Field noteOct 27, 20261 min read1 source

Meta Networking at the 2026 OCP Global Summit: Open Innovation for Scale-Up, Scale-Out, and Scale-Across AI Networks

In one minute

Meta published details of networking systems it presented at the 2026 Open Compute Project Global Summit, describing infrastructure used for its AI models and its gigawatt-scale Prometheus cluster.

What changed: Same event as OCP Summit 2026: Powering AI with Open Innovations in Scale-Up, Scale-Out, and Scale-Across Networks

The company described IcePack, its first liquid-cooled network rack, and IceCube switch blades with 102.4 Tb/s capacity: 128 800G ports using either Broadcom Tomahawk 6C or NVIDIA Spectrum-6 ASICs, running FBOSS. For scale-out fabrics, Meta said each GPU connects to eight network planes; its design moves from a three-tier to two-tier Clos topology and uses NIC-level packet spraying.

Meta also described BAG, a petabit-scale super-spine for connecting buildings, campuses and regions, plus SRv6 regional backbones built with FBOSS and SAI. It previewed an Ethernet scale-up configuration of 72 accelerators per rack on 102.4T silicon. Meta said MetaRoCE, its RDMA transport for AI networks, is going open source.

Separately, Meta, Microsoft and Alibaba co-led OSPA, an OCP initiative for a standardized switch-platform layer; Meta says version 1.0 has been submitted to OCP. The announcement does not state release dates, licensing terms, availability of MetaRoCE code, or independent performance results for these systems.

Evidence / sources

  1. 1engineering.fb.com/2026/10/26/open-source/ocp-summit-2026-meta-networking-talks
All Field Notes

Next reads

Read next

Open the graph
  1. 01topicComputeExplore the compute topic hub.
  2. 02archiveField NotesOpen the latest editorial analysis.
  3. 03source ledgerSource LedgerInspect the public source evidence graph.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract