Cornelis Raises $205M to Expand Open Networking for AI Clusters

Cornelis Networks announced on Monday, September 14, 2026, that it has raised $205 million in a funding round led by IAG Capital Partners. TechCrunch

The Wayne-based company builds networking technology to help AI chips communicate more effectively. It spun off from Intel in 2020 and offers an open architecture, which means customers can link different brands of GPUs and accelerators to its networking fabric, the shared system that moves data between chips.

The funding coincides with an expansion into scale-up networking with a product called Active Compute Fabric. Scale-up covers the fast, short-distance links inside and between tightly connected groups of accelerators, distinct from scale-out links across a larger data center. The technology is designed to let chips process and send information at the same time. Cornelis says Active Compute Fabric adds distributed compute at every hop, a transfer point in the network, so congestion control, or traffic management, and collectives, or coordinated data exchanges between chips, run in the network instead of on the accelerators. The product is shipping.

Cornelis plans to use the $205 million to scale production of its CN5000 and CN6000 network switches. SiliconANGLE The company said the capital will support its expansion and bring its next generation of products to market. It is working on a new generation expected to be released later in 2026.

CN5000 SuperNICs, specialized network cards for servers, provide one or two 400 Gb/s ports on a low-profile x16 PCIe Gen 5 adapter, a standard server slot. Cornelis previously unveiled the CN6000 800 Gbps Ethernet SuperNIC for AI and HPC at scale. Its switches provide 48 ports at 400 Gb/s for 38.4 Tb/s full duplex with air or liquid cooling. Director Class Switches provide 576 ports at 400 Gb/s for up to 230.4 Tb/s in a two-tier fat tree, a branching network layout, in one chassis. The cabling options include passive copper, active copper and active optical cables at 400 Gb/s per link. Cornelis

Cornelis OPX Software is open-source software from host drivers to fabric and switch management built on libfabric, a standard interface for network communication. Cornelis states MPI, OpenSHMEM and RCCL applications, common tools for splitting work across many processors, run without code changes on its fabric. The CN5000 Omni-Path product family won the HPCwire Readers' Choice and Editors' Choice award for Best HPC Interconnect Product or Technology.

Cornelis distributed the Active Compute Fabric, $205M funding and Qualcomm collaboration announcement as a Business Wire press release titled Cornelis Expands into Scale-Up Networking with Active Compute Fabric $205M in Funding and Qualcomm Collaboration at AI Infra Summit. Business Wire The company announced a collaboration with Qualcomm and presented the fabric expansion, funding and collaboration at AI Infra Summit. In April 2026, Cornelis issued a separate Business Wire announcement about expanding its global partner ecosystem with new federal integrators and distributors to accelerate deployment of CN5000 for HPC and data-intensive workloads.

The broader context here is the split between scale-up and scale-out in AI clusters. Scale-out, connecting many servers across a data center, has received most of the Ethernet attention. Scale-up, the high-bandwidth, low-latency domain inside and across tightly coupled accelerator complexes, has remained harder to open up.

In my view, that is why Cornelis is worth watching closely. An open fabric that preserves MPI, OpenSHMEM and RCCL compatibility without code changes lowers the operational cost of trying non-incumbent hardware.

Worth flagging for buyers is the execution question. Shipping current-generation SuperNICs and switches is one test. Ramping 48-port and Director-class supply, while delivering a next-generation release later in 2026, is another. Interoperability across a variety of GPUs and accelerators, plus the Qualcomm work, will need validation in production collectives, failure handling and tail latency, not only in peak throughput figures.

Looking further ahead, if that validation lands, the long-arc effect is positive for operators. More choice in interconnects tends to improve pricing, supply resilience and architectural flexibility for HPC and AI builds. That outcome is not guaranteed. It is, however, the practical change this funding is intended to enable.