OCP Summit 2025: The Open Future of Networking Hardware for AI

This post details Meta's advancements in networking hardware for AI training clusters, presented at OCP Summit 2025. Key contributions include the evolution of Disaggregated Scheduled Fabric (DSF) to a dual-stage architecture supporting up to 18,432 XPUs for scale-out interconnect, and the development of a new Non-Scheduled Fabric (NSF) architecture based on shallow-buffer Ethernet switches for large AI clusters. The post also announces the Minipack3N, a new 51 Tbps OCP switch based on NVIDIA's Spectrum-4 ASIC, and the introduction of 2x400G FR4 LITE and 400G/2x400G DR4 optics for 400G/800G optical interconnections. Meta is also a founding participant in the Ethernet for Scale-Up Networking (ESUN) initiative within OCP, contributing to the advancement of Ethernet for AI accelerator architectures.