France’s Eviden teams with AMD to create a new supercomputer driving Europe’s AI goals

Alice Recoque becomes Europe’s next AI powerhouse

France’s new Alice Recoque supercomputer is set to become Europe’s second exascale machine, designed as both a scientific workhorse and an “AI factory.”

Installed at CEA’s TGCC center, it will help Europe push back against U.S. dominance by providing researchers with massive computing power for AI training, simulations, and data-intensive projects, all within a strongly European-controlled infrastructure.

Eviden and AMD form a strategic supercomputing duo

On the hardware side, Eviden and AMD are teaming up to deliver the whole system stack. Eviden supplies its BullSequana XH3500 architecture, networking, cooling, and integration, while AMD brings next-gen EPYC CPUs and Instinct MI430X GPUs.

It’s a classic split of European system design and American chip muscle, tuned specifically for AI workloads and high-precision scientific computing.

Exascale performance unlocks staggering computing power

Alice Recoque is designed to exceed one exaflop of peak performance. One exaflop equals one billion billion floating point operations per second, roughly equivalent to the combined peak performance of about ten million typical modern desktop PCs, depending on their configuration.

For scientists and AI teams, this leap into exascale territory means complex models, simulations, and training runs that once took days or weeks can now be completed within far tighter timelines.

Project funding ties into Europe’s AI sovereignty push

This supercomputer is not just a tech upgrade; it’s a political statement. The project costs approximately 554 million euros over five years.

It is funded by the EuroHPC Joint Undertaking, as well as the Jules Verne consortium, led by France, with partners from the Netherlands and Greece.

The goal is clear: to maintain critical AI and HPC capacity under European control, rather than relying on foreign cloud giants for these resources.

European networking replaces Nvidia’s Jupiter blueprint

Unlike JUPITER, which is built around NVIDIA InfiniBand networking, Alice Recoque will use Eviden’s BXI v3 interconnect as its primary fabric, reflecting a different vendor and sovereignty trade-off.

By relying on a European-designed fabric across 94 racks and hundreds of kilometers of cabling, France and its partners reduce dependence on a single U.S. vendor and gain more control over performance tuning and upgrades.

Next-generation AMD chips target sovereign AI workloads

Under the hood, the system will use next-gen AMD EPYC “Venice” CPUs and Instinct MI430X GPUs, along with AMD FPGAs for specialized acceleration. Each GPU is designed with a massive HBM4 memory capacity and extreme bandwidth to support AI models efficiently.

Support for lower-precision AI data types, such as FP4 and FP8, enables researchers to train huge models while extracting more performance per watt from every GPU.

Energy efficiency becomes a first-class design goal

What’s striking is how hard the design leans into green computing. Alice Recoque uses about 25 percent fewer racks and components than other exascale systems, while promising up to 50 percent better energy efficiency per GPU.

Direct liquid cooling with warm water handles 100 percent of rack components, and Eviden’s Argos software constantly monitors and optimizes energy use so performance gains don’t come with runaway power bills.

European components strengthen digital sovereignty

A significant portion of the system’s hardware is expected to come from European suppliers, reflecting a broader trend toward regional production.

Alongside AMD’s technology, the design also includes a dedicated section built around a processor developed by SiPearl, adding another layer of locally driven innovation.

That mix of local CPUs, networking, and system manufacturing helps the EU reduce geopolitical risk and keep sensitive AI and scientific workloads on infrastructure it can fully audit and regulate.

Climate and energy research gets a considerable performance boost

One of Alice Recoque’s headline missions is to supercharge climate and energy modeling. With exascale capacity, researchers can run far higher-resolution climate simulations, test extreme weather scenarios, and stress-test future energy grids with unprecedented detail.

That kind of modeling directly informs climate policy, disaster planning, and renewable integration strategies, which require both precision and speed to stay ahead of real-world changes.

Medical research and digital twins take center stage

Healthcare is another big winner here. The system is expected to power complex simulations and “digital twins” for personalized medicine, as well as virtual models of organs, treatments, or even entire patients.

By combining AI and HPC, researchers can test therapies, optimize drug combinations, and analyze large genomic datasets significantly faster, transforming what used to be multi-year research programs into much tighter, iterative cycles.

Europe’s AI models finally get homegrown training muscle

Instead of relying on U.S. hyperscalers to train large language models and other AI systems, European researchers will be able to use Alice Recoque as an “AI factory.”

This means building foundational AI models in local languages, in accordance with EU data rules, with complete visibility into how and where the models are trained.

For startups and labs, it lowers the barrier to competing with global players in terms of model quality and scale.

Design choices strike a balance between extreme power and a practical footprint

Physically, Alice Recoque is massive but tightly optimized. It spans 94 racks, weighs around 280 tons, and occupies roughly 174 square meters for racks and services.

Yet, despite that scale, it multiplies CEA’s computing capacity by about 50 while only increasing electrical power needs by roughly five times, demonstrating the significant engineering effort that has gone into squeezing performance out of every square meter and kilowatt.

You might want to see how this push for extreme performance echoes in the broader industry by checking the reaction to Zuckerberg’s latest $100 million offers.

A new benchmark for Europe’s digital ambitions

When Alice Recoque comes online, it will stand as a flagship for Europe’s broader digital and AI ambitions.

It demonstrates how regional collaborations, targeted public funding, and carefully selected industrial partners can develop infrastructure that competes with the largest U.S. and Asian platforms.

In practice, it provides Europe with a powerful, sovereign tool to transform data, models, and research into real-world innovation more quickly than ever.

You might want to see how this race for next-gen power continues by taking a look at AMD’s new Helios AI racks.

What do you think about Eviden teaming up with AMD to create supercomputers in the surge of AI? Please share your thoughts and drop a comment.

Read More From This Brand:

• Qualcomm’s next big chip could outmuscle AMD and Intel on memory
• Stocks rise in early trading with AMD leading Wall Street’s AI-fueled gains
• IBM and AMD team up to shape the future with quantum computing

Don’t forget to follow us for more exclusive content on MSN.

If you liked this story, you’ll LOVE our FREE emails. Join today and be the first to get stories like this one.

This slideshow was made with AI assistance and human editing.