10X Backbone: How Meta Is Scaling Backbone Connectivity for AI

This post details Meta's '10X Backbone' initiative to scale its network connectivity for AI workloads. It introduces three key techniques applied to the Express Backbone (EBB): DC Metro Architecture for simplified and standardized fiber deployment, IP Platform Scaling (scaling up with larger chassis and faster interfaces like 800G, and scaling out by adding planes or devices per plane), and IP/Optical Integration using ZR technology to reduce power consumption and improve rack density. It also outlines solutions for AI cluster expansion, including FR, LR, and ZR plugs with DWDM for varying distances.