Scaleway to build AMD-powered AI supercomputer in Sweden

Scaleway is building an AI supercomputer in Sweden, its first facility of this kind outside France. The machine will use AMD Instinct MI355X chips. It is designed for commercial use, rather than only for academic research. Customers across Europe will be able to access it.

AI supercomputers provide large amounts of computing power for developing and operating artificial-intelligence systems. Startups can use them to train models, test applications, and run demanding services. The article does not specify the machine’s size or exact capacity. It does identify the chip, networking, storage, cooling, and construction partners.

The system will be assembled and maintained by Bull. Giga Computing will supply servers, HPE will provide networking, VAST Data will provide storage, and Motul will supply cooling liquid. Scaleway describes the project as the first AMD-based AI supercomputer for commercial use in Europe, expanding infrastructure choices for European customers.

Scaleway chose Sweden mainly because of its largely low-carbon electricity and its AI sector. Electricity is central to large computing facilities, so the energy profile made Sweden attractive for locating an AI supercomputer. The article does not provide a specific emissions figure or describe the facility’s exact power source.

The location also fits Scaleway’s commercial expansion. The company entered the Swedish market in May, when it signed an agreement with telecom operator Tele2 to sell cloud and AI services there. That relationship gave Scaleway an existing business connection in the country.

Tele2’s business-to-business head, Stefan Trampus, said the project sends a clear signal about Sweden’s attractiveness for AI infrastructure. The choice therefore combines environmental considerations, local AI activity, and market access. Sweden becomes the base for Scaleway’s first AI supercomputer outside France, serving customers across Europe.

Nvidia-based systems are far more widespread in global AI computing than AMD-based systems. The article states that Nvidia’s chips power most AI computing globally. It does not provide a percentage, shipment total, or direct comparison of the installed numbers for Nvidia and AMD.

AMD is nevertheless gaining a place in European infrastructure. Its Instinct MI355X is described as the chipmaker’s answer to Nvidia’s Blackwell line. AMD chips already power research supercomputers such as LUMI in Finland. Oracle and Vultr also rent AMD chips to European customers through their cloud services.

Scaleway’s Swedish machine would add a commercial AMD-based supercomputer in Europe. That matters because customers could access AMD hardware without relying only on Nvidia systems. The broader shift remains limited in the article’s description: Nvidia still dominates global AI computing, while AMD is expanding its presence through supercomputers, cloud rentals, and open software.

The supercomputer could give European startups, AI companies, and businesses more choice in how they obtain AI computing. Scaleway says the system will be open to customers across the continent. Its AMD hardware adds an alternative to the Nvidia systems that dominate global AI computing.

The key mechanism is access to AMD Instinct GPUs and the ROCm software ecosystem. Developers can use ROCm with widely used AI programming tools that also work on Nvidia chips. This may make it easier for some teams to use AMD infrastructure without abandoning familiar development tools. The article does not promise lower prices or faster performance.

The project could strengthen Europe’s AI infrastructure by adding commercial capacity and another supplier ecosystem. Bull said it would enhance Europe’s AI capabilities. The machine will also be assembled and maintained by a European partner, while companies from several technology sectors supply its components and systems.

The project brings together several companies with different technical responsibilities. Bull, the supercomputer builder, will assemble and maintain the machine. Bull already collaborates with Scaleway through the AION consortium and described the contract as a positive strategic step.

Giga Computing will provide the servers that hold and run the computing hardware. HPE will supply networking, which connects the system’s components and supports communication with users and services. VAST Data will provide storage for data and software. Motul will supply the cooling liquid used to manage the equipment’s heat.

AMD is the chip provider, supplying Instinct MI355X AI chips, while Scaleway is developing the facility and making it available to customers. Together, these roles cover construction, maintenance, compute hardware, connections, data storage, and cooling. The article identifies no other suppliers or detailed contract values.

ROCm and CUDA are software ecosystems that AI developers use to write and run applications. AMD’s ROCm is open-source, while Nvidia’s CUDA is proprietary. This difference affects how freely developers and customers can use the software with different hardware.

According to the article, ROCm runs widely used AI programming tools that also work on Nvidia chips. That can help developers move between AMD and Nvidia systems. CUDA, by contrast, runs only on Nvidia chips. Applications developed for CUDA are therefore locked to Nvidia hardware, according to the article.

The distinction gives AMD a way to compete beyond its physical chips. Scaleway says its Swedish system will offer AMD Instinct GPUs together with the open ROCm ecosystem. Nvidia’s established CUDA ecosystem remains important because Nvidia powers most AI computing globally. The article does not claim that ROCm already has the same adoption as CUDA.

GPUs, or graphics processing units, are processors designed to perform many calculations at the same time. Although first associated with graphics, they are also well suited to the repeated mathematical operations used by artificial-intelligence models. This explanation uses established computing knowledge; the article names GPUs but does not define them.

During training, an AI model processes large datasets and repeatedly adjusts its internal settings. GPUs can perform many of those calculations in parallel, reducing the time needed compared with using only general-purpose processors. After training, GPUs can also run a model’s calculations when people use an AI application. Their parallel design is the key mechanism.

The article applies this role to AMD Instinct GPUs and Nvidia’s AI chips. Scaleway’s Swedish supercomputer will make AMD GPU capacity available commercially. The article does not provide performance comparisons, training times, or the number of GPUs that the facility will contain.