Cornelis, an AI infrastructure company that spun off from Intel in 2020, announced on September 14, 2026 that it has raised $205 million in a funding round led by IAG Capital Partners. The company also unveiled Active Compute Fabric, a new networking technology aimed squarely at Nvidia's dominance in AI data centers.
The Problem Cornelis Solves
In today's AI data centers, a significant portion of expensive GPU time is wasted. Chips sit idle while waiting for data to arrive from storage or other processors. This bottleneck means that even organizations spending millions on Nvidia GPU clusters are getting far less performance than their hardware is theoretically capable of delivering.
Cornelis's Active Compute Fabric addresses this by creating a networking layer that allows chips to process and transmit data simultaneously, rather than sequentially. The company describes it as eliminating the traditional trade-off between computation and communication in distributed AI workloads.
Why Open Architecture Matters
The critical differentiator is that Cornelis uses an open architecture, meaning its networking fabric works with GPU and accelerator hardware from any vendor. While Nvidia chips can technically run on other networking fabrics, they are heavily optimized for Nvidia's own networking software stack (InfiniBand, NVLink), making it much more practical — and more tempting — for customers to use the full Nvidia ecosystem.
Cornelis is betting that customers want an alternative. By offering networking infrastructure that is vendor-agnostic, the company is part of a new wave of AI infrastructure startups aiming to decompose Nvidia's vertically integrated stack piece by piece.
Already Shipping
Notably, Cornelis is not just announcing a concept. The company has already started shipping its Active Compute Fabric product and is working on a next-generation version expected to be released later in 2026. This puts it ahead of many AI infrastructure startups that remain in stealth or prototype stages.
The Competitive Landscape
The $205 million raise positions Cornelis among the best-funded challengers to Nvidia's ecosystem. The company joins a growing list of startups — including optical networking firms, custom chip designers, and software-defined infrastructure providers — that are betting Nvidia's roughly 90% market share in AI compute is vulnerable to disaggregation.
Nvidia itself has responded to such challenges by continuously expanding its own networking capabilities, including its $7 billion acquisition of Mellanox Technologies in 2020 and ongoing development of its Ethernet-based Spectrum-X platform.
Whether Cornelis can gain meaningful traction against a incumbent with Nvidia's resources and ecosystem lock-in remains to be seen. But with $205 million in fresh capital and a shipping product, it is one of the more credible challengers in the AI infrastructure space.
