Volantis raises $88m for photonic chip-to-memory architecture

Volantis has raised $88m in a Series A funding round to develop an AI inference system built around a photonic memory interconnect.

The round was co-led by Lachy Groom and Abstract Ventures, with investment from John Doerr, VXI Capital, Triatomic and Susa Ventures. The company’s founding team includes former employees of Nvidia, AMD, Broadcom and Ayar Labs.

Volantis is developing a system called A-1. Its proposed architecture uses an optical fabric to connect processing devices with multiple off-chip memory devices, aggregating memory capacity and bandwidth as more devices are added.

The company says the links use custom micro-vertical-cavity surface-emitting lasers, or micro-VCSELs, rather than external lasers. VCSELs can be fabricated using the established gallium arsenide supply chain, which Volantis says reduces its dependence on indium phosphide components.

Volantis is targeting link energy below one picojoule per bit. It also claims A-1 will eventually support models containing more than 20 trillion parameters at speeds of up to 10,000 tokens per second for each user.

Those figures remain development targets. Volantis has not published independent benchmark results, complete system specifications or customer evaluation data for A-1.

The distinction between chip-to-chip and chip-to-memory photonics is important. Memory traffic operates over relatively short distances but at very high aggregate data rates. Cost, laser efficiency, packaging tolerances, thermal behaviour and reliable optical coupling are likely to determine whether the architecture is commercially practical.

Volantis plans to use the funding to expand its engineering team and move A-1 towards deployment. The first integrated inference engines are scheduled for delivery to customers during 2027.