Scaleway announced two GPU deployments outside France on October 13, during its AI-Pulse 2026 event in Paris. One is an AMD-based supercomputer in Sweden. The other is an Nvidia-backed GPU zone in Copenhagen. Neither came with capacity numbers.

The company did not disclose the compute capacity of the Swedish system. It did not disclose the number of GPUs. It did not provide a deployment date. Those omissions matter for any customer planning capacity around these announcements.

What does the Swedish deployment actually contain?

Scaleway will build an AI supercomputer in Sweden based on AMD Instinct MI355X GPUs (AMD's data center accelerator line). The company calls this its first supercomputer deployment outside France. It positions the system for startups, AI companies, and enterprises across Europe.

CEO Damien Lucas framed the location as strategic. He said: "We are excited to deepen our work with AMD and bring the power of AMD Instinct technology to companies across Europe at this scale. Doing it from Sweden makes this milestone particularly meaningful, as we continue to grow our presence and investment across Europe."

Note the phrasing. "At this scale" is a claim without a number attached. Scaleway has not said what scale means here.

AMD's EMEA senior vice president, Stephanie Dismore, added: "Scaleway is expanding the AI infrastructure choices available to customers across Europe with AMD Instinct GPUs and our open ROCm software ecosystem." ROCm is AMD's open-source compute software stack, positioned against Nvidia's CUDA.

Vast Data will supply the data platform for the system. Scaleway disclosed no contract value or storage specifications.

The Swedish deployment builds on an existing partnership with Tele2, announced in May, under which the two companies planned a joint cloud and AI offering in Sweden. Tele2 operates two data centers in the country. Scaleway has not said which site will host the supercomputer.

How does the MI355X compare?

AMD launched the MI355X in June 2025. The chip uses 3nm manufacturing and the CDNA 4 architecture. It carries 288GB of HBM3E memory (high-bandwidth memory for accelerators) with 8Tbps of memory bandwidth. FP64 compute reaches 79 teraflops. Total board power runs up to 1,400W.

AMD claims advantages over Nvidia's GB200 and B200: 1.6x more memory capacity, 2x higher peak FP64 performance, and 1x more memory bandwidth. The claims are AMD's own, not independent benchmarks. The company also cites a 4x generation-on-generation AI compute increase and a 35x generational gain in inference.

Vendor comparisons like these use favorable workload profiles. Buyers should treat them as a starting point, not a verdict.

What is the Copenhagen GPU zone?

Scaleway says it will launch a GPU zone in Copenhagen offering Nvidia HGX B300 compute. The company defines GPU zones as "locations designed to provide large-scale GPU capacity for AI training and inference, complementing its existing cloud Availability Zones."

How a GPU zone differs from the Swedish supercomputer remains unclear. Scaleway has not clarified the distinction. DCD has contacted the company for comment.

Lucas tied the Copenhagen deployment to European sovereignty concerns. He said: "Access to high-performance AI computing is becoming a strategic priority for European organizations, and they increasingly want that capacity delivered by providers they can trust. With our new GPU Zone in Copenhagen, we're responding to that demand by bringing the latest Nvidia technologies closer to customers, while staying true to our commitment to European technological independence."

There is a tension worth noting. The sovereignty argument pairs with Nvidia hardware, a US supplier under US export controls. Scaleway did not address that contradiction in its announcement.

What remains unanswered?

The announcements leave gaps:

• No GPU counts or total compute capacity for either site
• No timelines for deployment or availability
• No pricing information for customers
• No clarification on the difference between a GPU zone and a supercomputer
• No explanation of how the Nvidia and AMD offerings will be positioned against each other

Two GPU vendors, two countries, one event, and no hard capacity figures. Customers in the Nordics now have a promise of choice. They do not yet have a date or a price.