Cornelis Networks

Keeping AI and HPC accelerators fed with data

Cornelis Networks builds end-to-end networking fabrics for AI and HPC systems, combining adapters, switches, software and cabling. Its shipping CN5000 Omni-Path fabric uses adaptive routing and congestion management to move data across large clusters with low latency. The company is also developing Active Compute Fabric capabilities that perform more communication work inside the network; some acceleration features are in development, distinct from the product generation already shipping.

Cloud providers, enterprise AI operators, research centers, national laboratories and supercomputing teams deploy Cornelis fabrics to connect compute nodes. The fabric coordinates data movement across the cluster and reduces communication bottlenecks that can leave expensive GPUs waiting.