The Linux Foundation forms the Ultra Ethernet Consortium with AMD, Cisco, HPE, Intel, Meta, Microsoft, and others to optimize Ethernet standards for AI and HPC
UEC to deliver on Ethernet-based open, interoperable, high-performance full-communications stack architecture to meet the growing network demands of AI & HPC at scale. Rakesh Chopra : AI/ML will influence all aspects of our lives, and this is creating a seismic shift in our society. — It is not easy for technology to rise to meet this challenge. … Kevin Wollenweber : Rakesh is always spot on. AI with all of its promise will require all of us to continue to innovate around open, efficient technologies. … Mark Hill : As of yesterday there are now three public examples of the software-hardware pathfinding my group contributes to at Microsoft Azure: … Mark Papermaster : Great industry partnership to align on a standard to connect high performance AI and HPC computers with high performance networks! — #ethernet #ai … Twitter: @broadcom : We're proud to support the Ultra Ethernet Consortium, which is bringing together leading companies for industry-wide cooperation to build a complete #Ethernet-based communication stack architecture for high-performance #networking. Learn more: https://www.broadcom.com/... Markus Schaffhauser / @m_schaffhauser : Exciting news! 🚀 Eviden joins forces with industry giants as a founding member of the Ultra Ethernet Consortium. Together, we will deliver ethernet-based open, interoperable, high-performance full-communications stack architecture.⚡ Find out more 👉 https://www.linuxfoundation.org/ ... [video] Glenn K. Lockwood / @glennklockwood : After reading the Ultra Ethernet whitepaper ( https://ultraethernet.org/...), it's clear that the goal is to put RoCE in the ground and solve network challenges specifically for AI workloads (e.g., broadcast/incast-heavy). Subtext: break the NVIDIA-Mellanox hegemony. @thoefler : @ultraethernet set out to bring the benefits of proprietary #HPC interconnects back to #datacenter #ethernet! Converegence is now officially starting - read the white paper https://ultraethernet.org/.... Some more background at https://arxiv.org/... and https://ieeexplore.ieee.org/ ... :-) [image] @linuxfoundation : We are excited to announce the launch of the Ultra Ethernet Consortium, to deliver ethernet-based open, interoperable, high-performance full-communications stack architecture for the growing network demands of AI & HPC at scale. Learn more: https://www.linuxfoundation.org/ ... @ultraethernet [image]
Context & Ripple Effects
The consortium extends a familiar open-interconnect playbook in this coverage: Intel, Google and Microsoft had earlier backed CXL as an open CPU-to-accelerator interconnect, while the new group focuses on the network fabric linking AI and HPC systems.
Its significance is the breadth of the founding coalition across chipmakers, networking vendors and large infrastructure operators. The effort later progressed to a published UEC 1.0 specification, showing that the initial alliance was positioned to turn interoperability goals into a defined standard.
First-order effects
- AMD, Cisco, HPE, Intel, Meta, Microsoft and the other founders gain a common forum to define an Ethernet-based communications stack for AI and HPC deployments.
- Ethernet suppliers and system builders receive a coordinated target for interoperability and performance work, rather than relying solely on vendor-specific networking approaches.
Second-order effects
- The initiative puts pressure on competing AI-networking approaches to demonstrate openness, compatibility and performance across multi-vendor deployments.
- It aligns with the later Ultra Accelerator Link standards push: as accelerator links are standardized inside servers, the network between servers becomes a separate, consequential interoperability layer.
Third-order effects
- If adoption broadens, AI infrastructure can become more modular: customers could combine compute, servers and network equipment from more vendors, subject to the consortium delivering usable specifications and implementations.
- The deeper shift is from isolated component competition toward standards battles across the full AI system, where control of interfaces can shape which hardware ecosystems remain interoperable.
The trend: AI infrastructure vendors are increasingly organizing open standards for both in-server and between-server interconnects to support heterogeneous AI and HPC systems at scale.