The EU’s sovereign AI plans: high hype, low voltage? Part 1
Europe wants to build its own AI factories – but there are still a lot of unanswered questions.
Build it and they won’t come? The strange geography of European compute
From the UK to India, Singapore to South Korea, every government now wants sovereign AI. It has become the mantra of 2025, as leaders have woken up to three converging realities: the extreme concentration of power in generative AI; AI’s rapid penetration into social media, public services and critical infrastructure; and a worsening geopolitical climate. The conclusion is unavoidable: if AI is the new substrate of the state, you had better control it – or someone else will.
Because the real risk isn’t just falling behind. It’s dependency. What happens when the AI systems running your hospitals, power grids, tax authorities, and business transactions sit on compute you don’t control – and can be switched off, priced up, or politically weaponised by another country?
Sovereignty isn’t just ‘having AI’. It’s being able to operate it on your own soil even if foreign suppliers pull the plug on the hardware-software stack behind modern AI models.
Europe has understood this. Since 2024, the EU has announced 19 ‘AI factories’ across 16 Member States, and in February 2025 it added plans for five gigafactories. The EU’s ambition is to lay the foundations for European tech sovereignty – a core theme of the upcoming Franco-German sovereignty summit in Berlin on 18 November.
But does the factory strategy make sense? We ran the numbers and looked under the bonnet: where they’re being built, whether or not they form a network, and how they fit the EU’s own narrative on AI compute.
First, the ‘where’. You’d expect factories to be placed in regions with strong AI ecosystems – talent, research hubs, and start-up density. Yet, the map of AI factories barely overlaps with Europe’s actual AI excellence hubs. Even current labour-market demand for advanced AI skills shows no spike in those regions. This already clashes with the Commission’s pitch of ‘dynamic ecosystems’ where researchers and entrepreneurs collaborate to “create cutting-edge AI’. Remote compute access makes this less fatal, but it’s still a mismatch compared with what the Commission implied – that factories would host entire research and innovation ecosystems.
Next, the energy. In the US, hyperscalers cluster their factories and data centres where electricity is cheap. Europe doesn’t follow that logic. Only Finland and Sweden offer energy prices competitive with the US or China. Everywhere else, the cost is significantly higher. It gets worse when we consider energy conditions across Europe. The factory sites tend to coincide with the most energy-efficient areas for data centres in the countries they’re in. That’s great but they still don’t benefit from being in regions with the highest share of renewables and the cheapest electricity.
Finally, connectivity and collaboration. Are the factories plugged into each other? Do they already collaborate scientifically or technologically? Do they interface with existing AI excellence hubs? Is there at least high-bandwidth optical fibre linking them? Across the board, the answer is no. The network effects don’t exist – either socially or physically.
So, are Europe’s AI factories just cathedrals in the desert? One defence is the classic ‘build it and they will come’. But what happens if they come – and there’s nothing there to sustain them?
Andrea Renda, CEPS Director of Research