Cornelis secures $205 million for open AI networking technology

Cornelis raises $205 million for AI network infrastructure

Cornelis, an AI infrastructure company developing networking technology for AI chips, has raised $205 million in a funding round led by IAG Capital Partners. The company also unveiled Active Compute Fabric, a network technology intended to address GPU time lost while processors wait for data to arrive.

The company is positioning its fabric as a way for chips to process information and send it across the network at the same time. Cornelis has already started shipping its product and is developing a new generation, which it expects to release later this year.

Data movement is central to AI cluster efficiency

Large AI workloads require chips to exchange data continually. Cornelis says a meaningful portion of GPU time can be spent idle while waiting for that data, making networking behaviour an important part of how effectively accelerator capacity is used.

Active Compute Fabric targets that bottleneck by combining communication with processing rather than treating data transfer as a separate waiting period. The announcement centres on network infrastructure rather than a new GPU, reflecting the role that interconnect technology plays in AI systems built from many accelerators.

An open-architecture alternative to a full GPU stack

Cornelis spun out of Intel in 2020 and is competing in a market led by Nvidia. Its approach uses an open architecture, allowing customers to use different GPU and accelerator hardware with Cornelis networking fabric.

Nvidia chips can technically operate on other networking fabrics, but they are optimised for Nvidia's own software. That integration can make Nvidia's complete GPU stack easier and more attractive for customers to deploy.

Cornelis is among AI infrastructure companies seeking to challenge Nvidia's market position through individual layers of the stack. Its funding provides resources as it ships the current product and prepares the next generation.

Business implication

For organisations planning or expanding AI clusters, the announcement is a reminder to evaluate networking alongside accelerator selection: the path data takes between chips can affect how much available GPU time is spent doing useful work.