Cornelis, the AI and high-performance computing networking company spun out of Intel, has raised $205 million to take on Nvidia's grip on data center networking. The round was led by IAG Capital Partners, according to SiliconANGLE.

The Wayne, Pennsylvania-based company announced the funding at the AI Infra Summit alongside a new architecture called Active Compute Fabric, and a collaboration with Qualcomm Technologies on rack-scale AI data center networking.

From Intel Lab to Nvidia Challenger

Cornelis was formed in 2020 when the team behind Intel's Omni-Path interconnect spun out as an independent company. It is now led by CEO Lisa Spelman, a longtime Intel executive who took the top job in 2024.

The pitch is openness. Nvidia sells a tightly integrated stack of GPUs, networking and software. Cornelis is building on open standards, including Ethernet, UALink for scale-up connections inside a rack and Ultra Ethernet for scale-out connections across racks, so data center operators are not locked into a single vendor.

Putting Compute in the Network

The core problem Cornelis is attacking is idle GPU time. In large AI clusters, expensive chips often sit waiting for data to arrive, as TechCrunch noted in its coverage of the round.

Active Compute Fabric embeds programmable compute, including RISC-V cores, directly into the network fabric so it can perform workload-specific operations while data is in transit. Target workloads include KV cache acceleration and mixture-of-experts routing, two bottlenecks in modern AI inference.

"The payload does not arrive the way it left," Chief Marketing Officer Brandon Draeger said, describing how the fabric processes data on the way to its destination. Cornelis says pre-production simulations show up to a 50% reduction in network traffic.

Where the Money Goes

Cornelis will use the capital to scale production of its CN5000 platform, which began shipping in 2025, and its CN6000 line, including an 800-gigabit SuperNIC. The funding also supports the company's expansion into scale-up networking and the rollout of Active Compute Fabric, with a new product generation expected later this year.

The raise dwarfs the $29 million Series B Cornelis announced earlier in its life, a sign of how much investor appetite has grown for the infrastructure layer around AI chips. With hyperscalers and neoclouds spending heavily to squeeze more output from every GPU, networking has become one of the most contested corners of the AI buildout.

Cornelis also argues that better networking changes the economics of AI clusters. If operators can keep GPUs busy instead of waiting on data, they can serve more inference requests or finish training runs sooner on the same hardware, a direct way to improve returns on some of the most expensive equipment in the data center.

The Qualcomm collaboration gives Cornelis a high-profile partner as it moves toward rack-scale systems, where networking, compute and memory are increasingly designed together rather than bought as separate parts.

Whether an open, standards-based approach can pry customers away from Nvidia's integrated systems is the question Cornelis now has $205 million to answer.

Source

TechCrunch — "AI infrastructure company Cornelis raises $205M to chip away at Nvidia's dominance"