Meta Introduces New Open Networking Infrastructure for AI at OCP Summit 2026

Next-Generation AI Networking Hardware

At the 2026 Open Compute Project Global Summit, Meta presented new open networking infrastructure designed for AI workloads. Key hardware introductions include IcePack, Meta's first liquid-cooled network rack, and IceCube, a 102.4T switch platform. These components are designed to support high power density and improve thermal efficiency for gigawatt-scale AI clusters.

Innovations in Fabric and Protocols

Meta also introduced the Flattened Planar Fabric (FPF), a multiplane backend scale-out fabric that alters how GPUs connect to the network, enabling larger and more efficient GPU clusters with lower latency. Additionally, Meta announced MetaRoCE, an open-source AI transport protocol. These advancements are intended to support Meta's AI models like Muse Wonder, Muse Spark, and Muse.

Scale-Across and Multi-Site Clusters

For multi-site clusters, Meta detailed its petabit-scale backend aggregation (BAG) super-spine, which addresses large-scale training job failures. The company also provided an overview of its SRv6 regional backbones, built using FBOSS and SAI, to handle longer distances in networking.

Ethernet for Scale-Up Networking and OSPA Initiative

Meta's MTIA 300 system utilizes Tahan, an Ethernet-based 51.2T scale-up switch, delivering 6.4 Tbps per accelerator. The company previewed a next-generation 72-accelerator-per-rack design on 102.4T silicon, moving towards an ESUN-based implementation. Furthermore, Meta launched the Open Switch Platform Architecture (OSPA) initiative within OCP, inviting industry participation to build open, efficient, and sustainable AI infrastructure.