Cornelis raises $205 million for a network fabric that computes while it moves data

Cornelis Networks raised $205 million in a round led by IAG Capital Partners, the company said, as it launched a new networking product called Active Compute Fabric.

The fabric embeds programmable compute directly into the network rather than just forwarding packets, handling tasks such as assembling data for disaggregated inference, coordinating requests across mixture-of-experts models, accelerating collective operations and compressing data in transit, according to Cornelis and SiliconANGLE. Pre-production simulations showed the approach cutting overall network traffic by up to 50%, per SiliconANGLE.

Cornelis, which spun out of Intel in 2020, also announced a strategic collaboration with Qualcomm to jointly evaluate rack-scale inference hardware, TechCrunch and SiliconANGLE reported. CEO Lisa Spelman said "the fabric has to become an active part of the compute system," and CMO Brandon Draeger said accelerator utilization typically sits "near half of installed capacity."

The company plans to use the funding to scale production of its CN5000 and CN6000 switches and speed up deployment of the new fabric, TechCrunch reported.

Cornelis is one of several networking startups betting that as GPU clusters get more expensive, wasted GPU time waiting on data becomes the bigger cost to fix, not the chips themselves.