EU Selects EUROPA Consortium to Build 400-Billion Parameter Open-Source AI Model

The European Commission has officially selected the EUROPA consortium, spearheaded by the Italian technology firm Domyn, to build a massive, open-source artificial intelligence model. The announcement marks the culmination of the EU's Frontier AI Grand Challenge, a flagship competition launched earlier this year to foster homegrown digital infrastructure. By backing a sovereign AI initiative, European leaders are signaling a decisive shift from merely regulating foreign technology to actively funding and developing their own.[1][6]

The EUROPA project is mandated to develop a foundational model with more than 400 billion parameters. This immense scale places the planned system in the same weight class as the most advanced proprietary models currently deployed by tech giants in the United States and China. However, unlike existing frontier models that are built primarily around English fluency and treat other languages as secondary layers, the EUROPA model will be structurally different.[2][4]

From its inception, the system will be natively trained to comprehend and generate text across all 24 official languages of the European Union. This multilingual foundation is designed to directly serve European public institutions, healthcare systems, and researchers who operate in their native tongues and currently lack a comparable open-source option. The approach ensures that the nuances of Europe's linguistic diversity are baked into the model's core architecture rather than translated as an afterthought.[1][6]

Henna Virkkunen, the European Commission's Executive Vice-President for Tech Sovereignty, Security and Democracy, framed the victory as a turning point for the continent. "Europe can lead in advanced AI on its own terms," Virkkunen stated following the announcement. She emphasized that the project demonstrates Europe's ability to match the world's best capabilities while staying true to its values of openness, trust, and strategic autonomy.[1][5]

The consortium taking on this monumental task is led by Domyn, an Italian scale-up formerly known as iGenius. Founded and led by CEO Uljan Sharka, the company has a proven track record in sovereign AI development, having previously released the open-source 'Italia' model and partnered with NVIDIA to build the Colosseum supercomputer. Domyn's expertise in highly regulated sectors made it a natural leader for an alliance that also includes the prestigious German research organization Fraunhofer-Gesellschaft.[3]

As the winner of the Grand Challenge, the EUROPA consortium will not just receive funding; it gains access to the raw computational power necessary to train a frontier model. The group has been awarded up to 2.5 percent of the total supercomputing capacity of the European High Performance Computing Joint Undertaking (EuroHPC) for an entire year. This access bridges the critical gap between theoretical AI design and the physical infrastructure required to execute a massive training run.[2][3]

To maximize this computational windfall, the consortium plans to utilize efficient, modular architectures, such as the Mixture-of-Experts (MoE) approach. This technique allows the model to activate only the specific neural pathways needed for a given query, achieving state-of-the-art performance without the unnecessary energy waste associated with running dense models. It is a technical choice that aligns closely with the EU's broader sustainability and climate goals.[4]

The decision to make the EUROPA model entirely open-source is perhaps its most consequential feature. By releasing the model weights publicly, the European Commission is treating foundational AI as a common good rather than a proprietary product. This open-access strategy is intended to democratize advanced technology, allowing local startups, academic researchers, and small-to-medium enterprises to build customized applications without paying licensing fees to foreign corporations.[3][5]

The geopolitical timing of the announcement is stark. It arrives during a period when several leading US-based AI laboratories are facing internal restructuring, tightening security protocols, and stepping back from previous open-source commitments due to national security concerns. By contrast, the EU is doubling down on transparency, betting that an open ecosystem will accelerate innovation across its member states.[5]

Industry analysts note that while winning the Grand Challenge is a significant milestone, it represents a commitment rather than a finished product. The gap between announcing a 400-billion parameter model and delivering a reliable, aligned system that researchers can actually use is vast. The consortium will need to demonstrate immense engineering discipline to manage the complexities of a supercomputer-scale training run.[2]

If successful, the EUROPA project will fundamentally alter the global artificial intelligence landscape. It promises to prove that a publicly backed, multi-national coalition can build digital infrastructure that rivals Silicon Valley's finest, ensuring that Europe's digital future is written in its own languages and governed by its own values.[1][4]