The European Union plans to open a faster route to its public supercomputers for AI startups that want to train models. The proposal links access to this computing capacity with the bloc’s push for AI development it considers ethical and responsible.
A new route to computing capacity
EU president Ursula von der Leyen announced the initiative during the annual State of the Union address. The bloc already lets industry use its high-performance computing systems through application calls, but the process asks applicants to explain their need for substantial computing time, data storage and support.
A European Commission spokesman said that access policy would be adjusted to create a dedicated, swifter route for small and medium-sized enterprises and AI startups. Ethical criteria already used to assess research projects could also help evaluate candidates for access under an AI-specific scheme.
The EU’s existing systems include pre-exascale and petascale supercomputers. The article identifies eight supercomputers located around the bloc, including Lumi in Finland, MareNostrum 5 in Spain and Leonardo in Italy. Two more powerful exascale systems, Jupiter in Germany and Jules Verne in France, were expected to come online in the future.
Access tied to responsible development
The proposal is part of a broader EU approach that pairs support for innovation with governance. The Commission has promoted voluntary principles while formal rules are still being negotiated, and von der Leyen said companies should commit to the principles of the AI Act before it takes effect.
The AI Act is described as a risk-based framework for regulating AI applications. The EU was also working with the United States and other international partners on an AI Code of Conduct, with the aim of helping bridge differences between national approaches.
Von der Leyen framed the EU’s work around guardrails, governance and guiding innovation. Alongside potential benefits such as better healthcare, productivity and climate action, she pointed to concerns about serious AI risks and argued for a global framework.
She also proposed a body that could bring scientists, technology companies and independent experts together to inform policymakers about AI’s risks and benefits. The idea drew on the role of the IPCC in providing climate science to policymakers, and would build on work including the G7 Hiroshima Process.
Potential benefits and open questions
Thierry Breton, the EU’s internal market commissioner, said access to European supercomputers could help startups reduce model training time from months or years to days or weeks. The initiative would identify promising European AI startups and provide access to computing capacity.
That could make a scarce resource available to selected companies and support the EU’s effort to encourage AI development aligned with European values. How much competitive advantage the program would deliver remained unclear in the source article.
The supercomputer plan also sits alongside other EU efforts to support AI development, including Testing and Experimentation Facilities for AI, Digital Innovation Hubs, regulatory sandboxes under the incoming AI Act, the European Partnership on AI, Data and Robotics, and the HorizonEurope research program.
Bringing more voices into AI governance
Breton also announced that he would convene the European AI Alliance Assembly in November. The gathering was intended to include startups, businesses that use AI, consumers, NGOs, academic experts and policymakers, alongside large technology companies.
The European AI Alliance began as an online discussion forum and also held in-person meetings and workshops. The Commission said the alliance had not met for the past two years, and that the November assembly would take place during an important stage in the AI Act’s adoption process. Discussions would cover implementation of the AI Act and AI Pact, as well as efforts to promote excellence and trust in AI.
Together, the planned computing access and renewed stakeholder discussions show how the EU is combining infrastructure support with a push for shared expectations around AI. The details of the startup access process and its impact were still to be worked out.