Volantis raises $88 million for technology linking AI chips to memory through laser beams

San Francisco semiconductor startup Volantis has raised $88 million in venture capital, as Reuters reports, to develop a technology that links artificial intelligence compute chips to memory chips through laser beams of light.

According to Reuters, the main limit of Nvidia’s and AMD’s AI chips is how well the compute parts communicate with the memory chips where the AI model lives. Both companies use expensive HBM (high-bandwidth memory) chips, but even Nvidia’s best current offering holds only 8 memory chips per GPU because of the limited range of electrical wires.

Volantis is developing the transfer of data between compute chips and memory chips through laser beams — this removes the range limit and would allow around 220 memory chips around a single GPU. The company uses VCSEL lasers (vertical-cavity surface-emitting lasers), the same technology that powers facial recognition in hundreds of millions of iPhones; Apple has invested heavily in its supply chain over the last decade.

Chief executive and co-founder Tapa Ghosh said in an interview: “Advanced packaging must always be respected — it is never a trivial thing — but it is not necessarily something new to do. Nobody will win the Nobel Prize if our project works, but the good news is that they won’t have to.” The company aims to avoid supply-chain bottlenecks and to deliver next year a chip that speeds up tasks such as AI programming.

According to Reuters, the round was led by Stripe veteran Lachy Groom and Abstract Ventures. It was also joined by John Doerr, the investor who backed Google and Amazon in their early days, together with VXI Capital, Triatomic and Susa Ventures. Angel investors include AI podcaster Dwarkesh Patel, AI chip veteran Naveen Rao and Anthropic researcher Sholto Douglas.