Cornelis Networks offered a glimpse into its latest SuperNIC, the CN6000 – an 800 Gb/s Ethernet adapter designed to support AI and high-performance computing (HPC) networking.

Teased during the recent SC25 event, the CN6000 offers double the performance of the previous-generation CN5000, with Cornelis claiming it can support up to 1.6 billion messages per second.

Powering the new SuperNIC is a “hardware-accelerated” iteration of RoCEv2 (remote direct memory access over converged Ethernet version 2). The underlying component can track millions of concurrent RoCEv2 operations, removing bottlenecks often seen in traditional RoCEv2 architectures deployed in large GPU clusters due to memory-heavy overheads.

Cornelis claims the addition “helps organizations accelerate AI training, maximize infrastructure utilization, and achieve faster results, all while reducing data center power and operational costs.”

The SuperNIC is also Ultra Ethernet compliant, ensuring the hardware can be seamlessly integrated into consortium-supporting Ethernet-based infrastructures.

“As AI factories scale toward gigawatt-sized deployments, traditional Ethernet RoCEv2 architectures are reaching their limits,” Cornelis Networks CEO Lisa Spelman explained. “With the CN6000 SuperNIC we’re eliminating that bottleneck by bringing our proven high-performance architecture to Ethernet for the first time, combining limitless RoCEv2 scalability with support for Ultra Ethernet and Omni-Path.”

Production deployments of the CN6000 are scheduled to follow customer sampling, which is anticipated to commence in mid-2026.

In addition to next-gen SuperNICs, Corenlis confirmed it’s working on a new family of CN6000-series-based Omni-Path switches for HPC and AI. Details on those switches were light at the time of writing.

Shortly before the unveiling of its CN6000, its prior-generation CN5000s were picked to support a new computing cluster at Lawrence Livermore National Laboratory (LLNL).

Cornelis’s hardware will support Lynx, an upcoming supercomputing cluster that’s being used as part of the Advanced Simulation and Computing Program (ASC), which performs continual assessments and certification on the U.S. nuclear stockpile.

Public details on the program’s newest cluster are limited, though an LLNL official revealed to SDxCentral in July that it’ll be powered by Dell’s C6620 and R760 servers, which feature Intel Sapphire Rapids CPUs. Lynx is expected to be delivered in fall 2026.