The European Commission's roughly €20 billion proposal to build several very large artificial intelligence data centres — so-called gigafactories — has come under increasing criticism. The idea, first outlined by European Commission President Ursula von der Leyen in February last year, envisages four to five megacentres each equipped with about 100,000 graphics processing units (GPUs) to train AI models.
The Commission frames the initiative as a matter of digital sovereignty. Thomas Regnier, a spokesperson for the European Commission, says the aim is to avoid Europe's dependence on infrastructure located on other continents. The plan is also presented as a way to compete with major global projects such as OpenAI's Stargate data centre in Norway.
Questions about demand and the intended users
Lawmakers and experts doubt there is sufficient demand to make use of such scale. Sergey Lagodinsky, a German Green Group MEP, told a Brussels event: “Nobody could explain to me what the business plan for these gigafactories is.” He added that while people can say “more compute capacity is needed in Europe,” they struggle to specify exactly for what.
Nicoleta Kyosovska of the Centre for European Policy Studies, co-author of a paper titled “Temples of innovation or cathedrals in the desert?”, warned that it is far from clear who the gigafactories’ customers would be. She noted that in Europe only the French company Mistral appears able to develop large language models that would demand this level of compute — and that Mistral is building its own infrastructure rather than relying on shared megacentres.
Private investments and the Commission's timetable
Mistral has separately announced significant private investments: in February it announced a €1.2 billion data centre investment in Sweden, and in March it raised $830 million to build a facility near Paris equipped with close to 14,000 GPUs. These moves show some European actors are pursuing their own infrastructure paths.
The Commission’s formal tender for the project has already been postponed twice; the call is currently expected in the spring.
Dependency on US technology and the question of scale
Critics also warn the gigafactories could paradoxically increase Europe’s reliance on US technology, because the GPU market is effectively dominated by US firm Nvidia. Eighteen Members of the European Parliament have written to the Commission asking how the project will reduce Europe’s strategic dependence if the data-centre hardware segment is “dominated by a single supplier.” The Commission did not directly answer that query, but reiterated that creating a sovereign environment under European law is valuable in itself.
Observers also question the EU’s proposed scale. The €20 billion envelope is small relative to private sector plans: OpenAI launched a $500 billion capacity-expansion plan last year, and Anthropic announced $50 billion in infrastructure investment. Jeff Campbell, Cisco’s director of government strategy, said that compared with investments elsewhere in the world, the scales are entirely different and “it would not hurt to think at least double.”
Alternative approach: build on industrial strengths
Some, including Bulgarian MEP Eva Maydell, argue that the EU should not try to match US investments in large language models but instead leverage its existing industrial strengths. Maydell noted that beyond large language models there is “a whole world” where European industry — for example in battery technology — can gain a real advantage.
Conclusion
While the Commission’s gigacentres are presented as a route to strengthen digital sovereignty, lawmakers and experts question whether there is clear demand, who would use the capacity, and whether the plan reduces strategic dependency. Delays to the tender, the dominance of Nvidia in GPUs, large private-sector investments elsewhere, and the relative modesty of the €20 billion budget in global terms all raise doubts about whether this approach will effectively boost Europe’s competitiveness.
