European Sovereign Compute for Defence AI
Chips, accelerators and HPC dependencies in the European model-to-weapon stack
European military AI sovereignty cannot be assessed only at the level of models, algorithms or defence applications. Its first constraint is compute. Training, fine-tuning, deploying and operating AI systems for defence require access to processors, accelerators, memory architectures, interconnects, software stacks, high-performance computing infrastructure, secure data-centre capacity and edge-inference hardware that can be governed under European strategic, industrial and security priorities. If these layers remain dependent on non-European accelerators, non-European software ecosystems and non-European fabrication capacity, then European AI models and defence-AI applications remain structurally exposed, even when they are developed, trained or deployed by European firms. The relevant question is therefore not whether Europe possesses strong AI companies, advanced supercomputers or important semiconductor assets. It is whether Europe can control the compute chain that connects chips, accelerators, HPC systems, AI Factories, sovereign cloud, classified deployment and tactical military workloads inside the European model-to-weapon stack.
This report analyses that chain from the semiconductor policy base to military use cases. It begins by defining Europe’s compute-sovereignty problem, distinguishing territorial control, access governance, processor design, accelerator hardware, fabrication, software-stack dependence, classified deployment and tactical-edge inference. It then examines the European Chips Act, the Chips Joint Undertaking, EuroHPC, JUPITER and the AI Factories as the institutional and infrastructural base of Europe’s sovereign-compute strategy. JUPITER is treated as the central case study because it shows both the strength and the limits of Europe’s position: a European public exascale infrastructure governed through EuroHPC, but still dependent on non-European accelerator technology for the AI-critical layer. The report then assesses SiPearl and the European processor path, the near-term opportunity in inference and edge AI, the role of Axelera AI, VSORA, Openchip and Kalray, the strategic leverage of ASML, and the embedded, secure and edge-compute relevance of STMicroelectronics, Infineon and NXP. It concludes by translating these dependencies into defence workloads and by assessing whether Europe is moving towards full sovereign compute, hybrid sovereignty, inference-first autonomy or a form of dependency concealed behind European infrastructure.