Ultra Ethernet Consortium officially democratizes networking for AI!

High-speed data flow concept with colorful light trails over modern server racks in a secure data center environment.

Nokia is delighted to congratulate the Ultra Ethernet Consortium (UEC) on the public release of UEC Specification 1.0. This is a transformative milestone that will reshape the future of AI and high performance computing (HPC) network infrastructure. 

As a General Member of the UEC, Nokia is proud to contribute to this groundbreaking collaborative effort that brings together the industry's leading minds to address the most demanding networking challenges of our time. Learn more about UEC 1.0 here.

“With strong partners like Nokia, we know the UEC will drive the kind of standard to ensure broad AI adoption,” said J Metz, Chair of the Ultra Ethernet Consortium.

Ultra Ethernet: A true InfiniBand alternative

The UEC 1.0 specification fundamentally transforms Ethernet into a high-performance networking solution that can rival InfiniBand in AI and HPC environments. Picking up where RDMA over Converged Ethernet (RoCEv2) leaves off, UEC 1.0 addresses the limitations of congestion management imposed by the unique characteristics of AI workloads running at scale – a limitation that has long challenged hyperscalers and cloud providers. 

While RoCEv2's reliance on lossless fabrics and Priority-based Flow Control created operational complexities, UEC's Ultra Ethernet Transport (UET) protocol represents a fresh approach designed from the ground up to deliver AI and HPC performance while preserving Ethernet/IP ecosystem advantages. 

Ultra Ethernet incorporates advanced congestion management, multipath routing and packet spraying capabilities. The breakthrough comes from Ultra Ethernet's ability to handle massive AI data volumes while maintaining ultra-high throughput with minimal packet loss, approaching InfiniBand's wire-rate performance on industry standard merchant silicon developed by Broadcom while delivering smooth throughput at 800Gbps, 1.6Tbps and beyond.

Crucially, Ultra Ethernet can reduce hardware expenses through greater operational simplicity and by avoiding the vendor lock-in inherent to InfiniBand’s proprietary interconnect. As well, with superior scalability to 1,000,000 endpoints, there is no loss of AI network features. 

Ultra Ethernet momentum

This UEC launch follows Broadcom’s announcement of its Tomahawk 6 switch series, which delivers 102.4Tbps switching capacity. As a UEC steering member and our strategic partner, we congratulate Broadcom on this achievement. 

“With an impressive switching capacity of 102.4Tb/s and extensive range of AI networking topologies and features, Broadcom's Tomahawk 6 series provide a valuable tool set in support of Nokia's expanding portfolio of modern data center switches for AI networks.” - Jeff Jakab, Vice President of Hardware, Nokia 

Industry analysts are also taking notice of this Ethernet momentum, with Dell'Oro Group projecting that Ethernet will surpass InfiniBand in AI back-end networks by 2027.

Graphic of Ethernet vs. InfiniBand in AI Back-End Networks


Source: Dell’Oro AI Networks for AI Workloads January 2025 report

Nokia's Commitment and Real-World Success

Our dedication to advancing AI networking infrastructure is already delivering results. Nscale, the hyperscaler engineered for AI, recently selected Nokia to deliver, in partnership with AMD and Lenovo, an IP network solution to support AI workloads at Nscale’s sustainable data center in Stavanger, Norway.

Nscale required a reliable, high-performance network to support training and inferencing on large-scale GPU clusters that rank among the Top 500 supercomputers in the world. 

As Nscale CTO David Power noted: "Nokia's data center fabric enables us to scale our GPU clusters while maintaining the reliability and performance needed to serve our customers with cutting-edge AI services." 

Our work with Nscale exemplifies how Nokia's data center switching solutions are already enabling the next generation of AI infrastructure that UEC 1.0 will further enhance.

The convergence of Nokia's proven data center switching capabilities with UEC's revolutionary transport protocol creates unprecedented opportunities. Nokia looks forward to working closely with our customers and partners as well as the broader UEC ecosystem. Together, we will bring these transformative capabilities to market, helping customers transition from legacy protocol limitations to the enhanced capabilities that UEC 1.0 provides. Learn more about Nokia Networking for AI workloads here.

Congratulations to the entire Ultra Ethernet Consortium community on this remarkable achievement. The future of AI and HPC networking is brighter with UEC 1.0, and Nokia is excited to continue leading this transformation.

Chalon Duncan

About Chalon Duncan

Chalon leads the AI data center networking go-to-market strategy at Nokia. He is an industry veteran with broad experience across the full stack of data center technologies having worked at leading IT infrastructure vendors like Cisco, Brocade, Juniper and with technology alliance ecosystem partners.

Connect with Duncan on LinkedIn and X

Article tags