crossroads magazine Mr Grotz. “A model that works for everyone – from a ten-person fintech startup to a ministry – scales dif- ferently.” Luxembourg’s national supercomputer, MeluXina, is a good example of this. When it was launched in 2021, most European supercomputers were primar- ily designed for research simulations. MeluXina was designed for business use from day one. Its architec- ture enables simulation, modelling, data analytics and AI workloads on a single system. It was AI-ready before AI became the dominant supercomputer workload – and this didn’t happen by coincidence. In addition, MeluXina’s hybrid architecture sup- ports workloads of different sizes. This makes it ac- cessible to companies that don’t need full capacity for their projects. AI Factories: The bridge to impact But making infrastructure available doesn’t automat- ically translate into use. Particularly smaller compa- nies find it difficult to navigate the complex landscape of digital infrastructure. “The demand is real,” said Mr Grotz. “But for some businesses, the ramp-on is the bottleneck.” To bridge this gap, the EU has set up a network of 19 AI Factories that give access to computing power, data and talent. Luxembourg’s own AI Factory targets companies at all stages of their AI journey, whether they are just starting or seeking the power and exper- tise to scale their growth. Replacing “think big” with “think smart” NVIDIA’s CEO Jensen Huang described the global AI boom as the “largest infrastructure buildout in hu- man history”. Most of the world’s AI computing power is in the US, while Europe is investing heavily to catch up. Luxembourg will never compete on raw scale, but scale has never been the point. “Small countries that think clearly and invest consistently can lead in the computing continuum – not despite their size, but because of it,” explained Mr Grotz. “If you have access to the best data, the right infrastructure and the domain expertise that matches your use case, you don’t need to have the big- gest model.” OVHai LLM is the perfect example of this approach. In 2025, it rolled out a new, highly efficient LLM (large language model) architecture. Instead of consulting all available data when responding to a prompt, it consults only specific memory blocks. This allows it to provide faster responses using smaller machines that consume less energy. The algorithm was trained on the European network of supercomputers, of which MeluXina is a part. “We have proven that, with a team of 15, computing power made available by the Euro- pean Commission and European data, it is possible to develop successful ‘made in the EU’ foundation mod- els,” Mr Debeugny said. Mr Grotz extends an invitation to companies world- wide looking to innovate in the AI space for the Eu- ropean market. “Luxembourg is not large enough to pretend that what it builds will be enough to serve Europe. But it is positioned – by size, openness and the concentration of its ecosystem – to be a testbed, a reference model and a proof of concept for inclu- sive sovereignty at national scale.” The infrastructure is built. The framework is in place. The question now is who decides to use it. AI SOVEREIGNTY