Beyond what, a report by CEPS investigates where, when and how AI factories should be built, calling for appropriate stewardship and clear investment choices.
Europe is accelerating its investment in advanced computing infrastructure, betting on a network of AI factories and gigafactories to strengthen competitiveness and reduce dependencies in a rapidly shifting geopolitical landscape. The ambition is high: combine computing power, data, and talent to build a new layer of AI-driven industrial capacity. Yet, as the latest report of the Centre for European Policy Studies (CEPS) highlights, Europe’s path towards this goal is neither straightforward nor guaranteed.
Over the past two years, the European Commission has moved from incremental upgrades of the European High Performance Computing network (EuroHPC) to a far more proactive agenda. The AI Continent Action Plan, Apply AI, and AI in Science strategies, together with plans for up to five gigafactories equipped with at least 100’000 advanced AI chips each (approximately four times the size of the AI Factories), signal a strong political commitment to sovereign AI capabilities. Meanwhile, private actors – most prominently Nvidia according its recent announcements – are rapidly advancing their own (giga-)factory deployments on European soil, further reshaping the continent’s computing landscape and eventually creating competition with the EU-supported AI Factories network.
But ambition alone does not guarantee competitiveness. A central finding of the CEPS report is that most AI factories are not located in Europe’s leading AI hubs. Instead, they tend to be placed where energy efficiency and existing HPC infrastructure are favourable rather than where Europe’s scientific and entrepreneurial activity in AI is strongest. This is the case in Switzerland, where the ‘Alps’ supercomputer is located in Lugano (Ticino), whereas the excellent AI Hubs are in Zurich and Lake Leman region. Still, this is mitigated by the fact that, beyond Switzerland being a small country, Alps can be accessed remotely and is operated by ETH Zurich, thus closely connected to an AI Hub. While this reflects rational constraints around energy, land, and cooling, it raises questions for the other AI Factories about their ability to anchor vibrant innovation ecosystems.
A second challenge relates to specialisation. Factories have been assigned priority sectors ranging from health to manufacturing to agriculture. Yet in several regions, these thematic designations do not fully align with local strengths or AI penetration levels. Some facilities appear well-positioned, such as those in Spain or Finland, where strategic importance for the region is correlated with investments in targeted areas, while others risk diffusing resources across domains where regional capacity is thin. The result is a patchwork of strategies that may not translate into a cumulative European advantage.
Collaboration is another critical gap. Patent co-authorship data reveal weak links between factory regions, especially compared with Europe’s established AI hubs. Without stronger cross-border connectivity, these facilities risk becoming well-equipped but isolated nodes, with limited spill-overs into Europe’s wider R&I landscape.
Finally, Europe faces a structural dependency dilemma. The gigafactory model, as currently envisioned, rests heavily on Nvidia’s vertically integrated ecosystem – an architecture spanning hardware, software, networking, and even factory design. While this brings short-term performance gains, it risks reinforcing precisely the dependencies Europe seeks to reduce. As the CEPS analysis argues, Europe may have little choice in the immediate term but to “Buy American” to meet urgent compute needs. Yet this must be matched with a determined effort to “Build European” by diversifying suppliers, advancing alternative AI architectures beyond GPUs, strengthening European chip design capabilities, and linking factory development with research tracks on trustworthy and resource-efficient AI. At the same time, the broad geographical dispersion of factories across Member States, partly the result of political balancing, risks diluting critical mass. Without greater strategic concentration, Europe’s distributed infrastructure may struggle to evolve into a coherent and sovereign AI ecosystem.
Europe’s AI factories are, without doubt, a bold step. But their success will depend on more than hardware deployment. Truly moving from “Buy American” to “Build European” requires stronger inter-regional collaboration, more mobile and competitive talent pools, and governance models that support long-term technology sovereignty. Ultimately, as the authors caution in their report’s title, Europe must decide whether these facilities become sanctuaries of innovation or cathedrals in the desert.