Skip to content

Scale-across Networking: Panel discussion on routing and transport equipment implications#

OCP Global Summit 2026 digest Program Core pick 12–15 October 2026

Abstract Session recording Materials

Jeff Rahn (Meta), Yawei Yin (Microsoft), Xiguang - Henry Wu (Broadcom), Tad Hofmeister (Google)

  • Why: The PM's DC/WAN interconnect territory: with power and space capping single-site clusters, operators are extending AI fabrics across metro/regional distances to reach 0.5M–1M XPUs — a problem that diverges sharply from conventional cloud DCI.
  • What: A multi-operator panel (Meta, Microsoft, Google, Broadcom, with Arista referenced) on the scale-across use case: real-world rollout experience and the architectures being built for clusters beyond 100k XPUs, where propagation delay, higher oversubscription, and massive synchronized bursts change the equipment requirements (from abstract only).
  • How: Panel focuses on concrete technology implications: coherent pluggables (ZR/ZR+) to overcome fiber scarcity, deep packet buffering to absorb regional latency, wire-rate security, and advanced congestion control.
  • Where applicable: Multi-operator (not single-vendor) and forward-looking — panel format means less depth per topic, but it is the best cross-industry read on where AI DCI equipment requirements are heading. Pair with the Alibaba AI DCI/WAN TE talk for the deployment-level counterpart.