Volantis, a San Francisco semiconductor company building an AI inference system around photonic memory links, has raised an $88m Series A. Lachy Groom and Abstract Ventures co-led the round, which the company announced on 1 October 2026.
John Doerr, VXI Capital, Triatomic and Susa Ventures also invested, along with angel investors Dwarkesh Patel, Naveen Rao and Sholto Douglas. Unite.AI reported that the round brings Volantis's total funding to $97m, with earlier backing from Sam Altman, Jeff Dean and Dylan Patel among others. No valuation was disclosed.
“Today's hardware forces a tradeoff between running the largest, most sophisticated models and running them fast.”Tapa Ghosh, CEO and Co-Founder, Volantis
What Volantis is building
Volantis's first system, A-1, targets the memory bottleneck in AI inference: the largest models need more memory capacity and bandwidth than current accelerators can provide while still running fast. The company says A-1 is being designed to run models with more than 20 trillion parameters at up to 10,000 tokens per second per user, and to raise memory capacity and bandwidth together by nearly two orders of magnitude. The Next Web pointed out that these are design goals, not measurements from a working machine.
The system uses an optical fabric to connect many memory chips into a single pool, so bandwidth grows as memory is added, and it allows cheaper off-chip memory to be used. Instead of external lasers it uses custom micro-VCSELs, small lasers built on the gallium arsenide supply chain, which the company says avoids constraints in the supply of indium phosphide. Volantis claims its end-to-end links use less than one picojoule per bit.
SiliconANGLE reported that the A-1 appliance is about a third of the size of a standard server rack, with 10 terabytes of memory and 250 terabits per second of memory bandwidth. Electrical interconnects today reach about five millimetres, which limits how many memory modules can sit around a chip. Volantis says its optical links reach more than 200 millimetres, allowing more than 220 memory chiplets per accelerator.
The team
Co-founder and chief executive Tapa Ghosh is a Thiel fellow and former Y Combinator founder, according to The Next Web. Co-founder and chief technology officer Roy Meade previously ran Micron's high-bandwidth memory programme and was a vice president at Ayar Labs. The wider founding team includes semiconductor and photonics engineers from Nvidia, AMD, Broadcom and Ayar Labs, whose earlier work includes the first CoWoS product, the first high-volume tunable VCSELs and early silicon photonics co-packaged optics systems.
In a blog post cited by SiliconANGLE, Ghosh said faster inference could let a coding agent finish a task in 30 seconds rather than 30 minutes.
What the money is for
Volantis said the Series A will pay for the development and commercialisation of A-1 and its photonic memory architecture, a larger engineering team and the work needed to move the system towards customer deployments.
The company plans to deliver its first integrated inference engines to customers in 2027 and said it will reveal further parts of the architecture as A-1 approaches commercial release. It has not disclosed its headcount or the year it was founded.
Company website: volantissemi.ai