For its first supercomputer outside France, Scaleway picks AMD over NVIDIA
Set to be installed in Sweden, it joins the thousands of NVIDIA GPUs planned in Copenhagen, and Scaleway now offers two Chinese AI models. What has been announced, and what hasn't yet.
Translated from French by AI. The original French article is the reference version.

For its first AI supercomputer outside France, Scaleway did not choose NVIDIA. The machine, which will be installed in Sweden, will run on AMD Instinct MI355X graphics processing units (GPUs). Scaleway, the cloud subsidiary of the iliad group, unveiled it on October 13 at Station F in Paris, during its annual ai-PULSE conference. According to Reuters, it is Scaleway's first system of this kind outside its home country.
How many chips, how much power, and when? On these points, Scaleway's press release is silent: it gives neither the computing power, nor the commissioning date, nor the amount of the investment.
Paris, Copenhagen, and now Sweden
According to Journal du Net, Scaleway now organizes its high-performance computing offering into three zones, for building and running AI models. Paris and Copenhagen rely on NVIDIA, and the Danish zone will host "thousands" of HGX B300 GPUs. The third is in Sweden, with AMD. Journal du Net puts it at 696 MI355X GPUs, a figure that does not appear in the press release.
Around the Swedish machine, a chain of partners. Integration and maintenance are handled by Bull. The servers come from Giga Computing, the network from HPE, the storage from VAST Data and the cooling liquid from Motul. Why Sweden? Scaleway highlights its largely low-carbon electricity. The company is already present there through a partnership with the operator Tele2, announced in May.
And a third chip supplier is on the way. Journal du Net reports that the Jotunn 8 inference chips from French start-up VSORA are due to be installed in the first half of 2027. According to the same outlet, the first customers have already signed.
Two Chinese models in the catalog
Scaleway's on-demand inference service is adding three models: LightOn OCR 3, from French company LightOn, for document extraction, as well as GLM 5.3 (Z.ai) and DeepSeek V4.1 Flash, two Chinese models. According to Journal du Net, Mistral Large 4, unveiled by Mistral AI on October 6, is due to arrive in the coming weeks. The catalog already includes more than 40 models.
The message focused on independence more than on sovereignty. Damien Lucas, Scaleway's CEO, emphasized an "immunity" from extraterritorial laws. According to Journal du Net, the AMD and VSORA offerings provide redundancy against the risk of a US "kill switch", meaning an access cutoff decided in Washington.
Dependence does not disappear, though. The chips are still designed in the United States, at NVIDIA as well as at AMD.
What to check before committing
- For CIOs: having several GPU suppliers (NVIDIA, AMD, soon VSORA) reduces dependence on a single manufacturer. However, software must be compatible with ROCm, AMD's development environment.
- Chinese models: being hosted in Europe by Scaleway does not remove the need to assess them before adoption, in particular their biases and their restrictions on certain content.
- Timeline: no commissioning date has been given for Sweden. It will need to be confirmed before building a project on it.
Method: AI-assisted monitoring, verified sources, article reviewed and approved by Frédéric LENQUETTE.
Share this article