Volantis raises $88M to to build photonic memory layer for next-gen AI

Volantis has pulled in $88 million in Series A funding to push a photonic architecture for AI inference. The round was co-led by Lachy Groom and Abstract Ventures, with backing from John Doerr, VXI Capital, Triatomic, Susa Ventures, and angels including Dwarkesh Patel, Naveen Rao, and Sholto Douglas.

The San Francisco startup is building A-1, its inference system for large AI models. Volantis says the target is ambitious: systems that can handle models exceeding 20 trillion parameters and serve up to 10,000 tokens per second per user, while cutting inference costs. Those are product goals, not deployed performance numbers, and the first integrated inference engines are slated for customer delivery in 2027.

The pitch is really about memory. Big models need huge capacity for weights and context, plus enough bandwidth to keep feeding compute. Today, that usually means choosing between fast-but-small on-chip SRAM and larger memory setups that still run into bandwidth, cost, and energy limits. Volantis argues even newer approaches such as 3D DRAM still sit on the same tradeoff curve.

Its answer is a photonic interconnect that links compute chips to memory and forms a shared memory pool, with bandwidth rising as memory is added. The company says the architecture uses custom micro-VCSELs rather than external lasers, leaning on the gallium arsenide VCSEL supply chain and avoiding some indium phosphide constraints. It is also aiming for end-to-end optical links under one picojoule per bit.

Volantis was founded by Tapa Ghosh and Roy Meade. Ghosh is a Thiel Fellow and former Y Combinator founder with four patents, while Meade previously led Micron’s HBM programme and was a vice president at Ayar Labs. The new money is set to go into developing and commercialising A-1, expanding the engineering team, and moving toward customer deployments.