Cornelis Networks Inc. introduced a new data center network architecture called Active Compute Fabric, alongside a strategic collaboration with Qualcomm Technologies Inc. and a $205 million funding round led by IAG Capital Partners. The company announced the fabric at the AI Infra Summit. Cornelis develops congestion-free networking systems for artificial intelligence and high-performance computing workloads, competing with established providers such as Cisco Systems Inc. and Arista Networks Inc.
Active Compute Fabric is an open architecture that places programmable compute inside the network fabric itself, rather than leaving the network to only forward data packets. Cornelis said the rapid growth in AI model compute cluster sizes has created bottlenecks that leave expensive AI accelerators idle while they wait for data. Chief Executive Lisa Spelman said compute, memory and storage have become more workload-aware in recent years, while traditional networks have not made the same change.
The architecture combines in-fabric acceleration with lossless transport and programmable compute, allowing the network to work on data as it moves and to adapt in real time to shifting workloads. It is built on open standards including Ethernet and UALink for scale-up and Ultra Ethernet for scale-out, so organizations can deploy it with existing compute architectures. Chief Marketing Officer Brandon Draeger said the fabric can assemble KV cache data for disaggregated inference, coordinate expert dispatch for mixture-of-experts models and accelerate collective operations such as AllReduce, compressing gradients in transit before data reaches its destination.
Draeger said partial results from many endpoints can be combined inside the fabric so a single reduced result arrives instead of thousands of separate contributions, and compressed gradients move at a fraction of their original size. He said this cuts network traffic by up to 50 percent, based on Cornelis pre-production simulations, and reduces the time GPUs spend waiting on communications and synchronization. Accelerator utilization in large AI deployments commonly sits near half of installed capacity, and the architecture aims to recover a meaningful share of that waste, though gains will vary by model, cluster size and customer stack.
Spelman said Cornelis is expanding into scale-up and scale-out to bring its network architecture closer to the AI accelerators it feeds. Qualcomm will join Cornelis at the AI Infra Summit in Santa Clara to announce the collaboration, which targets networking for rack-scale AI data centers. Draeger said the companies are in advanced stages of joint technology evaluation focused on rack-scale inference across scale-up and scale-out environments.
Cornelis and Qualcomm argue the network, now about 15 percent of the cost of an AI system, should be a first-order design decision made alongside the accelerator rather than after the compute choice. Cornelis will use the $205 million to scale production of its CN5000 and CN6000 network switches and accelerate deployment of Active Compute Fabric.
More company and startup news from TechManNews.
