Volantis Raises $88M Series A for Photonic Chip-to-Memory Inference
San Francisco photonics startup Volantis closed an $88 million Series A co-led by Lachy Groom and Abstract Ventures to build optical chip-to-memory interconnect for AI inference. First integrated engines are claimed for 2027 customer delivery; A-1 performance targets remain company claims without independent benchmarks.
Another sized bet that the inference bottleneck is memory reach, not just FLOPs — paid in venture dollars, delivered (claimed) in 2027.
Volantis announced an $88 million Series A co-led by Lachy Groom and Abstract Ventures, with John Doerr, VXI Capital, Triatomic, Susa Ventures, and angels Dwarkesh Patel, Naveen Rao, and Sholto Douglas. The company release (PR Newswire, October 1, 09:00 ET) is corroborated on financing and thesis by a Thomson Reuters report (Stephen Nellis) carried via syndication. CEO/co-founder Tapa Ghosh; team background cited as NVIDIA / AMD / Broadcom / Ayar Labs alumni in company materials.
Architecture pitch — company targets
Thesis: photonic interconnect (custom micro-VCSELs, no external lasers) linking compute to large off-chip memory pools so capacity and bandwidth rise together versus SRAM/HBM shorelines. First system A-1 is company-claimed to target models >20 trillion parameters at up to 10,000 tokens/sec/user and lower $/token; Reuters recounts a pitch of packing on the order of 220 memory chips around a GPU via optical reach. Volantis says first integrated inference engines go to customers in 2027. No independent benchmarks verified for this draft.
Proceeds: develop and commercialize A-1 and the photonic memory architecture, expand engineering, move toward deployments — per the release.
Limits
- Round size and investor list: Volantis PR Newswire, corroborated by Reuters (syndication full text; reuters.com canonical returned 401 this run).
- A-1 parameter, tok/s, tok/$, memory-chip-count, and 2027 delivery figures are company targets / interview claims, not measured shipping hardware.
- Traders Union secondary in the digest window is thinner and unused here.