Beyond the Cloud: Ethernet’s Role in AI and HPC Infrastructure

The rapid advancement of Large Language Models (LLMs) has driven an exponential demand for AI workloads, requiring tens of thousands of GPUs to operate as a unified computing system. At the core of this infrastructure lies the network—where traditional cloud architectures fall short in meeting AI’s unique demands. This presentation explores the critical networking requirements for AI workloads and how Ethernet based networking fabric delivers the efficiency and scalability needed for next-generation AI clusters.
We examine why hyperscalers are increasingly adopting Ethernet for the backend fabric of AI infrastructure, highlighting its flexibility, cost efficiency, and performance advantages in distributed training environments. Additionally, we discuss emerging innovations from the Ultra Ethernet Consortium (UEC) and their implications for high-performance AI and HPC workloads.
Attendees will gain insights into the future of AI and HPC networking, learning how cutting-edge technologies are addressing the challenges of scalability, tail-latency, and congestion management in large supercomputing environments.