Cornelis Raises $205M to Build Open, Programmable Networking for AI Infrastructure

AI infrastructure networking company Cornelis has raised $205 million in funding as it expands from scale-out networking into scale-up networking and introduces its new Active Compute Fabric architecture.

Based in Wayne, Pennsylvania, Cornelis is developing networking technology designed to improve the efficiency of increasingly large AI and high-performance computing systems. The company says its latest architecture brings programmable compute directly into the network fabric rather than limiting networking infrastructure to simply moving data between endpoints.

Making the network part of the compute system

As AI clusters become larger, communication and synchronization between accelerators can become a significant source of wasted computing capacity.

Cornelis' Active Compute Fabric combines lossless transport, in-fabric acceleration, and programmable compute. This allows the network to process data as it moves through the system, support collective operations, adapt to workloads, and potentially take on new functions as AI algorithms evolve.

The company argues that this approach can improve accelerator utilisation by moving certain workloads into the network fabric instead of allowing GPUs to remain idle while waiting for data.

Cornelis estimates that in a 100,000-GPU system, around half of GPU hours could be spent waiting for data. Based on its modelling, the company estimates this could represent approximately $1.68 billion in annual wasted capacity and 500 GWh of power.

The company notes that these figures are projections based on pre-production modelling and public industry data rather than measured production results.

Open standards and scale-up networking

Cornelis is positioning Active Compute Fabric around open industry standards rather than a proprietary infrastructure stack.

The architecture supports UALink and ESUN for scale-up networking and Ultra Ethernet specifications for scale-out networking, while being designed to work across a broad range of accelerators.

The company's move into scale-up networking is intended to give customers greater flexibility when building rack-scale AI systems.

Qualcomm collaboration

Cornelis is also collaborating with Qualcomm Technologies around the future of AI infrastructure networking.

Tony Pialis, executive vice president and general manager of Data Center at Qualcomm Technologies, will join Cornelis CEO Lisa Spelman during her keynote at AI Infra Summit.

The companies share the view that networking will become increasingly important as AI systems move toward rack-scale architectures, particularly as operators look for ways to improve accelerator utilisation and reduce the economics of idle compute.

$205M backs the next phase

The newly announced $205 million will support Cornelis' expansion into scale-up networking and the development and commercialisation of its next-generation products.

The company plans to use the capital to scale production, deepen customer partnerships, and accelerate its go-to-market strategy.

Cornelis says its CN5000 is already shipping, while the CN6000 is sampling with customers, with expanded availability expected in Q4 2026.

The company will also preview its next-generation scale-up and scale-out networking roadmap at AI Infra Summit.

The bigger AI infrastructure opportunity

Cornelis believes open-standard networking across both scale-up and scale-out environments could represent a more than $55 billion opportunity by 2030.

As AI infrastructure becomes increasingly dominated by massive GPU clusters, improving the efficiency of communication between accelerators could become just as important as increasing raw compute performance.

Cornelis' strategy is therefore built around a simple premise: the network should become an active component of the AI compute system, not merely the infrastructure connecting it.