AI Infrastructure Company Cornelis Announces Completion of $205 Million Financing Round, Led by IAG Capital Partners. The company also launched a network product named Active Compute Fabric, aiming to improve data transmission efficiency between AI chips and reduce idle time while waiting for data such as GPU.

Financing and new products announced simultaneously

Cornelis is mainly developed for the network technology that interconnects AI chips. The company states that currently, a significant amount of GPU computing power is wasted waiting for data to arrive, which limits the utilization rate of the clusters. The newly released Active Compute Fabric aims to solve this issue by allowing chips to continue transmitting data while processing information.

Such capabilities are particularly crucial for large-scale AI training and inference clusters. As the scale of GPU clusters expands, network layer latency and throughput are becoming important factors affecting overall efficiency.

Entering the AI infrastructure after the split from Intel

Cornelis was spun off from Intel in 2020. The company's current focus of competition is not on directly manufacturing GPU, but rather on providing alternative solutions for the connection layer between GPU and accelerators.

One of its key features is an open architecture. According to the company, customers can integrate different types of GPU and accelerator hardware into the network architecture of Cornelis, without being bound to a single chip manufacturer's complete system.

Aiming for choices outside of the Nvidia ecosystem

In the AI infrastructure market, Nvidia still holds a dominant position. Although the chips from Nvidia can theoretically run on other network architectures as well, their software and hardware systems are more compatible with their own complete stack, which makes it easier for customers to continue adopting Nvidia's overall solutions.

Cornelis attempts to make a breakthrough from this point, aiming to gain market space with its open and interconnected capabilities. TechCrunch classifies it as one of a group of emerging AI infrastructure companies that are gradually challenging Nvidia's dominant position within the AI clusters.

Products have begun to be shipped.

Cornelis indicates that the existing products have already begun to be shipped, and the company is also working on the next generation of products, which are expected to be released later this year.

From the perspective of industry trends, as the training costs of AI continue to rise, there is an increasing focus on network efficiency, cluster utilization rates, and compatibility with heterogeneous hardware. This also presents more opportunities for infrastructure providers that are not part of the Nvidia system.