Breaking

Cornelis Raises $205 Million and Unveils Active Compute Fabric for AI Data Centers

September 14, 2026 at 16:11 EDT

Infra & Chips
Foundation Models

Cornelis announced on September 14, 2026, that it has raised $205 million and introduced Active Compute Fabric, an open networking architecture designed to make AI and HPC clusters spend less time waiting on data. The company is positioning the system as a scale-up and scale-out fabric that adds programmable compute, lossless transport and in-fabric acceleration to the network layer rather than treating the network as a passive data pipe.

September 14, 2026 · Cornelis

A $205M Push to Make AI Networks Compute

Cornelis introduced Active Compute Fabric, a scale-up and scale-out architecture that moves selected AI communication work into the network itself.

$205M

New funding to support scale-up networking and next-generation products.

Q4 2026

Expected expanded availability for CN6000 after customer sampling.

$55B+

Cornelis' cited 2030 opportunity for open-standard AI networking.

CN5000 collective communication claim

A proportional view of Cornelis' stated AI collective communication comparison versus RoCE.

1xRoCE baseline

6xCN5000 claim

Cornelis says the gain comes from a lossless Omni-Path architecture and simulations of all-to-all AI traffic.

Fabric workflow

Data moves → fabric accelerates → collectives offload → accelerators wait less

Validation question

The strongest Active Compute Fabric figures are projections or pre-production simulations, with results expected to vary by workload and deployment.

Why it matters

In a modeled 100,000-GPU system, Cornelis estimates about half of GPU hours may be spent waiting for data, equal to $1.68 billion a year in wasted capacity and 500 GWh of power.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.