Cornelis Networks Raises $205 Million to Challenge Nvidia's Networking Monopoly

The bottleneck for large-scale AI is no longer just how many GPUs you own — it's how much time those GPUs spend idle, waiting for data. Cornelis Networks, an Intel spinoff, just raised $205 million to attack exactly that problem.

The round and the announcement

Cornelis Networks announced a $205 million round on September 14, 2026, led by IAG Capital Partners, alongside the launch of Active Compute Fabric, a new network architecture built specifically for AI clusters. The announcement came at the AI Infra Summit, paired with a strategic collaboration with Qualcomm Technologies.

Cornelis originated as an Intel spinoff in 2020 and has since positioned itself as a provider of congestion-free data center networking for AI and HPC (high-performance computing) workloads.

What sets Active Compute Fabric apart

The core technical pitch is that the network stops being just a data-transport layer and starts doing useful work while data travels between accelerators. Active Compute Fabric combines lossless transport, in-network acceleration, and programmable processing, so operations happen simultaneously with data movement instead of waiting for it to finish.

The result the company promises is a reduction in the amount of time GPUs sit idle waiting for data to arrive — a problem that compounds exponentially as clusters scale to tens of thousands of accelerators.

What it means for companies buying AI compute

For most companies in Guatemala and Latin America, this doesn't translate into a direct purchasing decision — almost no organization in the region operates its own cluster of thousands of GPUs. But it matters indirectly: the emergence of viable alternatives to Nvidia's proprietary networking (NVLink, InfiniBand via Mellanox) puts pressure on prices and speeds up the availability of cheaper inference compute in the cloud — something that does directly affect the cost per token any company pays when it consumes frontier model APIs.