Executive Summary
- During the SC25 conference, Cornelis Networks CEO Lisa Spelman discussed the critical role of high-performance interconnects in the evolution of supercomputing. As AI and HPC workloads scale to unprecedented levels, the demand for low-latency, high-bandwidth fabric technologies is driving innovation in networking architecture. Cornelis Networks is positioning itself at the forefront of this shift, focusing on open, scalable solutions that challenge established proprietary interconnects.
Strategic Deep-Dive
The SC25 conference highlighted a critical bottleneck in the quest for Exascale and beyond: the network. Cornelis Networks, led by CEO Lisa Spelman, is addressing this by developing fabric technologies that prioritize low-latency communication between thousands of compute nodes. Spelman emphasizes that in 2026, the performance of an AI cluster is determined as much by the cables and switches as it is by the GPUs themselves.
Cornelis Networks’ strategy centers on providing high-performance, cost-effective alternatives to InfiniBand and proprietary Ethernet solutions. Their architecture focuses on minimizing the “tail latency” that can cripple large-scale parallel processing tasks. By refining the hardware-software interface, Cornelis aims to ensure that the massive compute power of modern accelerators isn’t wasted waiting for data packets.
This focus on the “future of the fabric” is essential for the sustainability of supercomputing centers that face increasing power and budget constraints, offering a path toward more modular and vendor-agnostic HPC environments.
