Volantis raises $88M to rewire AI accelerators with light

Today's AI accelerators are limited less by how fast they compute than by how fast they can move data to and from memory over copper wires. That bottleneck, the memory wall, is what Volantis, a startup backed by Sam Altman, Jeff Dean, Dylan Patel, and John Doerr, says it plans to break with light.

Volantis closed an $88 million Series A co-led by Lachy Groom and Abstract Ventures, bringing total funding to $97 million, according to a company press release. Its first system, A-1, is an inference accelerator that pools memory optically across chips rather than leaning on each accelerator's own HBM. Design targets include large model support — the company's press release says exceeding 20 trillion parameters while its blog says over 10 trillion, a discrepancy it has not resolved — at up to 10,000 tokens per second per user, with the company projecting its optical fabric could lift memory bandwidth and capacity by nearly two orders of magnitude.

Those are targets, not measured results. Volantis is pre-product: no tape-out date, no customer, no third-party benchmark. The team includes veterans of NVIDIA, AMD, Broadcom, and Ayar Labs, and the wider race includes Lightmatter and the silicon photonics work inside Intel and Nvidia.

The Register has more on the company's plans.