AI infrastructure company Cornelis Networks has raised $205 million in new funding as it seeks to expand its networking technology for artificial intelligence and high-performance computing workloads. The funding round was led by IAG Capital Partners and comes alongside the launch of the company’s new Active Compute Fabric, a networking architecture designed to improve the efficiency of large-scale AI systems.
Cornelis is developing technology aimed at addressing a growing challenge in AI infrastructure: keeping large numbers of processors and accelerators supplied with data efficiently. In large AI clusters, GPUs can spend significant amounts of time waiting for data to arrive rather than performing computations. Cornelis is positioning its networking technology as a way to reduce this idle time and improve the utilisation of expensive AI computing infrastructure.
The company’s new Active Compute Fabric takes a different approach to conventional networking. Rather than using the network only to transfer data between computing devices, Cornelis is designing its network fabric to perform programmable computing functions as data moves through the system. This allows some workloads to be handled within the network instead of relying entirely on GPUs for those operations.
According to Cornelis, the approach can help improve the utilisation of GPUs in AI systems. The company has said that GPUs in some environments operate at utilisation rates of around 42% to 54%, leaving significant computing capacity unused while systems wait for data. By moving certain processing capabilities into the network, Cornelis aims to increase GPU utilisation and make existing AI infrastructure more productive.
The new architecture also expands Cornelis’ focus from scale-out networking to scale-up networking. Scale-out networking connects multiple systems across larger clusters, while scale-up networking focuses on high-speed connections between processors and accelerators within a system or rack. Cornelis sees an opportunity among customers looking for alternatives to Nvidia’s proprietary networking ecosystem.
For its scale-up offering, Cornelis is using open standards including UALink and ESUN. The company believes an open architecture can give AI infrastructure operators more flexibility in selecting accelerators and networking technologies instead of tying their systems closely to a single hardware and software ecosystem.
Cornelis’ strategy comes as Nvidia maintains a dominant position in AI computing infrastructure. Nvidia’s GPUs are widely used for AI training and inference, while its broader software and networking ecosystem has helped strengthen the company’s position in the market. Cornelis is not attempting to compete with Nvidia simply by developing another GPU; instead, it is targeting the networking layer that connects and coordinates AI computing resources.
The company was established in 2020 after spinning out of Intel’s Omni-Path business. Cornelis has since developed networking products for AI and high-performance computing, including SuperNICs, switches and software. Its first generation of standalone products began shipping in 2025, and the company has continued expanding its networking portfolio.
Alongside the funding and Active Compute Fabric announcement, Cornelis has also announced a collaboration with Qualcomm Technologies focused on AI infrastructure and rack-scale networking. The partnership is intended to support the development of networking solutions for accelerator-based AI systems.
The new capital will provide Cornelis with additional resources to expand its technology, manufacturing and customer deployments as demand for AI infrastructure continues to grow. The company is targeting customers across hyperscalers, AI companies, enterprises, government organisations and academic institutions that are building large-scale AI and high-performance computing environments.
Cornelis’ latest funding and product launch reflect a broader shift in the AI infrastructure market. As companies continue investing billions of dollars in GPUs and AI data centres, improving how efficiently those computing resources communicate and operate has become increasingly important. Cornelis is betting that programmable, open networking can play a larger role in addressing those infrastructure challenges while giving customers more alternatives to Nvidia’s tightly integrated ecosystem.
