{"slug": "bull-kalray-unite-to-develop-ultra-ethernet-ai-interconnects", "title": "Bull, Kalray unite to develop Ultra Ethernet AI interconnects", "summary": "Bull, the supercomputing giant, has partnered with fabless data processing unit developer Kalray to develop Ultra Ethernet-based interconnect technologies for AI infrastructure, aiming to support very large-scale AI and HPC deployments. The collaboration will integrate Kalray's Networking Engine subsystem into a joint solution expected in 2028, targeting eight million endpoints for Bull's next-gen BXIv3 interconnect. Bull CEO Emmanuel Le Roux called the development a strategic milestone and a European sovereignty play against non-European dominance.", "body_md": "Bull, the supercomputing giant, has teamed up with fabless data processing unit developer Kalray to create Ultra Ethernet-based interconnect technologies for AI infrastructure.\n\nThe former Atos-owned brand will work with Kalray on open networking efforts designed to support “very large-scale” infrastructures for both AI and high-performance computing.\n\nA statement announcing the partnership said the efforts will “accelerate the evolution” of Bull’s BXI networking tech. [Version three](https://www.bull.com/hubfs/Documents/Factsheets/Bull-BXI-V3-Datasheet.pdf?hsLang=en) is already in development and marks the first time the French firm will use Ethernet, replacing the prior generation’s proprietary link protocol. It also promises improved graphics processing unit (GPU) utilization and advanced congestion management.\n\nA [paper](https://arxiv.org/html/2601.19413v1) published on NET4EXA, the European-centric interconnect project, suggests Bull’s next-gen interconnect offering is expected to scale to eight million endpoints, compared to just 64,000 (64K) nodes for BXIv2. Bringing in Kalray aims to help that reach that goal, providing the high-throughput, low-latency required for sizable AI deployments.\n\nA fellow French firm, Kalray specializes in hardware capable of handling massive data flows. Per the partnership, it will provide Bull a license to its software and hardware. Its Networking Engine subsystem will be integrated into a joint commercially available solution set to be available in 2028.\n\n“The development of new interconnect solutions represents a strategic milestone in meeting the needs of future generations of supercomputers and AI infrastructures,” Emmanuel Le Roux, Bull’s CEO, said. “The collaboration we’ve signed with Kalray enables us to develop advanced technologies – particularly for the cloud and AI – and to build high-performance solutions.”\n\nThe joint development project comes as Bull was one of the founding members of the Ultra Ethernet Consortium (UEC).\n\nThe group is rallying around the open protocol to reduce vendor lock-in, adding improvements like a modern remote direct memory access (RDMA) approach to provide low-latency transport for high-throughput environments. Bull joined the likes of AMD, Broadcom, and Cisco in backing the project, which has since [published its first standard](https://www.sdxcentral.com/news/ultra-ethernet-consortium-launches-10-specification-sdx/) and added new members including [NTT](https://www.sdxcentral.com/news/ntt-joins-ultra-ethernet-consortium-to-advance-ai-and-hpc-networking/).\n\nThe French pair want Ultra Ethernet to power advanced networking capabilities for AI, but are also treating their efforts as a sovereignty play.\n\n“In a sector largely dominated by non-European technologies, we are particularly proud to offer a European alternative by producing a chip that incorporates the most advanced technological innovations available today,” Le Roux added.\n\nBull was [acquired by the French state](https://www.datacenterdynamics.com/en/news/french-state-completes-acquisition-of-atos-ai-hpc-and-quantum-group-bull/) in a deal worth around $455 million (€400 million) from Atos in a show of technology strength for the continent. It operates the only supercomputer manufacturing plant in Europe, while its BullSequana high-performance computing (HPC) systems form the backbone of some of the world’s most powerful supercomputers, including the continent’s only exascale system, [Jupiter Booster](https://www.sdxcentral.com/news/ericsson-taps-europes-fastest-supercomputer-to-fuel-ai-development-for-6g/).", "url": "https://wpnews.pro/news/bull-kalray-unite-to-develop-ultra-ethernet-ai-interconnects", "canonical_source": "https://www.sdxcentral.com/news/bull-kalray-unite-to-develop-ultra-ethernet-ai-interconnects/", "published_at": "2026-07-29 10:47:52+00:00", "updated_at": "2026-07-29 11:00:05.478685+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-infrastructure", "ai-chips"], "entities": ["Bull", "Kalray", "Ultra Ethernet Consortium", "Emmanuel Le Roux", "Atos", "BXI", "AMD", "Broadcom"], "alternates": {"html": "https://wpnews.pro/news/bull-kalray-unite-to-develop-ultra-ethernet-ai-interconnects", "markdown": "https://wpnews.pro/news/bull-kalray-unite-to-develop-ultra-ethernet-ai-interconnects.md", "text": "https://wpnews.pro/news/bull-kalray-unite-to-develop-ultra-ethernet-ai-interconnects.txt", "jsonld": "https://wpnews.pro/news/bull-kalray-unite-to-develop-ultra-ethernet-ai-interconnects.jsonld"}}