Intel spin-off Cornelis Networks raised $205 million in new funding and announced plans to enter the scale-up networking market, according to a Monday announcement from the Wayne, Pennsylvania-based startup. The company is moving beyond its traditional scale-out networking business to compete with Nvidia in an area critical to rack-scale AI computing platforms. The expansion comes with potential backing from Qualcomm, signaling a new front in the battle over AI infrastructure interconnects.
The fresh capital will fund the creation of future scale-up and scale-out products, along with manufacturing and customer rollouts for Cornelis' new CN6000 family of scale-out offerings, according to the company. Cornelis revealed a new Active Compute Fabric architecture designed to offer companies building rack-scale computing platforms an open substitute for Nvidia's proprietary NVLink interconnect technology. The startup plans to support both the Ultra Accelerator Link standard from the UALink Consortium and another open standard called ESUN for its scale-up products, while future scale-out networking products will support the Ultra Ethernet standard. Members of the UALink Consortium include Amazon Web Services, AMD, Apple, Google, HPE, Intel, and Microsoft.
"There are so many other compute solutions, accelerators, GPUs, XPUs that need scale-up capabilities as they build out their rack-level technology and need partners that are really focused on delivering to open standards," Cornelis CEO Lisa Spelman told CRN. Qualcomm's data center general manager Tony Pialis said the company's "vision for an open, programmable fabric aligns" with the need to improve utilization and AI economics through a "more integrated approach across compute, memory and networking." The collaboration between Cornelis and Qualcomm centers on integrating their technologies, with Qualcomm planning to launch its first rack-scale platform this year featuring its AI200 accelerator chip, followed by successor products in coming years.
The startup said its Active Compute Fabric will improve compute utilization within rack-scale platforms by integrating programmable compute into the architecture, allowing the network to operate on data as it moves through the system, adapt to changing workloads, offload collective operations, and take on new functions as AI algorithms and software evolve. According to Spelman, this means "driving AI-capable compute into your [network interface card] and into your switch," creating "a whole other pool of compute that is meant to deliver the AI workload." An executive at a large U.S. systems integrator told the publication that many enterprise customers would welcome new data center suppliers who embrace open standards because it could help with product choice, supply, and pricing, noting that avoiding dependence on one architecture carries cost, availability, and dependency advantages.
The timing of Cornelis' entry reflects a growing movement by major tech companies to support an alternative to Nvidia's NVLink, which the graphics chip maker has spent over a year promoting as the standard method for scale-up networking to competitors building rack-scale solutions with their own accelerator chips. Unlike Cornelis' traditional scale-out networking business focused on high-speed interconnects for servers across large clusters, the company's scale-up products may not result in channel-ready offerings directly from the startup, but could over time give solution providers more choices in the kinds of rack-scale platforms and configurations they can offer customers. For Cornelis, the challenge is clear: several companies including Amazon Web Services, Arm, and even Qualcomm have already signed up to support NVLink, giving Nvidia a way to remain a crucial supplier for the AI infrastructure market while opening another revenue stream. Still, the open architecture push could reshape how enterprises build and buy AI systems, particularly as rack-scale platforms become the dominant computing model.