Meta has joined Nvidia, OpenAI, and AMD to introduce ESUN, an open Ethernet-based network for scaling up AI superclusters. The initiative aims to compete with InfiniBand while remaining compatible with the Ultra Ethernet Consortium. Engineers believe this approach will simplify complex GPU interconnect systems.
In a significant move for AI infrastructure, Meta announced its participation in a new alliance with Nvidia, OpenAI, and AMD to launch ESUN, an Ethernet solution designed for scale-up networks in next-generation AI superclusters. This development, reported on October 19, 2025, positions ESUN as a direct competitor to InfiniBand, the dominant high-speed networking technology often used in GPU clusters.
The title of the TechRadar article highlights the 'massive tech alliance' taking on InfiniBand with an 'open Ethernet dream.' ESUN is intended to provide an alternative that simplifies the intricate interconnect systems required for large-scale AI computing. Unlike proprietary options, this Ethernet-based approach emphasizes openness and interoperability.
Key to the initiative is its compatibility with the Ultra Ethernet Consortium, ensuring ESUN can integrate with broader industry standards. Engineers involved hope that by leveraging Ethernet's widespread adoption, the network will reduce complexity in GPU setups, potentially lowering costs and improving scalability for AI applications.
This alliance underscores the growing demand for efficient networking in AI development, where rapid data transfer between processors is crucial. No further details on implementation timelines or specific technical specs were provided in the available sources.