European technology companies Bull and Kalray have formed a strategic partnership to develop next-generation networking technologies for artificial intelligence (AI) and high-performance computing (HPC), positioning themselves around the emerging Ultra Ethernet standard. The collaboration aims to deliver high-bandwidth, low-latency networking solutions for AI factories, supercomputers and hyperscale cloud infrastructure as demand grows for more efficient AI infrastructure.
As enterprises race to deploy increasingly powerful AI models, networking infrastructure has become one of the biggest bottlenecks in scaling modern computing environments. Addressing that challenge, Bull, a provider of advanced computing platforms, and Kalray, a specialist in data processing and semiconductor technologies, have announced a long-term partnership to jointly develop networking solutions designed for future AI and HPC workloads.
The agreement centres on next-generation interconnect technologies and support for Ultra Ethernet, an emerging open networking standard that seeks to improve performance, interoperability and scalability for AI clusters and cloud-native infrastructure.
Unlike traditional Ethernet deployments, Ultra Ethernet is being developed specifically to meet the demands of distributed AI training and inference, where thousands of GPUs and accelerators must exchange vast amounts of data with minimal latency. The standard is backed by the Ultra Ethernet Consortium (UEC), whose membership has grown to more than 170 organisations across the semiconductor, networking and hyperscale cloud industries.
AI Infrastructure Is Driving Demand for Faster Networks
The rapid expansion of generative AI and large language models (LLMs) has shifted industry attention beyond processors to the networks connecting them. While GPU vendors such as NVIDIA, AMD and Intel continue advancing AI accelerators, networking has emerged as a decisive factor in determining how efficiently large-scale AI systems operate.
Hyperscale cloud providers and enterprise AI operators increasingly require high-throughput, low-latency interconnects capable of supporting distributed model training while reducing congestion and improving infrastructure utilisation.
Industry analysts at IDC forecast continued double-digit growth in AI infrastructure spending over the next several years, with networking equipment becoming one of the fastest-growing segments as organisations build larger AI clusters. Gartner has likewise identified AI infrastructure as a strategic enterprise investment area, extending beyond compute hardware to include storage, networking and orchestration technologies.
Against this backdrop, Bull and Kalray are positioning their collaboration as a European contribution to the rapidly evolving AI infrastructure ecosystem.
Combining HPC Expertise with Semiconductor Innovation
The partnership combines Bull’s experience in supercomputing platforms with Kalray’s networking and semiconductor intellectual property.
Under the agreement, Kalray will license hardware and software technologies developed over more than a decade, including its Kalray Networking Engine subsystem. The company says the platform will support networking speeds of up to 1.6 Tbps while incorporating 64 next-generation KV5 processor cores based on the open RISC-V architecture.
The collaboration also includes joint engineering efforts aimed at integrating these technologies into future generations of Bull’s networking portfolio.
Rather than developing entirely proprietary networking stacks, the companies are aligning product development with open industry standards, reflecting a broader trend across AI infrastructure providers seeking greater interoperability.
Commercial deployment of the first jointly developed products is expected in 2028, highlighting the long-term nature of the initiative as AI infrastructure requirements continue to evolve.
Ultra Ethernet Emerges as an Alternative Networking Strategy
One of the partnership’s most significant elements is its focus on Ultra Ethernet compatibility.
Traditional Ethernet has long served enterprise networking but was not originally designed for AI workloads requiring extremely high bandwidth, predictable latency and advanced congestion management.
The Ultra Ethernet Consortium aims to modernise Ethernet by introducing features specifically optimised for AI clusters, large-scale cloud computing and next-generation data centres.
Bull, a founding member of the consortium, plans to evolve its BXI networking technology toward Ethernet-based architectures while adding cloud-native capabilities such as virtualisation support, enhanced security and improved scalability.
The companies also intend to develop hybrid networking technologies capable of supporting both traditional HPC workloads and modern AI environments. Such architectures could enable scientific computing applications to continue benefiting from ultra-low-latency interconnects while allowing AI deployments to leverage open Ethernet-based networking.
This dual approach reflects the convergence of AI and traditional supercomputing, where many research organisations increasingly operate mixed workloads across shared infrastructure.
Strengthening Europe’s Position in AI Infrastructure
Beyond technology development, the partnership reflects Europe’s broader ambition to expand its role in semiconductor design and AI infrastructure.
The global AI networking market remains heavily influenced by US-based technology companies, particularly NVIDIA following its acquisition of Mellanox, alongside networking leaders such as Cisco Systems, Broadcom, Arista Networks and Marvell Technology.
European policymakers have increasingly encouraged regional investment in advanced semiconductors, networking technologies and sovereign computing capabilities to reduce dependence on overseas suppliers.
By combining semiconductor IP licensing with collaborative product development, Bull and Kalray are seeking to establish a stronger European presence in an AI infrastructure market expected to expand significantly throughout the decade.
While commercial products remain several years away, the partnership signals continued momentum behind open networking standards and highlights how networking innovation is becoming as strategically important as advances in AI processors themselves.
Market Landscape
AI infrastructure is rapidly evolving beyond compute hardware alone. As organisations deploy increasingly larger AI models, networking has become essential for maximising GPU utilisation, reducing latency and enabling efficient distributed training. Companies including NVIDIA, AMD, Intel, Broadcom, Cisco, Arista Networks, Marvell Technology, Microsoft, Google and Amazon Web Services continue investing heavily in AI networking and cloud infrastructure. At the same time, industry-backed initiatives such as the Ultra Ethernet Consortium are promoting open standards that reduce vendor lock-in while improving interoperability across enterprise data centres and hyperscale AI environments. Bull and Kalray’s collaboration reflects this wider shift towards open, high-performance networking architectures capable of supporting future AI factories and exascale computing.
Top Insights
- Bull and Kalray have partnered to develop next-generation networking technologies for AI and high-performance computing, focusing on scalable interconnects built around the emerging Ultra Ethernet standard.
- The collaboration combines Bull’s supercomputing expertise with Kalray’s semiconductor and networking technologies, targeting bandwidth of up to 1.6 Tbps for future AI infrastructure.
- Ultra Ethernet is gaining industry momentum as an open networking standard designed to optimise distributed AI training, hyperscale cloud deployments and large-scale data centres.
- Commercial products resulting from the partnership are expected in 2028, reflecting long-term investment in AI infrastructure and next-generation networking innovation.
- The agreement also supports Europe’s broader objective of strengthening sovereign AI and semiconductor capabilities in a market currently dominated by global technology vendors.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
