Volantis, a San Francisco semiconductor company building an AI inference system, said on October 1, 2026 that it had raised an $88 million Series A co-led by Lachy Groom and Abstract Ventures. The company said the money will fund development and commercialization of its first system, called A-1, and its photonic memory architecture.
Who is backing the round
According to the company, the round also drew participation from John Doerr, VXI Capital, Triatomic and Susa Ventures, along with angel investors Dwarkesh Patel, Naveen Rao and Sholto Douglas.
Volantis said its founding team includes semiconductor and photonics veterans from NVIDIA, AMD, Broadcom and Ayar Labs, and that their previous work includes the first CoWoS product, the first high-volume tunable VCSELs and early silicon photonics co-packaged optics systems.
What the company says it is building
Volantis said it is designing an AI inference architecture that eliminates the tradeoff between memory capacity and bandwidth. The company said its first system, A-1, is being designed to run models exceeding 20 trillion parameters at up to 10,000 tokens per second per user, while reducing inference cost per token. These are the company's claims; the system has not shipped.
The company described the underlying problem as a compromise forced by existing hardware: on-chip SRAM offers high bandwidth but limited capacity, while GPUs and other HBM-based systems offer greater capacity but with bandwidth limits. Volantis said even newer approaches such as 3D DRAM remain on the same tradeoff curve.
Volantis said A-1 is designed to increase memory capacity and bandwidth simultaneously by nearly two orders of magnitude.
The photonic interconnect
Volantis said it is building a category of photonic interconnect designed specifically to connect compute chips to memory, using an optical fabric that links large numbers of memory chips into a unified pool and aggregates their bandwidth as memory is added. The company said this lets A-1 use lower-cost off-chip memory.
According to Volantis, existing data center photonics has largely focused on chip-to-chip connections, while chip-to-memory connections require more than 100 times as much data to travel over much shorter distances, creating different energy and cost requirements.
The company said its architecture uses custom micro-VCSELs rather than external lasers, drawing on the existing gallium arsenide VCSEL supply chain and avoiding indium phosphide supply constraints. It said the micro-VCSELs are small, temperature-stable and low power, enabling end-to-end links consuming less than one picojoule per bit, and that further details will be unveiled as A-1 moves toward commercialization.
Executive comment and timeline
"Today's hardware forces a tradeoff between running the largest, most sophisticated models and running them fast. We started Volantis to eliminate that tradeoff," said Tapa Ghosh, CEO and co-founder of Volantis.
Volantis said it plans to deliver its first integrated inference engines to customers in 2027, and that the financing will support expanding its engineering team and moving the system toward customer deployments.
Key facts and where they come from
- Volantis announced an $88 million Series A co-led by Lachy Groom and Abstract Ventures.
today announced an $88 million Series A co-led by Lachy Groom and Abstract Ventures
- Other investors include John Doerr, VXI Capital, Triatomic, Susa Ventures and angels Dwarkesh Patel, Naveen Rao and Sholto Douglas.
with participation from John Doerr, VXI Capital, Triatomic and Susa Ventures. The round also includes angel investors Dwarkesh Patel, Naveen Rao and Sholto Douglas.
- The first system, A-1, is designed for models exceeding 20 trillion parameters at up to 10,000 tokens per second per user.
A-1, is being designed to run models exceeding 20 trillion parameters at up to 10,000 tokens per second per user, while reducing inference cost per token
- Volantis claims A-1 will raise memory capacity and bandwidth together by nearly two orders of magnitude.
Volantis is designing A-1 to increase memory capacity and bandwidth simultaneously by nearly two orders of magnitude
- The architecture uses custom micro-VCSELs instead of external lasers, based on the gallium arsenide supply chain.
Its architecture uses custom micro-VCSELs rather than external lasers, drawing on the existing gallium arsenide VCSEL supply chain and avoiding indium phosphide supply constraints.
- The company says end-to-end links consume less than one picojoule per bit.
enabling end-to-end links consuming less than one picojoule per bit
- First integrated inference engines are planned for customers in 2027.
Volantis plans to deliver its first integrated inference engines to customers in 2027.
- Founders come from NVIDIA, AMD, Broadcom and Ayar Labs.
The company's founding team includes semiconductor and photonics veterans from NVIDIA, AMD, Broadcom and Ayar Labs.
Read the original from Company press releases (PR Newswire) →