Scaleway plans Europe's first commercial AMD AI supercomputer in Sweden, selecting HPE networking

A. Key facts

On October 13, 2026, Scaleway announced at its annual AI conference in Paris that it would build an AMD Instinct MI355X-based AI supercomputer in Sweden. It will be Scaleway's first supercomputer outside France and Europe's first AMD-based AI supercomputer open commercially to startups, AI companies and other businesses.

Bull will assemble and maintain the machine, Giga Computing will supply servers, HPE will provide networking, VAST Data will supply storage and Motul will provide liquid-cooling fluid. Scaleway selected Sweden for its low-carbon electricity and local AI sector.

GPU count, aggregate compute performance, IT load or power capacity, contract value, Scaleway CAPEX and commissioning date were not disclosed. The project is company-confirmed through Reuters, but its detailed economic terms remain unavailable.

B. Business significance · InveDeck interpretation

For AMD, the project adds a commercial European cloud reference for MI355X and ROCm beyond research supercomputers. If customers move material training and inference workloads onto the system, it would be an important adoption signal for reducing dependence on Nvidia CUDA.

For HPE, it provides additional evidence that its networking portfolio can win in high-performance GPU clusters serving neocloud and sovereign-AI markets. Because GPU utilization and cluster efficiency depend on network performance, the strategic relevance is greater than that of a generic peripheral sale.

However, Scaleway also announced an Nvidia HGX B300 site in Copenhagen on the same day. The event therefore represents northern-European capacity growth and a multi-vendor customer-choice strategy, not a decision to replace Nvidia wholesale with AMD.

C. Quantifiable impact

The disclosed product specifications are 288GB of HBM3E memory and 8TB/s of memory bandwidth per MI355X GPU. Without a GPU count, total HBM capacity, system compute, rack count, power and cooling demand cannot be calculated.

The absence of contract value and shipment volume prevents estimation of AMD or HPE revenue, Scaleway CAPEX, supplier backlog or future cash-flow impact. No project-size assumptions have been introduced.

Despite its strategic distinction as Europe's first commercial AMD AI supercomputer, available information is insufficient to determine whether the order is large enough to move company-wide financial results. GPU count or MW disclosure is required for quantification.

D. Link to earnings and expectations

AMD reported second-quarter 2026 revenue of USD 11.5 billion and Data Center revenue of USD 6.7 billion, up 107% year over year. Management identified scaling Instinct deployments as a second-half growth driver. The Scaleway project supports that direction but does not change existing guidance.

HPE reported fiscal third-quarter 2026 revenue of USD 12.2 billion. Networking revenue was USD 2.9 billion, including USD 382 million from Data Center Networking, up 112.2% year over year. HPE expects that business to grow at a low-to-high-50s percentage CAGR through fiscal 2029.

The project is a customer-adoption proof point consistent with HPE's growth framework, but its undisclosed value means it should not be treated as evidence for a guidance increase. Neither AMD nor HPE changed earnings, CAPEX or FCF guidance with this announcement.

E. Industry and supply chain · InveDeck interpretation

Direct positive exposure includes AMD MI355X and ROCm, HPE data-center networking, Giga Computing servers, VAST Data storage and Motul liquid cooling. MI355X adoption is directionally positive for HBM3E and advanced packaging, but suppliers and volume were not disclosed.

For Nvidia, the project increases competitive pressure from an AMD alternative in European commercial cloud. Yet Scaleway's simultaneous HGX B300 site indicates expansion of total northern-European AI capacity and workload-level supplier diversification rather than a direct reduction in Nvidia demand.

European data sovereignty and demand for low-carbon electricity strengthen the location case for Nordic data centers. Without disclosed power capacity or power contracts, grid, cooling and data-center demand cannot be quantified.

F. What to watch next

• MI355X GPU count, server and rack configuration, and total IT load or MW
• Total contract value, AMD and HPE shipment volume, Scaleway CAPEX and commissioning date
• Actual ROCm customer workloads, utilization and price-performance versus Nvidia
• HPE networking product mix, revenue-recognition timing and Data Center Networking backlog
• Customer and workload allocation between the Swedish AMD system and Copenhagen Nvidia site