AI 基礎設施
Europe Signs Contract to Build LUMI-AI, Unifying AI and HPC Workloads with MI430X GPUs and 256-Core EPYC CPUs
EuroHPC and Bull have signed a €387.8 million contract to deploy LUMI-AI in Finland, built around AMD MI430X GPUs and sixth-generation EPYC CPUs. The system is expected to come online in 2027, but the projected tenfold increase in AI capacity remains a planning estimate, while node counts, interconnects, and storage configurations have yet to be fully disclosed.
The European High Performance Computing Joint Undertaking (EuroHPC JU) has signed a procurement contract with France’s Bull for LUMI-AI, confirming that the AI-optimized supercomputer will be installed at CSC’s data center in Kajaani, Finland. The €387.8 million budget covers procurement, delivery, installation, and maintenance. EuroHPC will fund half through the Digital Europe Programme, with the remainder provided by a consortium comprising Finland, Czechia, Denmark, Estonia, Norway, and Poland. According to the [EuroHPC announcement](https://www.eurohpc-ju.europa.eu/eurohpc-ju-signs-contract-deploy-lumi-ai-supercomputer-2026-08-31_en), the system is expected to become available in 2027.
The compute nodes will use AMD Instinct MI430X GPUs and sixth-generation EPYC processors with 256 cores per CPU. The MI430X targets both scientific computing and model training. AMD lists per-GPU maximum specifications of 432GB of HBM4, 23.3TB/s of memory bandwidth, 288 TFLOPS of FP64 performance, and 9.2 PFLOPS of FP4 performance. This range of numerical precision allows a single infrastructure platform to run conventional simulations, train models on simulation data, and then perform low-precision inference without splitting the workflow across two separate accelerator clusters. [AMD’s system overview](https://newsroom.amd.com/news/amd-instinct-gpus-epyc-cpus-power-lumi-ai-supercomputer/) also mentions a multi-tenant architecture and API-based access.
Officials estimate that LUMI-AI will deliver ten times the AI capacity of the existing LUMI system and nearly twice its HPC capability. It will serve as the computing core of the LUMI AI Factory, supporting startups, small and medium-sized enterprises, and research teams, while integrating with the LUMI-IQ quantum experimentation platform. The data center is also planned to use renewable energy and recover waste heat.
However, the tenfold capacity figure is a projection rather than a completed benchmark result. Officials have also not disclosed the total GPU count, Bull interconnect topology, storage throughput, availability targets, or the methodology used to calculate capacity. The factors research teams will need to watch most closely are the maturity of the ROCm software stack, multi-tenant isolation, scheduling policies, and cross-border resource allocation. These considerations often determine whether large models can scale efficiently more directly than peak FLOPS do.