Cornelis Networks Raises $205 Million to Challenge Nvidia's AI Networking Dominance

Cornelis Networks announced on September 14-15, 2026 a $205 million funding round led by IAG Capital Partners to fund its expansion into scale-up AI networking with a new architecture called Active Compute Fabric.

Making the Network Itself Do More Work

Active Compute Fabric is an open architecture combining programmable compute, lossless transport, and in-fabric acceleration across both scale-up and scale-out networks — offloading collective operations and other processing tasks to the network itself rather than requiring AI accelerator chips to handle everything, and adapting network behavior to workloads as AI software evolves.

Specific Performance Claims

• The 400Gbps CN5000 switch is shipping now, claiming 2x the message rate and 35% lower latency compared to other 400Gbps solutions
• Cornelis claims 6x faster collective communications compared to RoCE implementations
• The 800Gbps CN6000, adding Ethernet support without losing the proprietary architecture, is sampling with customers ahead of expanded availability in Q4 2026

A Qualcomm Partnership on Rack-Scale Infrastructure

Cornelis is also collaborating with Qualcomm Technologies around the direction of rack-scale AI infrastructure — a partnership that positions Cornelis's networking approach alongside an alternative accelerator ecosystem rather than solely serving Nvidia-based AI clusters.

What This Means for AI Infrastructure Buyers

A well-funded challenger making specific, measurable performance claims against the dominant AI networking approach is a concrete signal that AI cluster interconnect — often treated as a secondary decision behind accelerator chip selection — is becoming a competitive battleground in its own right, worth independent evaluation rather than assuming the default networking choice that comes bundled with a given AI chip vendor.