AMD and Eviden have been selected to build Alice Recoque, a next-generation exascale supercomputer that will become France’s first system of its kind and only the second exascale machine in Europe. The system will support the region’s growing needs in high performance computing and artificial intelligence, serving as a large-scale AI factory for research, industry and public institutions.
The project is led by Grand équipement national de calcul intensif and operated by the Commissariat à l’énergie atomique et aux énergies alternatives. It is funded through EuroHPC Joint Undertaking under the Digital Europe Programme, alongside the Jules Verne Consortium, which includes contributions from France, the Netherlands and Greece. The overall project cost is 554 million euros.
AMD will power the system with its next-generation EPYC CPUs and Instinct MI430X GPUs, designed to deliver more than one exaflop of HPL performance. The machine will also incorporate AMD FPGAs and Eviden’s BullSequana XH3500 platform, connected through Eviden’s BXI network with DDN storage.
Dan McNamara, senior vice president and general manager for Compute and Enterprise AI at AMD, said the collaboration reflects a shared commitment to European leadership in advanced computing. “The Alice Recoque supercomputer represents a major step forward for European sovereign AI, uniting national ambition, regional collaboration and AMD’s high-performance and AI compute technologies,” he said. He added that the system is designed for scale, efficiency and discovery across scientific and industrial domains.
Emmanuel Le Roux, group senior vice president and global head of advanced computing and AI at Eviden, said the project embodies Europe’s goals in sovereignty, sustainability and scientific advancement. “Alice Recoque represents another critical step toward Europe’s digital future. As a catalyst for scientific and industrial breakthroughs, from climate modelling and healthcare to advanced materials and AI innovation, it will empower researchers and industries across Europe,” he said.
Supporting research, industry and next-generation AI workloads
According to AMD and Eviden, Alice Recoque will address Europe’s most complex societal, scientific and industrial challenges. The system integrates hardware, advanced AI software and validated AI use cases to support users through an end-to-end compute lifecycle.
Its capabilities are intended to accelerate progress in areas that rely on large-scale simulation and data analysis, including climate science, materials research, energy innovation and digital twins for personalised healthcare. It will also support the development of new European AI models and workloads that require significant training capacity.
The system will use EPYC processors codenamed Venice and the new Instinct MI430X GPU, designed for sovereign AI and scientific computing. The GPU features 432 GB of HBM4 memory and 19.6 TB/s of bandwidth, with support for AI data types such as FP4 and FP8 to deliver high FLOP performance.
Delivering efficiency and sustainability at exascale
Alice Recoque will consist of 94 racks and is expected to rank among Europe’s top systems for double-precision workloads. Eviden said the design offers improved memory performance to support deeper insights, faster simulations and more efficient research output.
The system aims to reach exascale performance with 25 percent fewer racks and components than other comparable systems, while delivering up to 50 percent better energy efficiency per GPU. This approach supports Europe’s sustainability targets for green computing.
Eviden’s Argos software will provide real-time monitoring and energy optimisation, while its Direct Liquid Cooling technology will use warm water to cool all rack components, improving efficiency at scale. The technology suite is designed to reduce power consumption without compromising performance.
The combination of AMD CPUs, GPUs and FPGAs with Eviden’s integrated architecture is intended to deliver high compute density, improved workload energy efficiency and strong capacity for both AI and HPC workloads.
