Test report DSG-2244 · Rev A · tested October 15, 2026
AI Datacenter InfrastructureDevice under test
Scaleway adds two Milan availability zones, touts European AI reach
Scaleway has brought two new Milan-area availability zones online, giving its Italy cloud region three sites. Stockholm and Frankfurt each get three zones in 2027.
- Read
- 3 min
- Words
- 668
- Node
- 5nm
- Operator
- Priya Raman
Spec summary
- Scaleway added two availability zones in Basiglio and Milan, bringing the Italy region to three zones.
- The Italy cloud region originally opened in March 2026 with one zone in Settimo Milanese, Milan.
- Three availability zones in Stockholm and three more in Frankfurt are planned for 2027.
- Eight Jotunn8 processors in the YMIR server deliver 25.6 petaflops, 2.3TB of memory and 9kW TDP.
- Scaleway now lists AI compute capacity in Paris, Copenhagen and Stockholm.
Scaleway now operates three availability zones in its Italy cloud region. The two new zones went online in Basiglio and Milan, joining the original zone in Settimo Milanese. That first zone opened in March 2026, when the region debuted with a single-site footprint — a configuration that limited the company's resilience claims for nearly a year.
An availability zone (AZ) is a physically separate data center cluster with independent power, cooling and networking. Two zones in the same metropolitan area do not equal geographic redundancy at continental scale. Scaleway's own framing acknowledges the distinction. The company said in its announcement: "With three geographically separate and independent zones, Milan becomes a fully fledged cloud region, offering enterprise-grade resilience."
The word "enterprise-grade" is doing considerable work in that sentence. Scaleway did not publish uptime targets, inter-zone latency figures or per-zone capacity numbers. Buyers comparing Milan against hyperscaler regions in the same country will find no like-for-like service-level data in the announcement.
What comes next in the 2027 expansion?
The Milan buildout is not the end of the roadmap. During its AI-Pulse 2026 event, Scaleway said it will expand capacity across Europe in 2027. The plan includes:
- Three availability zones in Stockholm, Sweden.
- Three additional availability zones in Frankfurt, Germany.
- Continued operation of existing regions in France, Poland and the Netherlands.
Frankfurt already hosts Scaleway capacity; the 2027 work adds zones rather than a new market. Stockholm would be new territory. No opening dates beyond "2027" were given, and the company did not state how the Swedish sites will be powered — a question that matters for buyers with energy-origin requirements in their procurement rules.
Where does the GPU capacity stand?
The same event brought news of GPU capacity launches in Sweden and Copenhagen. Scaleway now lists AI compute in Paris, Copenhagen and Stockholm. The company says the buildout will eventually "represent tens of thousands of GPUs, tens of exaflops of computing power and tens of megawatts of capacity."
Every figure in that sentence is an approximation with no deadline attached. "Tens of thousands" spans anywhere from 20,000 to 90,000 GPUs. "Eventually" carries no date. Scaleway has not broken the totals down by accelerator model, by site, or by when each tranche becomes rentable. For capacity planning purposes, the statement functions as a direction of travel, not a commitment a customer can book against.
What about the Jotunn8 inference bet?
The more concrete numbers sit in Scaleway's silicon partnership. Earlier this year, the company said it is working with VSORA and ZML on a European AI inference offering based on the Jotunn8 processor. VSORA is a French semiconductor firm specializing in AI inference chips; ZML is the software partner. The Jotunn8 is currently in manufacturing.
The published specifications are specific. Eight Jotunn8 chips deployed together in the YMIR inference server deliver 25.6 petaflops of compute, 2.3TB of memory and a total TDP (thermal design power, the sustained heat a system must dissipate) of 9kW. For comparison, that memory pool matches what a well-configured dual-socket x86 server offers, but at an order of magnitude higher compute throughput.
Caveats remain. Manufacturing is not deployment. Scaleway gave no date for general availability, no pricing and no benchmark results against incumbent inference hardware. Until the YMIR ships in volume, the specs describe a design target rather than a product a customer can rent by the hour.
Why the Milan expansion matters now
Three zones in one metro area is the baseline European enterprises typically demand for production workloads. Scaleway has now met that bar in Italy, its fifth operating market alongside France, Poland, the Netherlands and Germany. The 2027 Stockholm and Frankfurt additions would extend that model.
What the announcement does not say is equally relevant: no revenue figures, no customer names for the Italian region and no utilization data. The expansion signals capital commitment. Whether demand follows is a question the 2027 numbers will have to answer.
via Data Center Dynamics (Source)
More from Priya Raman
Show full bio
Correspondent covering money, funding and market shifts around hardware layer of artificial intelligence at Die Signal.
379 articles
Same lot · LOT-C1CB
- DSG-207528nmScaleway Plans GPU Cluster Deployments in Sweden and Copenhagen
- DSG-43903nmScaleway to Deploy AMD MI355X Supercomputer in Sweden, Nvidia GPU Zone in Copenhagen
- DSG-57425nmScaleway to build AMD-powered AI supercomputer in Sweden
- DSG-58403nmSpaceX and Nvidia target 2028 for orbital data centers at scale
- DSG-617714nmCivo to build 40 UK Edge data centers, starting with 8MW in Hertfordshire