Eva-Maria Hempe est executive leader in AI commercialisation and market creation chez Nvidia. 
For them, European sovereignty will not be an ‘all-or-nothing’ proposition, but rather a nuanced management of interdependencies. Thus, the question of sovereignty in artificial intelligence cannot be reduced to mere technological independence from actors who provide both the infrastructure and the tools. The debate would instead centre on Europe’s ability to make informed strategic choices at every layer of the technology stack. For businesses, this means strategic autonomy to avoid being locked into single solutions. For governments, it is about controlling the dissemination of knowledge and the cultural influence conveyed by the models. Together, they have identified several structural and regulatory obstacles that are holding back European momentum.
The obstacles to sovereign AI
First and foremost is the dominance of foreign models. Today, more than 50% of the models downloaded from platforms such as Hugging Face are of Chinese or American origin.
The second obstacle is regulatory in nature. Although the EU AI Act is seen as a necessary framework for safety and trust, the uncertainty surrounding the rules – particularly regarding the protection of intellectual property (IP) during model training – is a cause for concern among creators. A misguided focus on training rather than outcomes could prompt developers to exclude European data to avoid complex legal risks.
Access to resources for SMEs is identified as the third obstacle. Whilst large corporations manage to cope, small businesses struggle to access the market and computing power, whilst facing fierce competition from US digital services that capture hundreds of billions of euros in European spending every year. The final obstacle highlighted is the transition from research to industry. According to the speakers, there is a barrier between academic infrastructure (supercomputers) and the private sector. Adoption remains the major challenge, as universities are not always accustomed to collaborating with businesses.
Ways forward for sovereign AI
In the face of these challenges, experts suggest several courses of action to strengthen Europe’s position. Firstly, focusing on specialisation and small language models (SLMs). Rather than seeking to dominate the global market for general-purpose models, Europe should focus on smaller, specialised and efficient models (Small Language Models). The use of proprietary data specific to European industries (finance, healthcare, manufacturing) offers a unique competitive advantage in creating real AI for real-world problems.
Open source is seen as a driver of autonomy. The adoption of open-weight models is considered crucial. It enables companies to build on existing foundations and fine-tune them locally, thereby ensuring that data and know-how remain within Europe. These models are also perceived as more explainable and more secure, potentially requiring lighter regulation.
Europe has invested heavily in supercomputers such as Meluxina in Luxembourg. The next step would be to transform these capabilities into AI Factories that businesses can access in record time (sometimes within a matter of days) to test and deploy models. The focus must be on a trusted, certified infrastructure capable of handling sensitive data. This is a niche in which Luxembourg has positioned itself with its AI Factory.
Although often cited as a hindrance, regulation—positioned as a ‘foundation of trust’—could become an asset if it provided the necessary basis for moving from innovative prototypes to large-scale solutions by clearly defining risks and responsibilities. This would enable entrepreneurs to build on a stable foundation.
One final point: stimulating local demand. To support its leading companies, Europe should make it easier for public bodies to procure local solutions and combat ‘sovereign washing’ – in other words, foreign solutions masquerading as domestic ones.
Their conclusion? By building on its strengths – its industrial data, research talent and computing infrastructure – Europe can become a leader in trusted and specialised AI applications.
