Cornelis Networks is taking its networking business deeper into AI infrastructure with a new fabric architecture aimed at giving system builders an alternative to proprietary interconnects such as Nvidia’s NVLink. The company introduced Active Compute Fabric along with $205 million in new funding to support its expansion and next-generation networking products.

Cornelis argues that as AI clusters grow, GPUs and other accelerators can spend more time waiting for data or for other processors to catch up, leaving expensive hardware underused. Active Compute Fabric is designed to address this by making the network part of the compute system rather than a transport layer. The architecture combines lossless transport, in-fabric acceleration and programmable compute, allowing it to process data as it moves, adapt to changing workloads and offload collective operations that would otherwise consume accelerator resources. Cornelis says the goal is to keep accelerators busier by shifting some of these tasks into the fabric.

Active Compute Fabric uses UALink and Ethernet for Scale-Up Networking, or ESUN, for scale-up connectivity and Ultra Ethernet specifications for scale-out. UALink is a newer open standard for connecting large numbers of AI accelerators in a tightly coupled system, while the Open Compute Project’s ESUN project adapts Ethernet for scale-up AI workloads. Together, the technologies drive Cornelis’s open standards-based approach, which CEO Lisa Spelman said is intended to give customers more choice and reduce dependence on proprietary architectures.

Cornelis Networks emerged in 2020 from Intel’s former Omni-Path interconnect business. The company develops networking technology for AI and HPC, including its current CN5000 Omni-Path platform and the upcoming CN6000 generation, which is now sampling with customers and is expected to become more widely available in Q4 2026. Last month, Cornelis expanded a collaboration with NEC after NEC evaluated CN5000 for HPC workloads and agreed to participate in early CN6000 evaluations.

Cornelis said the $205 million in new funding will support its expansion into scale-up networking and development of future scale-up and scale-out products. The company also plans to use the capital to scale production, support manufacturing and customer deployments of CN6000, deepen customer partnerships and accelerate its go-to-market efforts. The announcement did not disclose the funding structure, valuation or a full list of investors.

Separately, Cornelis said it is working with Qualcomm Technologies to validate future AI data center designs. Qualcomm says its upcoming Dragonfly AI300 inference platform will use UALink and ESUN for scale-up connectivity, the same open standards behind Cornelis’ Active Compute Fabric, although Cornelis has not identified the specific products involved in that validation work or provided a timetable. Qualcomm data center chief Tony Pialis is scheduled to join Cornelis CEO Lisa Spelman during her AI Infra Summit keynote this week to discuss networking’s role in AI infrastructure.

Cornelis has not provided a timeline for products based on Active Compute Fabric. Spelman is expected to provide more detail on the architecture at the summit, where the company also plans to demonstrate CN6000 and preview its next-generation networking roadmap.