Sources: EuroHPC Joint Undertaking
The EuroHPC Joint Undertaking has awarded Bull a €387.8 million contract to build LUMI-AI, a new AI-optimised supercomputer that will be installed at CSC’s data centre in Kajaani, Finland. The system is expected to enter service in the second half of 2027 and will operate alongside Europe’s existing LUMI supercomputer.
LUMI-AI is designed to deliver ten times the AI capacity of LUMI and nearly twice its high-performance computing capability. The investment comes as demand for advanced computing is already putting pressure on available resources across Europe.
More than 3,000 scientific projects had used LUMI by June 2025, with nearly 7,000 users consuming around 140 million GPU hours and more than 3.8 billion CPU core hours. More than half of LUMI’s computing resources were being used for AI-related research and innovation, while a Finnish allocation call received GPU demand roughly three times higher than the resources available.
Limited GPU capacity can mean fewer training runs, smaller simulations and longer waits for computationally intensive work. LUMI-AI is being added at exactly that point in the infrastructure chain, where the ability to run more work has become as important as the performance of the machine itself.
Europe’s AI Compute Shortage
AI development has made large-scale computing relevant to organisations that do not necessarily have the resources or workload to build their own infrastructure. A startup may need substantial GPU capacity while developing a model but have little reason to maintain a permanent cluster once that phase of development changes.
The LUMI AI Factory is designed around this gap. It provides AI computing and support to startups, SMEs and academic researchers, including access to high-performance computing, datasets, training and technical expertise. Its work covers areas ranging from manufacturing and health to communications, climate, materials and language technologies.
The economics are different when computing can be accessed instead of owned. Money that would have gone into servers, GPUs, power and cooling can remain available for engineering, data and product development. At the same time, large bursts of computing can be used when a project actually needs them.
That matters because AI development is heavily experimental. Training a model is rarely a one-run exercise. Researchers and developers often need to change datasets, architectures or parameters and run the workload again. More available compute creates room for that experimentation.
AI Training and Inference at Scale
LUMI-AI will use AMD Instinct MI430X GPUs and sixth-generation AMD EPYC 256-core processors within Bull’s BullSequana XH3500 architecture. IBM Storage Scale will provide high-performance storage, while Nokia networking and Bull’s BXI interconnect will support the wider system.
A supercomputer cannot turn accelerator performance into useful output if data cannot move through the system efficiently. Storage, networking and interconnects determine how quickly computing resources can be fed with data and how effectively thousands of components can work together.
LUMI-AI is intended for AI training and inference alongside large-scale simulations and data-intensive scientific applications. The same computing infrastructure can therefore support model development, scientific research and industrial workloads rather than being tied to one type of AI application.
The practical gain is additional computational headroom. Larger datasets can be processed, more experiments can be attempted and workloads that compete for limited resources on existing systems can be moved onto a machine built specifically for a much higher AI workload.
AMD vs Nvidia in European AI Infrastructure
LUMI-AI will be funded through a European programme and hosted on European soil, but its core processors and accelerators will come from AMD, an American semiconductor company. That combination says more about technological sovereignty than a simple European-versus-American narrative can.
Europe is increasing its control over the infrastructure through which advanced computing is provided while still depending on international suppliers for critical components. Owning the facility and coordinating access to it is not the same as manufacturing every technology inside it.
The AMD choice also continues an established architecture. The current LUMI system already uses AMD EPYC processors and AMD Instinct accelerators, so LUMI-AI builds on technology already present within the European supercomputing environment.
There is no evidence that EuroHPC selected AMD specifically to challenge Nvidia. The more defensible point is that Europe’s publicly funded AI infrastructure is expanding around an AMD platform rather than concentrating all of its accelerator capacity around one supplier.
That can influence more than procurement. Hardware choices shape software environments, developer expertise and the skills required to operate the machines. A broader accelerator base gives public computing programmes more room to work across different technologies as the AI hardware market changes.
Bull and Europe’s Supercomputing Strategy
Bull gives the project another strategic dimension. France completed its acquisition of Bull’s advanced computing business from Atos in March 2026, making the company state owned. LUMI-AI is also Bull’s largest contract to date.
A French state-owned company is now delivering a major European computing system funded through a multinational consortium. Yet the platform remains international, with AMD, IBM and Nokia supplying critical technologies.
