Cornelis Raises $205M for AI Networking
Cornelis raises $205M to challenge Nvidia in the networking layer connecting large AI and HPC clusters.
Cornelis has raised $205 million as it targets one of the less visible but increasingly important bottlenecks in AI infrastructure: high-performance networking.
What happened
The Intel spinout builds interconnect technology designed to move data efficiently between accelerators across large AI and high-performance computing clusters. The new financing was led by IAG Capital Partners and coincides with the launch of Cornelis' Active Compute Fabric and a new partnership with Qualcomm.
Its products compete with networking technologies that form part of Nvidia's broader AI infrastructure ecosystem.
Why it matters
Buying more GPUs does not automatically translate into better AI performance. As clusters scale to thousands of accelerators, communication between chips becomes a major constraint on utilisation, latency and cost.
That creates room for companies specialising in network fabrics, switches and interconnects even if they do not manufacture the underlying accelerators themselves.
The bigger picture
The AI infrastructure market is beginning to unbundle. Capital is moving not only into chips and data centres but also into networking, memory, cooling, power and orchestration. Cornelis is part of a wider effort to weaken Nvidia's control over the full AI stack by offering alternatives at specific infrastructure layers.