Cornelis launches 800G networking to take on Nvidia’s InfiniBand
Cornelis, an Intel’s 2020 spin-off, has rolled out 800G network cards and switches for AI clusters. The company says its CN6000 family beats Nvidia’s InfiniBand by 21% in latency for small messages, though independent tests are still pending. Dell, Lenovo, and Supermicro plan to start selling systems with the new cards in early 2027.
Editor, Lazyfounder
30 SEC SUMMARY
- Cornelis, an Intel spin-off, has launched 800G network cards and switches aimed at AI and supercomputing clusters, challenging Nvidia’s InfiniBand dominance.
- The CN6000 family claims a 21% latency edge over Nvidia’s InfiniBand for small-message passing, though independent tests are pending.
- The hardware revives Intel’s abandoned Omni-Path technology and supports chips from AMD, Intel, Nvidia, and Qualcomm.
- Dell, Lenovo, and Supermicro will begin selling systems with the CN6000 in early 2027, offering a multi-vendor alternative for AI infrastructure.
TABLE OF CONTENTS
CN6000 aims to dethrone Nvidia’s InfiniBand
Cornelis, the networking company Intel spun out in 2020, has unveiled its CN6000 family of 800G network cards and switches. These are built for AI and supercomputing clusters and designed to rival Nvidia’s InfiniBand, which currently leads the market for high-performance networking.
Latency claims need real-world validation
Inside Cornelis’s labs, the CN6000 cards hit 1.6 billion messages per second and finished a small-message benchmark 21% faster than Nvidia’s InfiniBand. These numbers come from Cornelis’s own tests—no outside lab has verified them yet.
Even if a single message arrives faster, real-world jobs may not finish 21% sooner. The gain depends on how much the workload relies on these small messages.
Omni-Path returns, now open to all GPUs
The CN6000’s main card uses Omni-Path, a technology Intel ditched in 2019 before spinning out Cornelis. Now, Cornelis has dusted it off and reshaped it for AI workloads.
Unlike Nvidia’s closed system, the cards work with GPUs from AMD, Intel, Nvidia, and Qualcomm. That gives data centers the option to mix and match vendors without locking into Nvidia’s stack.
First partner systems arrive in early 2027
Dell, Lenovo, and Supermicro will integrate the CN6000 into their AI servers and storage systems, with the first models shipping in early 2027. For enterprises, that timeline sets a clock: benchmarks and procurement cycles will need to kick off in late 2026 to meet deployment windows.
Lisa Spelman, Cornelis CEO, says the 2027 launch window gives customers a viable alternative just as AI clusters scale beyond single-vendor solutions.
Nvidia faces new rivals in AI networking
Cornelis is not alone in targeting Nvidia’s grip on AI networking. Huawei has its own high-speed fabric, and startups like Upscale AI are chasing similar speeds.
AMD, a Cornelis partner, is also assembling rack-scale AI systems to compete with Nvidia’s full-stack approach. That means by 2027, data centers will have at least three enterprise-grade networking options.
What this means
Lazyfounder analysis — our interpretation, not reported fact.
Cornelis’s 21% latency claim could push Nvidia to sharpen its pricing or accelerate InfiniBand updates if the results hold in real-world tests. For AI startups and enterprises, the CN6000 means a second major supplier by 2027, reducing reliance on a single vendor for high-performance networking.
Key takeaways
- Cornelis’s CN6000 cards and switches aim to break Nvidia’s stranglehold on AI networking with 800G speeds and a claimed 21% latency edge.
- The hardware works with chips from AMD, Intel, Nvidia, and Qualcomm, giving data centers a way to mix vendors.
- Dell, Lenovo, and Supermicro will roll out CN6000-based systems in early 2027, offering a timeline for enterprises to diversify.
- The revival of Intel’s Omni-Path shows how spin-offs can resurrect shelved technologies when demand spikes.
Related on Lazyfounder
Sources
- The Next Web · 2026-11-13
Cornelis launches CN6000 800G networking to rival Nvidia InfiniBand
This story is an original summary drafted with AI by Lazyfounder from the reporting listed above and checked by automated validation. Facts are attributed to their original publishers; sections marked as analysis are Lazyfounder's. Where a source is in another language, facts were machine-translated and quotations are reported, not reproduced. Read the original coverage via the links, and see our AI policy and corrections policy.
About the author
Editor, Lazyfounder
Tarun Mottlia edits LazyFounders, covering Indian startups, funding rounds, AI and product launches. Every story on the site is AI-assisted and checked against its cited sources before publication.
More stories by Tarun MottliaGet the LazyFounder Brief
Startup, funding and AI news in a five-minute read. Join the early-access list.
