The Open Compute Project Foundation (OCP) has unveiled plans for a networking-focused working group set to examine Ethernet for AI scale-up.

Announced at OCP’s Global Summit 2025, the Ethernet for Scale-Up Networking (ESUN) working group will share work on designing large-scale data center infrastructures.

The likes of AMD, Cisco, Hewlett Packard Enterprise Networking (Juniper), and Nvidia are all participating and will focus on open, standards-based Ethernet switching and framing for scale-up networking.

Participating vendors will take part in regular ESUN meetings examining network switches, including protocol headers, error handling, and lossless data transfer.

The working group will also examine XPU endpoints, looking into how designs impact workload partitioning, memory ordering, and load balancing.

OCP previously looked into advancing endpoint functionality for scale-up networking through its SUE-Transport (Scale-up Ethernet Transport) workstream, now known as SUE-T. The non-profit said the newly rebranded group will continue its efforts despite the formation of ESUN.

The ESUN group will initially focus on L2/L3 Ethernet framing and switching, and will actively engage with organizations like the Ultra Ethernet Consortium and standards such as IEEE 802.3 to align on open standards and best practices.

The open standard's AI moment

Scale up focuses on connecting GPUs or accelerators within a single machine or tightly coupled cluster. It is not to be confused with scale-out, which focuses on connecting multiple separate machines across a data center.

ESUN’s focus on AI scale-up comes amid increased demand for AI services. Compared to InfiniBand-based interconnects, the decision to drill down on Ethernet for scale-up aims to use the open standard to help operators reduce costs and improve interoperability.

As recently as 2023, InfiniBand held an 80% share of the AI back-end networks market. But recent Dell’Oro Group research suggests that increased interest in Ethernet-based offerings could drive nearly $80 billion in data center switch sales over the next five years.

Compared to proprietary alternatives, Ethernet is well-known to the vast majority of industry engineers and is vastly cheaper to interconnect hundreds of thousands of GPUs, which has helped drive its resurgence.

“By leveraging Ethernet’s mature hardware and software ecosystem, ESUN will encourage diverse implementations and drive rapid adoption across the industry,” an OCP statement reads.

Several of the vendors participating in ESUN have launched Ethernet-based offerings targeting scale-up for AI.

Broadcom, for example, was a leading voice in the aforementioned SUE efforts, helping to develop the SUE framework specification to provide communication between XPUs in a multi-XPU system.

To further target scale-up for AI, Broadcom launched its Tomahawk Ultra Ethernet-based switch back in July, which can support up to 77 billion packets per second to power high-intensity AI workloads.

Even Nvidia, which has long championed InfiniBand in its networking switches, turned to Ethernet with Spectrum-X, its end-to-end accelerated networking platform for AI clusters.

The full list of firms participating in ESUN includes: AMD; Arista; ARM; Broadcom; Cisco; HPE Networking; Marvell; Meta; Microsoft; Nvidia; and OpenAI.