Volantis Inc. has secured $88 million in a Series A funding round. The investment was led by Lachy Groom and Abstract Ventures, with participation from several other investors, including John Doerr of Kleiner Perkins and former Intel executive Naveen Rao.
The company is developing a chip architecture for AI inference that utilizes a custom optical memory interconnect. This design transmits data via light generated by VCSELs, which are made from gallium arsenide. According to Volantis, this technology offers over 30 times the memory bandwidth of existing accelerators.
The architecture extends interconnect wire range to over 200 millimeters, allowing for the integration of more than 220 memory chiplets. This capability addresses physical constraints that currently limit the number of memory modules on standard GPU designs.
Volantis plans to release the A-1, a data center inference appliance. The system occupies one-third of a standard server rack and provides 10 terabytes of memory with 250 terabits per second of bandwidth. It is designed to process up to 10,000 tokens per second for models with 20 trillion parameters.
