Meta joins Nvidia, OpenAI, and AMD to launch Ethernet for Scale Up Network — a groundbreaking collaboration aimed at transforming how high-performance AI clusters communicate. The initiative, part of the Open Compute Project (OCP), brings together some of the biggest names in tech to make Ethernet the foundation for scalable, efficient AI networking.
For years, InfiniBand has been the go-to interconnect technology for AI workloads, powering nearly 80% of GPU clusters. But with the launch of Ethernet for Scale Up Network (ESUN), that dominance is being challenged. The project’s mission is clear: develop open, high-performance Ethernet standards that simplify how massive AI systems connect, communicate, and grow.
By embracing open standards, ESUN aims to reduce dependency on proprietary systems and create a level playing field for innovation.
The new initiative isn’t just about hardware — it’s about cooperation at a level rarely seen among competitors. Alongside Meta, Nvidia, AMD, and OpenAI, several other major players have joined ESUN, including Cisco, Arista, ARM, Broadcom, HPE Networking, Marvell, Microsoft, and Oracle.
Their shared goal is to unlock the potential of Ethernet to handle the bandwidth and latency demands of AI workloads without the restrictions imposed by closed ecosystems.
Ethernet has always been the backbone of the internet — reliable, cost-effective, and well-understood by engineers worldwide. What makes Ethernet for Scale Up Network (ESUN) special is its focus on scalability and interoperability for AI infrastructure.
Supporters believe that adopting Ethernet as the default for large AI clusters could:
Lower infrastructure costs
Simplify deployment and maintenance
Improve compatibility across hardware vendors
Accelerate innovation through open collaboration
This move could make Ethernet the de facto choice for AI networking, reducing complexity while boosting performance across data centers.
As artificial intelligence continues to scale, the need for high-speed, low-latency interconnects grows. The ESUN project represents a forward-looking approach — one that prioritizes openness, flexibility, and community-driven development over closed, proprietary systems.
If successful, Ethernet for Scale Up Network could mark a turning point in AI infrastructure design, giving organizations greater control over how they scale their compute networks.
Meta’s participation, alongside Nvidia, OpenAI, and AMD, signals a unified commitment to a more open and efficient future for AI networking — one that may redefine the standard for data-driven innovation.
𝗦𝗲𝗺𝗮𝘀𝗼𝗰𝗶𝗮𝗹 𝗶𝘀 𝘄𝗵𝗲𝗿𝗲 𝗿𝗲𝗮𝗹 𝗽𝗲𝗼𝗽𝗹𝗲 𝗰𝗼𝗻𝗻𝗲𝗰𝘁, 𝗴𝗿𝗼𝘄, 𝗮𝗻𝗱 𝗯𝗲𝗹𝗼𝗻𝗴. We’re more than just a social platform — from jobs and blogs to events and daily chats, we bring people and ideas together in one simple, meaningful space.