Cornelis, the networking infrastructure startup spun out of Intel in 2020, just landed a major $205 million funding round led by IAG Capital Partners. The company is pushing a fresh product called Active Compute Fabric, a system designed to reduce latency wasted while GPUs wait for data—bringing communication and computation closer together. Cornelis is already shipping its initial fabric and has a second-gen version slated for release later this year.
Open Architecture vs. Closed Ecosystems
What sets Cornelis apart is its commitment to open architecture. Unlike typical designs that favor a proprietary stack, Cornelis’ fabric works across different types of GPU and accelerator hardware. That could present a direct challenge to NVIDIA, whose ecosystem has traditionally been strongest when customers stick to its own software and networking tools to maximize performance.
While NVIDIA’s chips technically run on third-party networking tools, the alignment between its hardware and software stack has made its own tools far more attractive to enterprise customers. Cornelis is hoping its more flexible fabric can disrupt this dynamic and draw in users who want more freedom and compatibility.
Where Cornelis Fits in the AI Infrastructure Arms Race
Over the past few years, companies pushing alternative AI infrastructure—across networking, chips, accelerators, and more—have ramped up as a counterweight to NVIDIA’s dominance. Cornelis is joining that trend by honing in on one of the overlooked chokepoints in AI workloads: the communication bottleneck between accelerators.
The Active Compute Fabric aims to let compute and data flow concurrently, rather than sequentially, which the company claims will reduce idle GPU time. Cornelis is already selling its first-generation product, and the upcoming iteration is expected to further boost performance and efficiency.
With $205 million in new capital and a vision for open-network AI infrastructure, Cornelis may be gearing up to chip away at NVIDIA’s stronghold on the market. Investors appear confident; customers seem curious. What remains to be seen is whether the broader ecosystem will embrace an alternative that departs from NVIDIA’s existing hardware-software synergy.
