€30bn AI computing labs: compute policy analysis

Compute policy rarely arrives with a clean interface. It arrives as a tender, an eligibility rule, a deadline calendar, and several committees deciding who gets near the machines. Research Professional News reports that the EU has opened a call to set up €30bn AI computing labs. Quartz reports that the tender covers up to seven AI gigafactories, with the European High Performance Computing Joint Undertaking leading an effort that would combine up to €10bn in public funding with at least €20bn in expected private investment. The useful question for builders is not whether Europe has discovered AI. It is whether public compute becomes a practical route to building serious systems without depending entirely on foreign AI infrastructure.

What Brussels is actually buying

According to Quartz, the EU launched a call for tenders to establish up to seven AI gigafactories across Europe, with construction set to begin in 2027. Eunews also reported that the call concerns seven gigafactories worth €30bn. That makes the headline look like funding, but the instrument is procurement for capacity. In practice, this means the access rules, permitted users, and application windows will matter as much as the total amount. EuroHPC JU’s existing AI for Science and Collaborative EU Projects Access call shows how this kind of capacity becomes usable, or does not. EuroHPC lists that call as open, with a publication date of 11 April 2025 and multiple deadline dates across 2026. The deadlines it names are 27 February 2026, 30 April 2026, 30 June 2026, 31 August 2026, 30 October 2026, and 11 December 2026, all at 10:00 CET. That is not a slogan. It is an operating calendar. The lesson is administrative, which is why it is easy to underrate. If a lab exists but your team misses the right access mode, consortium status, or deadline, the compute might as well be on the moon. Public infrastructure does not remove friction. It changes where the friction sits.

Who gets near the machines first EuroHPC

JU says its AI for Science and Collaborative EU Projects Access mode supports AI applications for science, with a focus on ethical artificial intelligence, machine learning, foundation models, generative AI, and large language models when AI models are part of the research workflow. It says the access mode covers scientific users, whether funded or not by national or European programmes, public sector users, and industrial users participating in research and innovation projects funded by EU programmes such as Horizon Europe or the Digital Europe Programme. EuroHPC also states that all other types of industrial users should target the AI for Industrial Innovation Access modes. Translated into builder language, this is the first scope test. Are you a scientific user, a public sector user, or an industrial user inside a funded research and innovation project? If not, your route is probably a different access mode, not a strongly worded email about European competitiveness. The call does not, on the evidence available, mean every startup receives frontier compute on request.

The policy lever beneath the funding

Quartz reports that the EU wants the gigafactory effort to expand the bloc’s computing capacity and reduce reliance on foreign AI infrastructure. That is the compute policy story. When public money is used to attract private investment into shared infrastructure, the policy question shifts from who can write the biggest cloud invoice to who qualifies for access, on what terms, and for which workloads. This is where procurement becomes market structure. A €10bn public contribution designed to attract at least €20bn in private investment, as Quartz reports, is not merely a cheque. It is a way to shape where capacity is built and who has a credible path to use it. For European researchers, public sector teams, and companies tied to EU funded research and innovation projects, that could make compute planning part of grant and product strategy rather than a late budget crisis.

What builders should do before 2027

Quartz reports that construction is set to begin in 2027, while EuroHPC’s existing access call already has deadline dates through 2026. That timing matters. The gigafactories are a future capacity story, but the habits needed to use public compute are current compliance work: map the workload, identify the access mode, check whether the project sits inside an eligible programme, and calendar the cutoffs. For product teams, the immediate task is boring and valuable. Keep a short file that explains what model work needs external compute, what data would be involved, whether the work is scientific, public sector, industrial research, or commercial, and which EU programme link, if any, supports it. For investors and boards, the diligence question is no longer only whether a company has model talent. It is also whether it has a realistic compute access plan. The next things to watch are the final tender terms, the access conditions for the new facilities, and whether the promised public private funding mix turns into capacity that applicants can actually schedule. Builders do not need to treat Brussels procurement as destiny. They do need to treat it as infrastructure policy with deadlines, and deadlines have a rude habit of enforcing themselves.