plans to deliver its first systems to customers in 2027. None of these performance figures has been shown publicly; they are the company's targets.

Tapa Ghosh, chief executive and co-founder, said: “Today's hardware forces a tradeoff between running the largest, most sophisticated models and running them fast. We started Volantis to eliminate that tradeoff.”

Reuters reported further detail on how many memory chips the design places around a GPU; the release does not give that number, so it is not repeated. The round follows CScale's $145m for optical links in AI racks, which ran on 1 October.

• Confirmed $88m Series A co-led by Lachy Groom and Abstract Ventures, with John Doerr, VXI Capital, Triatomic and Susa Ventures. Volantis (PR Newswire)
• Claimed A-1 will run models of more than 20 trillion parameters at up to 10,000 tokens per second per user, with first customer deliveries in 2027. Volantis (PR Newswire)