- Published on
OCP Summit 2026: Powering AI with Open Innovations in Scale-Up, Scale-Out, and Scale-Across Networks
- Authors
- Name
- DeepHorus
At the 2026 Open Compute Project Global Summit, Meta unveiled next-generation open networking infrastructure for AI scale-up, scale-out, and scale-across networks. The announcements include IcePack, Meta's first liquid-cooled network rack, and IceCube, a 102.4T switch platform. Meta says the work supports AI workloads such as Muse Wonder, Muse Spark, and Muse.
Meta co-founded OCP in 2011 and argues open networking is critical for scalable, sustainable infrastructure. At OCP 2025, it introduced AI scale-out fabrics and helped create the Ethernet for Scale-Up Networking initiative. This year's updates extend that strategy as AI clusters approach gigawatt scale.
IcePack is based on OCP's ORV3-HPRv1 rack standard and houses IceCube switch blades. IceCube delivers 128x800G non-blocking switching capacity and 102.4 Tbps, a 2x bandwidth increase over Meta's 51.2T platforms. It comes in two variants, using Broadcom's Tomahawk6C or NVIDIA's Spectrum-6 ASIC. Both run Meta's open-source FBOSS network operating system with OCP's Switch Abstraction Interface.
For scale-out, Meta introduced Flattened Planar Fabric, a multiplane backend fabric that changes how GPUs connect to the network. The design aims to enable larger GPU clusters with lower latency. For scale-across, Meta described petabit-scale backend aggregation super-spines and SRv6 regional backbones built with FBOSS and SAI.
In scale-up, Meta's MTIA 300 system uses Tahan, an Ethernet-based 51.2T switch delivering 6.4 Tbps per accelerator. Meta previewed a 72-accelerator-per-rack design on 102.4T silicon and pointed toward an ESUN-based implementation with future chips. Meta also launched the Open Switch Platform Architecture initiative within OCP, co-led with Microsoft and Alibaba.
OSPA aims to standardize the switch platform layer, and specification v1.0 has been submitted to OCP. Separately, MetaRoCE, Meta's RDMA transport protocol for AI networks, now has open-source implementations. Meta invited engineers, silicon vendors, system integrators, and operators to join the OCP community, framing the challenge as building open, efficient AI infrastructure.
