We’ve raised a Series A, led by Lachy Groom and Abstract, totalling $97M in funds raised to date, including from Sam Altman, Jeff Dean, Dylan Patel, John Doerr & others. We’re building a system for AI inference that uses photonics to break the memory wall.

We’ll enable giant models (over 10T parameters) at ultra-high interactivity speeds (up to 10,000 tok/s/user).

This will enable real-time frontier inference, restart scaling laws & enable entire code bases in context windows. As a starting point, imagine a coding agent that completes a task in 30 seconds rather than 30 minutes.

Doing this (frontier inference at high token speeds) requires both increased memory size and memory bandwidth.

When it comes to memory size and bandwidth, you can’t “have your cake and eat it too”. Chips using on-chip SRAM deliver high memory bandwidth in exchange for reduced memory size. And HBM-based GPUs/TPUs deliver high memory size in exchange for low memory bandwidth.

We deliver both. We have built a new category of optical interconnect that can boost both memory capacity and bandwidth by over an order of magnitude. To boot, optical memory pooling enables lower $/token due to using less expensive memory and having higher throughput.

Bringing optics to these extreme bandwidth levels required engineering a new optical technology stack that we call the optical fabric. For one, we eliminated external lasers, instead using custom, integrated micro-VCSELs. As another example, we eliminated traditional optical fiber to enable much higher bandwidth densities.

Prior to Volantis, our team built industry first products at places like NVIDIA, AMD, Broadcom and Ayar Labs — including the first CoWoS product, the first silicon photonics CPO systems, the first high-volume tunable VCSELs, and more. Many of us have worked on some of the most innovative chips of their time, but Volantis is by far our most ambitious project to date. If this excites you, come join us.