That combination is important. European technological sovereignty does not have to mean producing every component locally. It can also mean retaining control over system integration, infrastructure, hosting and access while using global suppliers for specialised technologies that Europe does not manufacture at comparable scale.
LUMI-AI is consequently European in its institutional structure without being entirely European in its technology stack.
Power and Cooling for AI Supercomputers
LUMI-AI will use Bull’s Direct Liquid Cooling technology, with heat generated by the system captured and reused through Kajaani’s district heating network. The facility is also designed to operate using renewable electricity.
As AI systems become denser, power and cooling become part of the computing equation. Thousands of accelerators cannot operate without a corresponding infrastructure for electricity, thermal management and data movement.
Kajaani adds another layer to that equation by using the heat produced by computing locally. Instead of removing the thermal output and treating it solely as waste, the project connects it to the district heating network.
The arrangement makes the economics of the facility broader than processor performance. Electricity consumption, cooling efficiency and the ability to reuse heat all affect what a large computing installation means for the place where it operates.
AI Factories, Supercomputers and Shared Compute
LUMI-AI is part of a wider European computing buildout. EuroHPC is overseeing 19 AI Factories across Europe and 13 AI Factory Antennas, while its AI Gigafactory programme targets much larger computing facilities.
The different layers exist because AI workloads vary enormously. Developing an application does not require the same infrastructure as training a frontier model, while scientific simulations can have entirely different computing and storage requirements.
LUMI-AI sits in the layer connecting substantial computing capacity with startups, SMEs, researchers and industry. The AI Factory model adds datasets, training and technical support alongside the hardware, making the resource more useful than GPU capacity alone.
This also gives the investment a wider purpose. Instead of creating another isolated supercomputer, Europe is connecting computing capacity to a network through which organisations across different sectors can access infrastructure and expertise.
The Economic Value of Europe’s AI Infrastructure
The €387.8 million contract adds capacity where existing demand is already stretching parts of Europe’s public computing infrastructure.
More GPUs will not automatically produce better AI. Engineering talent, data, software and useful applications still determine what organisations can build with the available hardware. But when compute becomes a limiting factor, additional capacity can remove one of the barriers between an idea and the amount of experimentation needed to develop it.
LUMI-AI puts that additional capacity inside a European access model. A startup does not need to build a supercomputer to use high-performance computing. A research project does not need to operate its own facility to run demanding simulations. Industry can access infrastructure designed for workloads beyond conventional computing environments.
Europe will still rely on global suppliers for important parts of the system. What changes is the infrastructure surrounding those components. LUMI-AI places more advanced computing capacity inside a shared European network, where access to compute becomes part of the region’s research and technology infrastructure rather than a resource available mainly to organisations large enough to build it themselves.
Frequently Asked Questions
1. Why is Europe building LUMI-AI when it already has LUMI?
LUMI is already operating under significant demand, with more than half of its computing resources being used for AI-related research and innovation. One Finnish allocation call recorded GPU demand at roughly three times the available capacity. LUMI-AI is designed to add substantially more AI and high-performance computing capacity.
2. What will LUMI-AI actually be used for?
LUMI-AI will support AI training and inference, large-scale scientific simulations and data-intensive workloads. Its computing resources will be available to startups, SMEs, researchers and industry through the LUMI AI Factory.
3. Why is LUMI-AI using AMD instead of Nvidia?
LUMI-AI will use AMD Instinct MI430X GPUs and sixth-generation AMD EPYC processors. The official procurement announcement does not say that AMD was selected specifically to challenge Nvidia. The choice also continues the AMD-based architecture already used by the existing LUMI system.
4. How will startups and SMEs benefit from LUMI-AI?
Startups and SMEs can access high-performance AI computing without having to build and maintain their own large-scale infrastructure. The LUMI AI Factory also combines computing access with datasets, training and technical support, allowing smaller organisations to use the infrastructure for AI research and development.
5. What does LUMI-AI mean for Europe’s AI infrastructure?
LUMI-AI adds a major AI-optimised computing resource to Europe’s wider AI Factory network. It gives European organisations access to more advanced computing capacity while keeping the infrastructure funded, hosted and coordinated through European institutions, even though important components such as AMD processors and GPUs still come from global suppliers.