Summary
- The €387.8m package covers acquisition, delivery, installation, and maintenance of LUMI-AI.
- The system will use AMD Instinct MI430X GPUs and sixth-generation EPYC processors.
- EuroHPC expects LUMI-AI to provide ten times the AI capacity of the existing LUMI supercomputer.
The EuroHPC Joint Undertaking has signed a €387.8m procurement contract with Bull for a new AI-optimised supercomputer at CSC’s data centre in Kajaani, Finland.
LUMI-AI is expected to be installed and made available to users during 2027. EuroHPC says the system will provide around ten times the artificial-intelligence capacity of the existing LUMI supercomputer, substantially increasing the compute available through the LUMI AI Factory.
The system will use next-generation AMD Instinct MI430X GPUs alongside sixth-generation AMD EPYC processors with 256 cores. It is intended to support AI workloads using large, dynamic, and confidential datasets as well as large-scale simulation and data-intensive scientific computing.
The procurement package covers acquisition, delivery, installation, and maintenance. EuroHPC will provide half of the €387.8m budget through the Digital Europe Programme, with the remaining 50% funded by the LUMI AI Factory consortium.
Kajaani adds another layer of compute
The consortium is led by Finland and includes Czechia, Denmark, Estonia, Norway, and Poland. CSC – IT Center for Science will host the machine at its Kajaani data centre, extending a site that already houses major European high-performance computing infrastructure.
LUMI-AI is primarily aimed at startups and small and medium-sized businesses carrying out research and innovation, although wider research users will also be able to access the system. Allocation of computing resources will be shared between EuroHPC and the consortium according to their respective investment.
That model puts the physical data centre underneath a wider industrial-policy programme. European governments are not merely subsidising AI applications; they are funding processors, networking, hosting infrastructure, maintenance, and access mechanisms intended to give European organisations compute that would otherwise have to be purchased from commercial platforms.
The engineering requirement is substantial even when the procurement headline is expressed in computing terms. New accelerator systems bring concentrated electrical loads, high heat densities, networking requirements, and commissioning complexity into an existing data centre environment. The contract announcement does not disclose LUMI-AI’s electrical demand or cooling architecture, so those facility impacts cannot yet be quantified.
They will be important to delivery. The performance of an AI supercomputer depends not only on processors but on power distribution, thermal management, storage, interconnects, and the ability to keep all of those systems operating under sustained load.
AI factories move towards gigafactories
LUMI-AI is part of a wider EuroHPC expansion. The Joint Undertaking says it is overseeing implementation of 19 AI Factories across Europe, supplemented by 13 AI Factory Antennas intended to extend access to the network.
Its mandate has also expanded beyond the current AI Factory programme. Regulation adopted in 2026 added new responsibilities around European AI Gigafactories and quantum technologies, and EuroHPC launched a tender process at the end of July to select consortia for future AI Gigafactories.
The distinction is one of scale as well as policy. As publicly supported AI infrastructure grows, host sites will face the same constraints as commercial data centres: electricity availability, cooling, grid connections, construction lead times, hardware delivery, and secure operation. Public procurement cannot avoid those physical limits simply because the capacity is being commissioned for strategic reasons.
Kajaani’s advantage is that the facility already operates at the centre of a European supercomputing ecosystem. That gives LUMI-AI an established host, operational capability, and user programme rather than requiring a new campus to be built from scratch.
The next test is installation. Bull was selected after a tender launched in May 2025, and the 2027 availability date now sets a firm delivery window for a system intended to become one of Europe’s larger publicly backed AI-compute assets.
