Cornelis Networks, the company spun out of Intel’s Omni-Path division in 2020, recently announced a $205 million funding round alongside Active Compute Fabric, a new scale-up interconnect architecture built on UALink and Ethernet for Scale-Up Networks (ESUN) — open standards positioned as an explicit alternative to Nvidia’s proprietary NVLink and InfiniBand.
The pitch is narrow and specific: AI clusters lose money when GPUs sit idle waiting for data to arrive across the fabric.
Active Compute Fabric moves some computation into the network itself — combining data transport with in-network acceleration — to cut the time chips spend waiting rather than processing. The round was led by IAG Capital Partners. Cornelis paired the funding news with a partnership to provide networking for Qualcomm’s new Dragonfly HBC accelerators, giving the company a second silicon partner beyond its existing government and HPC customer base.
The strategic subtext matters for anyone tracking vendor lock-in risk: as UALink and ESUN both gain real commercial backers rather than remaining paper standards, operators building next-generation AI clusters get a genuine open-standard alternative to Nvidia’s end-to-end proprietary stack — the first time that alternative has real venture capital and a named silicon partner behind it rather than just a standards-body press release.
Bottom line: Watch whether Cornelis can convert this funding and the Qualcomm tie-up into actual production deployments before Nvidia’s own Spectrum-X and NVLink Fusion roadmap closes the gap — the interconnect layer is where AI infrastructure’s next real competitive battle is being fought, not at the GPU layer alone.
