Europe has just taken a step forward in the global race for supercomputing and artificial intelligence. AMD and Eviden, the product brand of the Atos group, have been selected to build Alice Recoque, France’s first exascale supercomputer and the second in Europe. It is designed not only to carry out cutting-edge scientific research but also to serve as a true “AI factory” dedicated to the continent.
The system, which will cost around 554 million euros, promises over one exaflop of performance in double precision, significantly higher energy efficiency than similar systems, and an architecture explicitly designed to boost the so-called European sovereign AI: models trained and run on infrastructures under European control, with data and strategic capabilities within its borders.
A scale leap for science and AI in Europe
Alice Recoque will be installed at France’s major supercomputing center, operated by GENCI (Grand Equipment National de Calcul Intensif) and managed by France’s Atomic Energy and Alternative Energies Commission (CEA). It will be the second European exascale system after Germany’s Jupiter supercomputer, and aims to become a global benchmark for frontier scientific computing and AI.
An exaflop equals a trillion floating-point operations per second. It’s a hard number to grasp: more calculations per second than all of humanity could perform by hand over several years. This leap in capability will allow research teams to perform simulations and models that were previously infeasible or would require weeks of machine time.
The project is part of the European Union’s coordinated effort to strengthen its high-performance infrastructure through the EuroHPC initiative and the Digital Europe program, focusing on three main axes: technological sovereignty, industrial competitiveness, and green transition.
A European consortium with a €554 million budget
The development of Alice Recoque is not an isolated project but the flagship of a European consortium called Jules Verne, led by France through GENCI and CEA, with participation from SURF (Netherlands) and GRNET (Greece).
Funding combines resources from the EuroHPC joint undertaking—Europe’s tool for co-financing strategic supercomputers—and contributions from participating countries through the consortium. The total project cost is approximately €554 million, covering acquisition, installation, commissioning, operation, and system evolution during its lifespan.
This initiative carries a clear political message: in a context where the US and China compete for control over AI computing infrastructure, the European Union aims to develop its own capabilities to train advanced models without relying entirely on external commercial data centers.
An “AI-HPC Factory” powered by AMD chips
Alice Recoque is explicitly designed as an “AI-HPC Factory,” a solutions hub combining three pillars: high-performance simulation (HPC), large-scale data analysis, and artificial intelligence.
Technically, the system will rely on:
- NVIDIA CPUs AMD EPYC next-generation processors, codenamed “Venice,” optimized for memory and computationally intensive workloads.
- AMD Instinct MI430X GPUs, a new generation from the MI400 series, specifically designed for sovereign AI and scientific computing.
- AMD FPGAs, for specialized acceleration and orchestration tasks.
- Eviden’s BXI network, a high-performance interconnection linking all compute nodes.
- BulldSequana XH3500 platform, Eviden’s latest supercomputing rack series, along with high-capacity DDN storage.
Overall, the system will comprise 94 racks and surpass the exaflop performance in HPL (the metric used in the Top500 list to measure supercomputers in double precision). It will also feature exceptional memory capacity, crucial for complex simulations and large AI models.
Fewer racks, less energy: efficiency as a strategic weapon
One of AMD and Eviden’s key focuses is energy efficiency. According to project specifications, the architecture will enable:
- Using approximately 25% fewer racks and components than comparable exascale systems.
- Achieving up to 50% better energy efficiency per GPU, thanks to the combination of MI430X accelerators and optimized platform design.
Each AMD Instinct MI430X GPU features 432 GB of HBM4 memory and a bandwidth of 19.6 TB/s, along with support for advanced AI numeric formats like FP4 and FP8. These formats enable training and deploying massive AI models while consuming less energy and memory, without sacrificing accuracy in critical phases.
Efficiency extends beyond chips. Eviden will implement:
- Argos, its intelligent real-time energy monitoring and optimization software, which adjusts power consumption based on workload and operating conditions.
- Its fifth-generation direct liquid cooling, using tempered water to remove heat from 100% of critical components in each rack, reducing the need for traditional climate control systems and lowering environmental impact.
The result is a supercomputer that not only delivers brute-force power but also aligns with Brussels and EU member states’ increasingly important green computing objectives.
From climate crisis to personalized medicine
Apart from the raw numbers, the real value lies in what will be done with this machine. Alice Recoque is conceived to address some of Europe’s major challenges:
- Advanced climate modeling to simulate scenarios of climate change, extreme events, the impact of decarbonization policies, and the evolution of oceans and the atmosphere with unprecedented resolution.
- New materials and energy by accelerating the design of alloys, batteries, catalysts, and critical components for energy transition and advanced industry.
- Digital twins in medicine creating detailed computational models of organs, tissues, or even entire patients, aiming for more personalized and preventive healthcare.
- Next-generation European AI training language, vision, and multimodal models adapted to European languages and regulatory frameworks, with a strong emphasis on compliance and data protection.
The convergence of simulation, data analysis, and AI will enable applications such as combining physical climate models with AI trained on historical and observational data, or deploying connected digital twins of factories and cities integrated with autonomous decision-making systems.
Digital sovereignty: more than just a supercomputer
AMD and Eviden emphasize the importance of sovereignty. Alice Recoque is not just a fast machine but a crucial element of European strategic autonomy in key areas:
- Data and models trained on European soil, subject to EU data protection regulations and upcoming AI frameworks.
- Technology with a strong European footprint, thanks to Eviden’s role, participation of centers like CEA and GENCI, and integration into the EuroHPC strategy.
- Capacity to compete globally in AI, not only by using third-party services but by creating and controlling its own infrastructures.
Backing this effort are decades of experience from AMD, with over 55 years in developing high-performance processors and solutions, and from Eviden and Atos Group, with extensive backgrounds in supercomputing, cybersecurity, and critical systems.
The combination of scientific ambition, coordinated investment, and a clear focus on energy efficiency positions Alice Recoque as one of the most emblematic projects in the new wave of exascale supercomputers in Europe.
FAQs about Alice Recoque and Europe’s new exascale supercomputing
What is the exascale supercomputer Alice Recoque, and why is it strategic for Europe?
Alice Recoque is a next-generation supercomputer surpassing one exaflop in double precision, making it France’s first exascale system and the second in Europe. Its significance lies in enabling complex scientific and industrial simulations and training large AI models within an EU-based infrastructure, aligned with digital sovereignty and green transition goals.
What hardware will Alice Recoque use (CPU, GPU, memory, and network)?
The system will be based on AMD EPYC processors codenamed “Venice,” AMD Instinct MI430X GPUs from the MI400 series, AMD FPGAs, and the Eviden BullSequana XH3500 platform, connected via the high-performance BXI network. Each GPU will have 432 GB of HBM4 memory and 19.6 TB/s bandwidth, supporting FP4 and FP8 formats optimized for AI, distributed across 94 racks with high-capacity DDN storage.
How does Alice Recoque contribute to European “sovereign AI”?
As it is financed and governed via EuroHPC and the Jules Verne consortium, Alice Recoque provides European researchers and industries with a powerful infrastructure under regulatory control, enabling the training and deployment of advanced AI models—from language to climate, energy, and health systems—without complete reliance on non-EU cloud providers.
What role does energy efficiency play in Alice Recoque’s design?
Energy efficiency is fundamental. The architecture will achieve exaflop scale with roughly 25% fewer racks than comparable systems and up to 50% better efficiency per GPU, thanks to high-density HBM4 memory, the fifth-generation direct liquid cooling system using tempered water, and Argos—its real-time energy monitoring software. This aligns with Europe’s sustainability goals for high-performance scientific computing and lower carbon footprint.