Cornelis Networks, an Intel spin-off, has raised $205 million in a funding round led by IAG Capital Partners to scale its open-architecture AI networking technology. Alongside the funding, the company unveiled its Active Compute Fabric, which aims to eliminate costly GPU underutilization by transforming the network from a passive transport layer into an active participant in computation. Cornelis is also collaborating with Qualcomm to optimize large-scale AI inference and challenge proprietary vendor lock-in.
Cornelis $205 million funding round
- ▪Cornelis Networks announced on September 14, 2026, that it raised approximately $205 million in a funding round led by IAG Capital Partners
- ▪Cornelis Networks plans to use the $205 million in new funding to scale production, expand customer engagements, and accelerate its go-to-market efforts
Active Compute Fabric technology
- ▪Cornelis Networks plans to preview its next-generation CN7000 series, which will feature RISC-V-based cores and SRAM on every NIC and network switch controller to support the Active Compute Fabric
- ▪Cornelis Networks announced its Active Compute Fabric technology, a network technology designed to let chips process and send information simultaneously
- ▪The Active Compute Fabric architecture relies on three pillars: lossless transport to minimize congestion, in-fabric acceleration to offload collective operations, and programmable compute
AI network bottleneck issues
- ▪Traditional data center networks act as passive transport layers that fail to keep pace with the massive bandwidth and synchronization requirements of modern AI workloads
- ▪When network fabrics cannot handle heavy AI workloads, GPUs and other accelerators sit idle, creating a significant bottleneck that erodes return on investment
GPU underutilization costs
- ▪A data center with 20,000 GPUs utilized only 70% of the time loses the performance equivalent of 6,000 GPUs, representing tens of millions of dollars in stranded value
- ▪Underutilized accelerators directly increase training times, raise inference costs, and lower overall system efficiency in AI data centers
Open architecture networking approach
- ▪Cornelis Networks, which spun off from Intel in 2020, offers an open architecture that allows customers to use a variety of GPU and accelerator hardware with its networking fabric
- ▪Cornelis Networks is positioning itself as an open alternative to vendor lock-in by grounding its architecture in open standards, including UALink, Ethernet, and Ultra Ethernet
Qualcomm collaboration
- ▪Tony Pialis, lead of Qualcomm's data center business, will join Cornelis Networks CEO Lisa Spelman at the AI Infra Summit to discuss data center efficiency and inference economics
- ▪Cornelis Networks announced a public collaboration with Qualcomm to address passive fabric limitations and optimize large-scale AI system efficiency
Debatable claims
- ▪The AI industry should prioritize optimizing existing hardware efficiency over scaling up cluster sizes
- ▪AI data centers should prioritize open networking standards over proprietary ecosystems
Story comments
Loading comments…