Cornelis Networks has secured $205 million in funding to expand its Active Compute Fabric, an architecture that integrates RISC-V cores into network interface cards and switches. The technology aims to offload AI tasks like KV cache acceleration directly into the network fabric to improve GPU utilization for AI and HPC workloads.
This capital injection signals a shift toward intelligent networking where the fabric does more than move data; it actively processes it. By embedding RISC-V compute into the interconnect, Cornelis addresses the critical bottleneck of GPU underutilization, which currently hovers around 50% for many AI workloads. For the broader streaming and AI ecosystem, this provides a high-performance alternative to Nvidia's closed NVLink architecture, favoring open standards like UALink. As infrastructure costs for generative AI soar, efficiency gains in the network layer become a primary lever for maintaining margins. Watch for the adoption rate of the CN6000 among non-hyperscale enterprise and government labs seeking to optimize heterogeneous compute clusters.
Cornelis Networks entered the AI interconnect market with its Omni-Path heritage, but the competitive landscape has shifted dramatically. Nokia and Ericsson have diverged sharply on AI-RAN strategy, with T-Mobile US testing both approaches, illustrating how even telecom operators are weighing competing architectures for AI-driven infrastructure. In the data center interconnect space, the parallel tension is between Nvidia's proprietary NVLink and open alternatives like UALink, where Cornelis positions its Active Compute Fabric as a differentiated option that adds processing capability directly into the network path rather than relying solely on bandwidth scaling.
The business case for in-network compute is gaining traction among infrastructure investors and operators alike. Nokia announced partnerships with AWS and Databricks to build a unified data and control layer for autonomous networks, demonstrating that major vendors are stacking cloud, data, and orchestration components into integrated platforms. Nokia's Autonomous Network Fabric, which uses agents and digital twins to deliver observability and automation, represents the same architectural philosophy Cornelis applies at the silicon level: moving intelligence closer to where data moves rather than centralizing all processing. Nokia reported operators achieving automation rates above 90 percent and service delivery times under four hours with its current portfolio, suggesting the market rewards architectures that reduce latency between decision and action.
On the technical side, the performance claims from Cornelis align with broader industry measurements of GPU underutilization. Ericsson launched its AI in RAN commercial software subscription in June 2026, claiming up to 20 percent higher downlink throughput and 10 percent better spectral efficiency across more than 15 live deployments, showing that software-level optimizations in the network layer can yield double-digit efficiency gains without new hardware. Nokia is deploying agentic AI into its mobile core, with executives citing call setup time reductions from roughly 10 seconds to one or two seconds when AI handles paging and location tasks at the edge. These results reinforce the thesis that distributing compute into the network path, whether at the RAN, core, or data center fabric level, produces measurable latency and utilization improvements. For Cornelis, the CN5000 and CN6000 products must demonstrate similar concrete gains in AI networking bottlenecks to justify adoption over incumbent solutions from Nvidia and Broadcom. As the industry matures, NVIDIA NodeWright is also emerging as a key tool for managing the complexity of these large-scale GPU deployments. Recent market activity also highlights that optical interconnect hardware is becoming a critical component for linking AI accelerators at scale.
Cornelis Networks has raised $205 million to scale its Active Compute Fabric, which embeds RISC-V cores into network silicon. By offloading AI tasks like KV cache acceleration from GPUs to the network, the company aims to increase GPU utilization by 5% to 10%, offering an open-standard alternative to Nvidia's proprietary NVLink ecosystem.
The $205 million funding round is intended to scale the company's Active Compute Fabric, an architecture that embeds RISC-V cores directly into network silicon to perform AI acceleration tasks.
It offloads specific AI tasks, such as mixture of experts routing and KV cache acceleration, from the GPU to the network, aiming to reduce idle time and increase overall GPU utilization by 5% to 10%.
The CN6000 is an 800 gigabit SuperNIC launched by Cornelis Networks that supports both standard Ethernet protocols and proprietary Omni-Path technology.
Cornelis Networks is expanding into scale-up networking using open standards such as UALink and ESUN to compete with proprietary ecosystems like Nvidia's NVLink.
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source