Cornelis Networks announced a $205 million funding round on 14 September, led by IAG Capital Partners, alongside a collaboration with Qualcomm on the architecture of rack scale AI infrastructure. The Intel spinoff will use the money to scale production of its CN5000 and CN6000 network switches and to push out Active Compute Fabric, an open architecture it unveiled the same day for both scale up and scale out AI networking.
The idea behind Active Compute Fabric is that GPUs spend meaningful time idle waiting on data rather than computing, so pushing some of that communication work, collective operations and adaptive routing among them, into the network itself should keep expensive accelerators busier. The fabric combines lossless transport with in fabric acceleration and programmable functions the company says can adapt as AI software workloads change, rather than requiring new hardware each time.
The framing is explicitly anti-Nvidia. Nvidia’s Quantum and Spectrum networking gear, inherited through its Mellanox acquisition, already dominates the interconnects tying GPU clusters together, giving the company a second lever of lock-in beyond the chips themselves. Cornelis and its backers are betting that hyperscalers and neoclouds building out AI capacity want at least one qualified alternative before they sign another multi-year networking contract with their GPU supplier.
Cornelis has not disclosed a customer list, a shipping date for hardware built around Active Compute Fabric, or independent benchmarks showing how much idle GPU time the architecture actually recovers in a live cluster. Qualcomm’s involvement signals it wants a seat in AI data centre infrastructure beyond mobile chips, but neither company has said whether that collaboration extends to Qualcomm’s own AI accelerators or stops at the network layer.
