A San Francisco semiconductor startup says it has found a way to blast past one of artificial intelligence's biggest hardware headaches — not with faster chips, but with light. Volantis announced Thursday that it raised $88 million in a Series A funding round to build technology that uses laser beams instead of copper wires to connect computing chips to memory chips.
The round was co-led by Lachy Groom and Abstract Ventures, with participation from John Doerr, VXI Capital, Triatomic and Susa Ventures, according to a company announcement. The round also included angel investments from Dwarkesh Patel, Naveen Rao and Sholto Douglas, per reporting from Reuters. The fresh capital arrives as chipmakers across the industry scramble to solve what's become known as the AI memory wall.
Why Memory, Not Just Processing Power, Is the Bottleneck
Today's AI systems face a memory-bandwidth challenge: connecting processors with enough high-bandwidth memory is a key part of scaling performance.
Volantis says its approach sidesteps that constraint entirely. Its optical fabric connects large numbers of memory chips into a unified pool, aggregating their bandwidth as memory is added, the company said in its announcement. Reuters reports that Volantis's technology would let a single GPU pack 220 memory chips — nearly 28 times what Nvidia's current chips can hold.
Betting on Light Over Copper
The company's system relies on vertical-cavity surface-emitting lasers, or VCSELs, a type of laser technology. Volantis says its custom micro-VCSELs enable end-to-end links consuming less than one picojoule per bit, a power-efficiency figure detailed in its announcement.
Volantis says its A-1 architecture is designed to run models exceeding 20 trillion parameters at up to 10,000 tokens per second per user, while simultaneously increasing memory capacity and bandwidth by nearly two orders of magnitude, according to the company. Separately, Embedded.com reports that Volantis's photonic motherboard is designed to pool more than 100 terabytes of memory per chip at bandwidth above 30 terabytes per second. The outlet also reports that the company's first-generation motherboard uses silicon nitride wafers, with a next generation planned to use glass substrates, and that Volantis says it has protected its innovations with 25 patents.
Timeline: Samples This Year, Volume Next
Volantis plans to deliver its first integrated inference engines to customers in 2027, the company said in its funding announcement. Embedded.com reports a fourth-quarter 2026 market timeline for the first-generation motherboards. Reuters frames the 2027 timeline as Volantis aiming to deliver a chip that would speed up AI coding and other tasks.
Tapa Ghosh, the company's CEO and co-founder, framed the challenge as one of engineering discipline rather than scientific breakthrough. Ghosh told Reuters, “Advanced packaging is always to be respected — it's never trivial — but it's not necessarily a new thing to do.” Ghosh also told the outlet, “No one is going to win a Nobel Prize if our project works, but the good news is, they won't need to.”
Volantis was founded in 2022 by Ghosh and chief technology officer Roy Meade, according to Optics.org. The company previously raised a $9 million seed round, as reported by Optica-OPN, before closing Thursday's much larger Series A.
A Crowded Race to Replace Copper With Light
Volantis isn't alone in betting that light will replace copper inside AI data centers. Industry analysis from SemiEngineering forecasts that all high-bandwidth data-center interconnects could become optical within the next five years, with co-packaged optics expected to replace pluggable transceivers for scale-out connections and copper for scale-up connections.
Volantis faces well-funded rivals chasing the same prize. Lightmatter expects to ship co-packaged-optics parts built at TSMC and GlobalFoundries in 2028, according to Tom's Hardware. SemiEngineering also reports that Ayar Labs demonstrated an all-optical scale-up rack system alongside Taiwan's Wiwynn at the OFC conference in March.
