As the EU is struggling to catch up to other continents’ AI buildouts, doubts arise over whether its strategy is even the right one to begin with. 20 billion euros is being set aside for several AI gigafactories with 100,000 GPUs or more each. Legislators and experts are unsure if the effort will be enough, or even be targeted at the right use case. Instead, domain-specific objectives are being highlighted.
The European Commission’s €20 billion plan to build AI gigafactories is drawing widespread criticism ahead of its formal launch this spring, reports Politico. Commission President Ursula von der Leyen outlined the ambition in February last year, proposing four to five mega computing facilities, each powered by 100,000 GPUs, forming Europe’s answer to OpenAI’s 500 billion dollar Stargate project.
Pushing against the tide
The gigafactories build on an earlier pledge to construct 19 AI-focused supercomputers, known as AI Factories, across 16 countries. This effort would be four times larger. A formal call for proposals, delayed twice, is still expected this spring. Nevertheless, it’s already clear AI buildouts in Europe aren’t quite progressing as intended, with OpenAI backing out of its Stargate UK plans.
Critics question whether the demand is actually there, and if the plan even benefits European companies. A total of 76 bids for 60 sites across 16 countries were filed during an informal sense-check last year. But legislators and experts say which companies would use this computing power remains unclear. The Commission’s spokesperson Thomas Regnier pushed back. “This is not just about raw compute power, it’s about sovereign compute,” he said, adding that European industry demands environments where data is “fully protected under European law, without the possibility of third country interference.”
Where’s the demand?
Europe’s most prominent AI developer, Mistral, isn’t waiting for the gigafactories. In February, it announced a 1.2 billion dollar data center investment in Sweden. At the end of March, it raised hundreds of millions of euros for a facility near Paris. That raises the question of who the gigafactories are for. If not Mistral, who else would be looking to commit to large-scale AI training?
Nvidia dependency adds another layer of concern, which is especially the case if AI training is the main purpose of the gigafactories. A group of 18 European Parliament lawmakers warned the Commission about dependency on a single chip supplier, questioning how the gigafactory initiative would reduce Europe’s strategic dependencies. Regnier declined to address those concerns directly.
The problem is that just as these plans have been drawn up, a revolution is taking place in the AI compute space. With training losing its importance as time moves on, inferencing, the daily running of AI workloads, is gaining more traction. Microsoft, Google Cloud, AWS, Intel and many more IT players are realizing a disaggregated AI infrastructure will define the future. Examples include different chips for training versus inferencing, separating individual components of AI workloads themselves and offloading to CPUs when latency isn’t critical.
Just as the EU is pushing to catch up in yet another technological area, having previously focused on legislating it, the present is rapidly outpacing past plans. We’re curious if the proposed AI factories will receive a rhetorical shift just as the largely failed EU Chips Act did. The latter political move initially revolved around securing chip supplies in constrained times to avoid a reliance on Asia and North America, but it was later used to entice non-EU vendors like Intel and TSMC to build fabs in Europe.
A global funding gap
The 20 billion euros, at any rate, looks modest compared to competitors. OpenAI launched a 500 billion dollarcompute plan last year, and Anthropic announced 50 billion dollars in U.S. infrastructure investments. On top of that, the hyperscalers are each setting aside around 180 billion dollars for all of 2026 to realize the AI infrastructure growth they perceive to be necessary. As we previously reported, the gigafactory financing model requires 65 percent from private parties, and that structure is already causing delays in some member states.
Siemens recently warned that EU regulations are pushing its AI spending outside Europe, and it’s far from the first to do so. Meanwhile, as has become clear this year once again, European data center demand has outpaced supply for five consecutive years.