Cornelis lands $205M to push open networking alternative to Nvidia's AI stack
Cornelis has raised $205 million in a funding round led by IAG Capital Partners. The company's new Active Compute Fabric networking technology aims to reduce idle GPU time by enabling simultaneous processing and data transfer. With an open architecture that supports multiple chip vendors, Cornelis is positioning itself as an alternative to Nvidia's tightly integrated AI infrastructure.
The funding round, led by IAG Capital Partners, arrives as Cornelis ships its first-generation Active Compute Fabric product, with a next-generation version slated for release later this year. The company's core pitch centers on eliminating wasted GPU cycles: rather than forcing chips to pause computation while awaiting data, the fabric enables simultaneous processing and transfer, a design choice aimed squarely at improving utilization in large-scale AI workloads.
Cornelis traces its origins to an Intel spin-off in 2020, and its open-architecture approach deliberately contrasts with Nvidia's vertically integrated hardware-software stack. While Nvidia's chips can technically operate on third-party fabrics, the tight optimization of Nvidia's own software creates a strong pull toward its full ecosystem. Cornelis positions itself among a growing cohort of startups chipping away at that lock-in by offering customers flexibility across GPU and accelerator vendors.
This funding signals growing appetite for alternatives to Nvidia's dominant AI infrastructure model. If open networking fabrics gain traction, enterprises and cloud providers could gain more bargaining power and hardware choice, potentially lowering AI deployment costs over time. However, the practical impact may remain limited initially, since Nvidia's software ecosystem and performance optimizations are deeply entrenched. The broader effect could be a more competitive AI hardware market, benefiting researchers and startups that currently face high switching costs when building large-scale systems.