San Francisco semiconductor startup Volantis said on Thursday it has raised $88 million in venture capital to attack one of the hardest problems in AI hardware: how computing chips talk to the memory chips that store AI models. Its answer is to replace copper wires with beams of laser light.
The fundamental limit for AI chips from Nvidia and AMD is bandwidth between computing cores and memory. Current packaging encircles computing chips with expensive high-bandwidth memory chips, but the tiny electrical wires connecting them can only reach so far — even Nvidia's best offerings can hold only eight memory chips per graphics processing unit. Volantis says its laser-based interconnect would eliminate the reach problem, allowing 220 memory chips to be packed around a single GPU.
The trick is to use vertical-cavity surface-emitting lasers, or VCSELs — the same tiny lasers that power facial recognition in hundreds of millions of iPhones, with a supply chain that Apple spent a decade building up. By assembling proven components rather than exotic new materials, Volantis hopes to avoid the supply-chain bottlenecks that plague AI chip development.
"Advanced packaging is always to be respected — it's never trivial — but it's not necessarily a new thing to do," CEO and co-founder Tapa Ghosh said in an interview. "No one is going to win a Nobel Prize if our project works, but the good news is, they won't need to." The company aims to deliver a chip next year that would speed up tasks such as AI coding.
The round was led by Stripe veteran Lachy Groom and Abstract Ventures, with participation from John Doerr — the early backer of Google and Amazon — alongside VXI Capital, Triatomic and Susa Ventures. Angel investors include AI podcaster Dwarkesh Patel, AI chip veteran Naveen Rao and Anthropic researcher Sholto Douglas.
Sources: CNA
The fundamental limit for AI chips from Nvidia and AMD is bandwidth between computing cores and memory. Current packaging encircles computing chips with expensive high-bandwidth memory chips, but the tiny electrical wires connecting them can only reach so far — even Nvidia's best offerings can hold only eight memory chips per graphics processing unit. Volantis says its laser-based interconnect would eliminate the reach problem, allowing 220 memory chips to be packed around a single GPU.
The trick is to use vertical-cavity surface-emitting lasers, or VCSELs — the same tiny lasers that power facial recognition in hundreds of millions of iPhones, with a supply chain that Apple spent a decade building up. By assembling proven components rather than exotic new materials, Volantis hopes to avoid the supply-chain bottlenecks that plague AI chip development.
"Advanced packaging is always to be respected — it's never trivial — but it's not necessarily a new thing to do," CEO and co-founder Tapa Ghosh said in an interview. "No one is going to win a Nobel Prize if our project works, but the good news is, they won't need to." The company aims to deliver a chip next year that would speed up tasks such as AI coding.
The round was led by Stripe veteran Lachy Groom and Abstract Ventures, with participation from John Doerr — the early backer of Google and Amazon — alongside VXI Capital, Triatomic and Susa Ventures. Angel investors include AI podcaster Dwarkesh Patel, AI chip veteran Naveen Rao and Anthropic researcher Sholto Douglas.
Sources: CNA
