Every accelerator in a large AI cluster spends part of its life waiting for data it cannot fetch fast enough.
Cornelis Networks has banked $205M led by IAG Capital Partners, pairing the raise with a Qualcomm collaboration and a new architecture called Active Compute Fabric.
The Wayne, Pennsylvania company builds congestion-free networking for AI and HPC clusters, taking on Cisco and Arista. Deployments include the Texas Advanced Computing Center and the US Department of Energy.
Active Compute Fabric, unveiled at the AI Infra Summit, integrates programmable compute directly into the network. The aim is to offload collective operations so expensive accelerators spend less time idle waiting for data, which matters as training clusters grow past hundreds of thousands of chips.
CEO Lisa Spelman said the funding scales production, deepens customer partnerships and accelerates the rollout of the CN6000 platform.
Compute, memory and storage all became workload-aware in recent years while traditional networks lagged, leaving a gap Cornelis is attacking with in-network processing.
Why it matters: at current cluster sizes, networking decides how much of the hardware customers already paid for actually gets used.
