The OCP ESUN 1.0 Specification has been released!

Date of Post
March 10, 2026
Posted by
OCP Networking Project

The OCP Ethernet for Scale-Up Networking (ESUN) initiative has launched its first major specification for AI infrastructure: the “OCP ESUN – Network Operator Requirements – Base Specification 1.0.” This release comes just four months after we kicked off ESUN at the OCP Global Summit 2025.

OCP ESUN 1.0 highlights

ESUN is focused on enabling Ethernet to be the basis of high-performance scale-up networks inside the latest AI systems. Larger and more performant scale-up networks can greatly accelerate the latest AI jobs, such as distributed inference workloads and agentic workflows.

In the OCP ESUN 1.0 Specification, we have enabled:

  1. Highly reliable and lossless Ethernet:  Scale-up networks need lossless connectivity to ensure low latency packet delivery. ESUN1.0 specifies use of congestion management (Priority Flow Control and Credit Based Flow Control) and link level reliability (LLR: Link Level Retry) protocols.
  2. Efficient support for small messages: ESUN 1.0 defines the ESUN Header (EH), which replaces the standard IP/UDP header stack (typically 28–48 bytes) with a compact 4‑byte ESUN header, significantly reducing per-packet overhead and improving payload efficiency—especially for small messages common in scale-up AI communication.
  3. Efficient multi-hop scale-up topologies: A multi-hop topology is important as scale-up cluster size increases. Such topologies demand support for transient as well as persistent congestion control to avoid packet loss and high latencies. The OCP ESUN header provides this functionality through the newly defined ESUN Header.

With these changes, OCP ESUN enables an efficient, performant, open and widely supported scale-up fabric for AI clusters.

Here is the  OCP ESUN 1.0 Specification document published by OCP.

Wide Industry Participation in OCP ESUN

Today, the ESUN initiative has grown to include over 175 participating companies, up from the 12 founding members at its launch on October 13, 2025—AMD, Arista, Arm, Broadcom, Cisco, HPE Networking, Marvell, Meta, Microsoft, NVIDIA, OpenAI, and Oracle.

Meta and Microsoft jointly worked on the “OCP Network Operator ESUN Specification 1.0”. The collaboration includes structured participation across different levels:

  • Meta and Microsoft led the core specification effort with input from the participating ecosystem.
  • Contributing companies provided technical feedback through signed agreements and regular working group participation. These include a variety of companies from silicon vendors to infrastructure operators who contributed to implementation requirements listed in the specification document. The broader community of 175+ companies has provided industry support and representation with active participation and discussions as well.
  • The initiative maintains coordination with relevant standards organizations, including the Ultra Ethernet Consortium (UEC) and IEEE, to ensure compatibility with existing industry frameworks.

The working group meets bi-weekly and provides a forum for technical discussions and coordination across participating organizations. A dedicated OCP ESUN wiki and mailing list have been established to support ongoing collaboration.

For information about ESUN and working group participation, visit the OCP Networking ESUN Project page.