Scaleway chooses AMD for its first supercomputer outside France

Scaleway announced on 13 October that it is deploying an AMD-based AI supercomputer in Sweden, its first supercomputer deployment outside France. The system will use Instinct MI355X accelerators and AMD’s ROCm software, expanding the computing architectures available through the European cloud provider.[1]

The announcement does not specify a GPU count, customer rental price or a general booking date. It establishes the project and its technology choices, rather than a complete catalogue entry that a buyer can price immediately.[1]

The memory is only part of the machine

AMD lists 288 GB of HBM3E memory and peak memory bandwidth of 8 TB/s for each MI355X. Those figures describe different properties: how much information an accelerator can hold close to its processors, and how quickly that memory can supply it. Neither number, on its own, is an application-speed measurement.[2]

Consider two hypothetical bottlenecks. A model that does not fit in available accelerator memory needs a different placement or representation. A model that fits can still spend time waiting for data. Increasing capacity addresses the first problem; increasing bandwidth addresses the second. A well-matched system needs both, with enough communication capacity to keep multiple accelerators working together.

Scaleway names Bull for integration, deployment and maintenance, Giga Computing for servers, HPE for Ethernet networking, VAST Data for the data platform and Motul for cooling fluid. The partner list describes a complete cluster project, not simply a purchase of graphics processors.[1]

That distinction matters when comparing competing offers. “The same GPU” can sit behind different storage, network and scheduling arrangements. For a customer, the useful unit is the completed workload: how long it takes, what it costs and whether the software used to run it can move elsewhere.

A Swedish relationship predates the cluster

The project follows Tele2 and Scaleway’s May 2026 partnership. Tele2 described an offering that combines its private-cloud services with Scaleway’s public cloud, and said a major energy-sector company was the first customer. The new accelerator deployment therefore sits alongside an existing route to Swedish enterprise customers.[3]

The May announcement framed the partnership around European control and the ability to combine private and public infrastructure. Those are related to, but not identical with, the later choice of AMD hardware. One concerns who supplies and operates the service; the other concerns the computing architecture available inside it.[3]

A company could consequently have several reasons to consider the system: a preferred European provider, a Swedish deployment location, a need for AMD capacity, or a project already connected to Tele2’s private cloud. Treating all of these as one vague promise of “sovereignty” would obscure what each purchasing decision actually changes.

The commercial significance is a wider set of combinations. European hosting need not imply one accelerator supplier, and an alternative accelerator need not require a company to build its own data centre. Scaleway has announced another route between those choices. The next concrete information for prospective users is the capacity and service terms they can actually reserve.

Report an error or suggest a correction