Cornelis Raised $205 Million for AI Networking

The company introduced a new architecture designed to improve performance in massive artificial intelligence clusters.

Updated on Sept. 25, 2026 in Data Centers

Cornelis Raised $205 Million for AI Networking

Live Poll

Do you believe new AI hardware infrastructure will effectively reduce data processing costs?

Cornelis has secured $205 million in funding to advance its AI networking technology. The company also unveiled its Active Compute Fabric, an architecture that combines programmable compute with scale-up and scale-out networking standards.

Why it matters

As AI clusters grow in size, synchronization and communication demands often consume vital accelerator capacity. This hardware design aims to reduce accelerator downtime by performing compute functions directly within the network fabric.

The Active Compute Fabric architecture integrates UALink and ESUN standards for scale-up networking alongside Ultra Ethernet for scale-out operations. The system processes data in-transit through lossless transport and in-fabric acceleration.

The players

Cornelis

Based in Wayne, Pennsylvania, this company develops advanced networking hardware and software architectures for high-performance computing.

IAG Capital Partners

This investment firm provides analysis and projections regarding the economic growth of the artificial intelligence networking sector.

The details

The technology is designed to offload collective operations from accelerators, addressing inefficiencies in massive AI systems. Cornelis is currently shipping its CN5000 networking product, while its more advanced CN6000 model is currently in the sampling phase with customers.

The CN6000 product is expected to have expanded availability in the fourth quarter of 2026.
The market opportunity for open-standard AI networking is projected to exceed $55 billion by 2030.

The Tech Race

This development marks a significant move toward replacing traditional, rigid networking hardware with programmable, compute-capable fabrics. It positions the company to compete against legacy infrastructure by targeting the immense inefficiencies found in large-scale AI clusters.

Improved networking efficiency could eventually lead to lower operational costs for large-scale AI service providers and developers. This may translate into more accessible or affordable cloud computing resources for those building and deploying AI models.

The takeaway

The move to integrate compute power directly into the network suggests a future where data movement and data processing are no longer separate tasks. This evolution is essential for maintaining the viability of clusters that scale to 100,000 GPUs and beyond.