EU allocates up to €30 billion for AI gigafactories, but US tech giants spend twenty times more

The European Commission wants to build up to seven AI gigafactories across Europe, with a total investment of more than €30 billion in public and private funding. Of that amount, up to €10 billion will come from EU and national government sources; at least €20 billion must come from private parties. Public funding is in principle capped at 35 percent of project costs.

The scale of the initiative is significant by European standards, but the contrast with the American market is stark. Major US tech companies are collectively planning between $600 billion and $725 billion in AI-related capital expenditure in 2026 alone, an increase of roughly $150 billion compared with 2023. What the EU is mobilising in total is approximately twenty times less than what the US sector will spend this year alone.

The programme is being implemented through the European High Performance Computing Joint Undertaking, known as EuroHPC JU, which was established in 2020. Eighteen member states have signed a joint procurement agreement, while Germany, France, Italy, Spain, Poland and the Netherlands have not.

Two project categories with different scales

The programme is divided into two so-called lots, each with different requirements and funding levels. Lot 1 supports up to four projects. Each project receives a maximum of €100 million in EU funding in the first phase and a further €400 million in the second phase. The requirement is that the computing facility must contain at least as many advanced AI processors as the most powerful existing AI factory in Europe, and must triple that capacity in the second phase.

Lot 2 targets larger projects, a maximum of three. For these, up to €200 million per project is available in phase one and up to €800 million in phase two. These projects must offer up to twice the processor capacity of the current European maximum and quadruple that capacity in the second phase.

The gigafactories consist of advanced AI processors, software and cloud technology, high-speed network connections and energy-efficient data centres. They do not necessarily need to be located at a single site: projects may also be set up on a cross-border basis through distributed computing facilities in multiple member states.

Access for startups, SMEs and knowledge institutions

An explicit goal of the programme is to ensure that computing capacity does not end up exclusively with large companies. Startups, scale-ups, SMEs, industry, universities and public authorities are intended to gain access through the gigafactories to infrastructure for training, fine-tuning and running advanced AI models.

EuroHPC JU and the participating member states will jointly procure computing time from the selected factories. Consortia or so-called Special Purpose Vehicles, comprising technology companies, cloud providers, investors and public organisations, can apply as operators.

An earlier non-binding consultation phase attracted 76 responses from consortia across sixteen countries, indicating that market interest exists. The Commission has signed letters of intent with AMD, Nvidia and Qualcomm to secure access to hardware.

Strategic autonomy as the driving principle

Henna Virkkunen, Executive Vice-President for Technological Sovereignty at the European Commission, has repeatedly emphasised the importance of the gigafactories. The broader objective is twofold: to significantly expand available AI computing power in Europe and to reduce dependence on foreign technology.

AI technology developed or deployed through the gigafactories must comply with EU standards on data protection, security and ethics. That framework distinguishes the European programme substantively from comparable initiatives in the US or China, where such requirements are less central.

For the Dutch and broader European startup and investment community, the key question is how accessible the capacity will be in practice. The programme promises open access for smaller players, but the ultimate distribution of computing time will depend on the procurement agreements that EuroHPC JU concludes with the selected operators. The fact that the three chipmakers AMD, Nvidia and Qualcomm have already signed letters of intent provides some assurance on the hardware side. Whether European founders and researchers will also be able to access the capacity affordably and quickly enough remains to be seen as implementation unfolds.