Volantis Raises $88M To Advance AI Inference Architecture
Volantis has raised $88 million to develop a new architecture for AI inference. The Series A was co-led by Lachy Groom and Abstract Ventures. The round includes John Doerr, VXI Capital, Triatomic and Susa Ventures. Angel investors Dwarkesh Patel, Naveen Rao and Sholto Douglas also participated.
The San Francisco-based semiconductor company is targeting a major constraint in modern AI infrastructure. Its architecture aims to improve memory capacity and bandwidth at the same time. Volantis says this could support larger models at much higher inference speeds.
Volantis Targets The AI Memory Bottleneck
Large AI models require significant memory capacity during inference. They also need high bandwidth to continuously move data into compute systems. Current architectures force a compromise between these requirements. On-chip SRAM offers high bandwidth but provides limited memory capacity.
GPU systems use high-bandwidth memory to provide greater capacity. However, bandwidth can limit how quickly increasingly large models operate. Volantis aims to change this balance through its A-1 system. The company says A-1 could support models exceeding 20 trillion parameters.
It is also being designed to reach up to 10,000 tokens per second per user. Volantis says the architecture could also reduce inference costs per token. The company expects faster inference to benefit increasingly complex AI agents. Faster systems could allow coding agents to complete tasks in significantly less time.
Photonic Technology Connects Compute And Memory
Volantis’ photonic architecture connects compute and memory through scalable optical links designed to increase bandwidth and capacity while improving AI inference efficiency. Source: Created by Ventureburn
Volantis is developing a photonic interconnect designed specifically for compute-to-memory connections. Its optical fabric connects large numbers of memory chips into one unified pool. The approach allows bandwidth to increase as additional memory is added.
Volantis says this could increase capacity and bandwidth simultaneously. The company is using custom micro-VCSELs within its photonic platform. These components draw on an established gallium arsenide manufacturing ecosystem. Volantis says the approach avoids some supply constraints linked to indium phosphide.
Its micro-VCSELs are designed to be compact, temperature-stable and energy efficient. The company targets end-to-end links consuming less than one picojoule per bit. It also says its technology could connect up to 220 memory chips around a GPU. Volantis plans to reveal more architectural details as A-1 approaches commercialisation.
More News: Restate Raises $20M Series A to Expand Durable Execution Infrastructure
Volantis Prepares A-1 For 2027 Customer Deployments
The new funding will support A-1 development and commercialisation. It will also fund engineering expansion and customer deployment preparations. Volantis plans to deliver its first integrated inference engines to customers in 2027.
Its founding team includes veterans from NVIDIA, AMD, Broadcom and Ayar Labs. Their previous work includes major advances in packaging, VCSELs and silicon photonics. Volantis says the team is building on established technologies rather than relying on unproven breakthroughs.
CEO and co-founder Tapa Ghosh said inference speed will become increasingly important. He said AI agents will take on more work across businesses. Their completion speed could therefore influence how quickly companies operate.
The funding also arrives during continued pressure across AI memory supply chains. Rising AI demand has increased the importance of memory capacity and bandwidth. Volantis is positioning A-1 as an alternative approach to this infrastructure challenge.
The company is developing its first system, called A-1. Its photonic architecture targets the connection between compute and memory. The company now plans to move from development towards commercial customer deployments. The company combines semiconductor design with photonic technology.
To stay updated on crypto venture capital funding and market trends, visit our venture capital news section for more insights.
The post Volantis Raises $88M to Build New AI Inference Architecture appeared first on Ventureburn.