- Volantis has raised $88 million in Series A funding co-led by Lachy Groom and Abstract Ventures to develop a new photonic architecture for AI inference.
- Its A-1 system is designed to run models exceeding 20 trillion parameters at up to 10,000 tokens per second per user, while reducing inference costs.
- The San Francisco semiconductor startup plans to deliver its first integrated inference engines to customers in 2027 as AI workloads put increasing pressure on memory infrastructure.
Volantis has raised $88 million in Series A funding. The round was co-led by Lachy Groom and Abstract Ventures, with participation from John Doerr, VXI Capital, Triatomic and Susa Ventures. Angel investors including Dwarkesh Patel, Naveen Rao and Sholto Douglas also participated.
Volantis will use the funding to develop and commercialise its A-1 inference system, expand its engineering team and move towards customer deployments. The company expects to deliver its first integrated inference engines in 2027.
Building A-1 for trillion-parameter models
Volantis was founded by Tapa Ghosh and Roy Meade. Ghosh is a Thiel Fellow and former Y Combinator founder with four patents, while Meade previously led Micron’s HBM programme and served as a vice president at Ayar Labs.
“As AI agents take on more work, how fast they complete that work will increasingly determine how fast companies can operate,” said Tapa Ghosh, CEO and co-founder of Volantis. “Today’s hardware forces a tradeoff between running the largest, most sophisticated models and running them fast. We started Volantis to eliminate that tradeoff.”
Volantis says A-1 is being designed to support models exceeding 20 trillion parameters at up to 10,000 tokens per second per user. These figures represent the company’s product targets as it moves towards commercialisation, rather than performance from a deployed system.
Rewiring the connection between compute and memory
Large AI models require enormous memory capacity to hold model weights and context, alongside high bandwidth to continuously move that data into compute. Existing architectures typically force a compromise.
On-chip SRAM delivers high bandwidth but has limited capacity, while GPUs and other systems using high-bandwidth memory offer substantially more capacity but remain constrained by bandwidth, cost and energy requirements. Volantis argues that even emerging approaches such as 3D DRAM remain on the same underlying trade-off curve.
Its solution is a photonic interconnect designed specifically to connect compute chips with memory. The company’s optical fabric creates a unified pool of memory, aggregating bandwidth as memory is added. This allows Volantis to increase memory capacity and bandwidth together, rather than scaling one at the expense of the other.
The architecture uses custom micro-VCSELs instead of external lasers. Volantis says this approach leverages the established gallium arsenide VCSEL supply chain while avoiding some of the constraints associated with indium phosphide. The company targets end-to-end optical links consuming less than one picojoule per bit.
The photonics race heats up
Volantis enters a growing AI infrastructure market attracting major funding. Lightmatter raised $400 million at a $4.4 billion valuation to develop photonic computing and interconnect technologies for AI data centres. Ayar Labs raised $500 million in Series E funding at a $3.8 billion valuation, with its optical interconnect technology designed to tackle data movement between chips.
Celestial AI raised $250 million at a $2.5 billion valuation for its Photonic Fabric platform, which targets bandwidth and data-movement bottlenecks in AI infrastructure.
Volantis is approaching the problem from the memory side, betting that photonics can help AI systems scale both capacity and bandwidth as models become larger and inference workloads more demanding.
With A-1 systems targeted for customer delivery in 2027, the next test will be turning that architecture into a commercially deployable inference platform.
