Cornelis Networks has officially launched its CN5000 networking solution on June 3, 2025, addressing one of the most significant challenges in modern AI infrastructure: keeping high-performance computing chips adequately supplied with data.
The CN5000 family represents a major advancement in AI datacenter connectivity, supporting deployments with up to 500,000 endpoints through its proprietary OmniPath technology. According to Cornelis CEO Lisa Spelman, the solution is engineered to address compute underutilization and scaling challenges that have plagued AI datacenters for years.
"Networking should do more than just move data quickly – it should unlock the full potential of every compute cycle," Spelman stated. The CN5000's architecture delivers lossless data transfer and congestion avoidance capabilities that can dramatically improve infrastructure performance and efficiency.
The technology directly competes with Nvidia's approach following its $6.9 billion acquisition of Mellanox in 2020. While Nvidia's InfiniBand technology has dominated the market, Cornelis claims its CN5000 outperforms InfiniBand NDR with 2X higher message rates, 35 percent lower latency, and up to 30 percent faster simulation times for HPC workloads.
A key advantage of the CN5000 is its vendor-neutral approach. Despite having backing from Intel, the technology is designed to work with AI computing chips from any manufacturer, including Nvidia and AMD, using open-source software. This positions Cornelis as an independent alternative in a market increasingly dominated by vertically integrated solutions.
The CN5000 will begin shipping to initial customers this month and will be broadly available starting Q3 2025 from all major OEMs. Looking ahead, Cornelis has already announced that its next-generation chips in 2026 will also be compatible with Ethernet networks, further expanding its interoperability.