Summary
- The first phase of the EU’s AI gigafactory tender is reportedly smaller than the 100,000-chip scale attached to the original policy ambition.
- A staged procurement would bring power, subsidy structure, private finance, and construction deliverability into sharper focus.
- The Commission still expects a formal call in summer 2026, with first construction targeted for 2027.
The European Commission is moving towards a smaller first phase for its AI gigafactory procurement, after earlier plans for very large European AI data centres ran into the hard limits of finance, chips, power, and buildability.
Bloomberg Law reported that the first phase is expected to seek four data centres with at least 25,000 GPUs and three with at least 40,000 processors, citing people familiar with the matter. The report said private-sector finance would carry most of the cost, with EU and national subsidies potentially covering up to one-third.
That scale is below the 100,000 advanced-processor figure associated with the original AI gigafactory concept. On the Commission’s own AI gigafactories page, the policy remains framed around mobilising €20bn for several private-led facilities across the EU, with a call for tender expected in summer 2026 and construction of the first AI gigafactory targeted for 2027.
A smaller opening phase may make the programme more deliverable. Very large AI training campuses sit across several constrained markets at once: advanced processors, power grid access, electrical equipment, cooling plant, specialist construction, sovereign cloud demand, and public subsidy tolerance. A staged tender would allow Brussels and member states to test where private capital is prepared to take development risk before committing the programme to a larger build-out.
The Commission has described large-scale computing infrastructure as a bottleneck for Europe’s competitiveness and strategic autonomy. Its gigafactory programme sits alongside AI factories and antenna sites intended to give startups, researchers, industry, and public bodies access to EuroHPC-backed compute. The gigafactory layer is more industrial: large facilities capable of training and running frontier models, with heavy demands on power, networking, energy efficiency, automation, security, and supply chains.
Policy meets the power queue
The reported shift moves the physical data centre question deeper inside Europe’s AI policy. Processor counts alone will not decide the programme. Each site will need land, grid connection, energy procurement, backup power, transmission reinforcement where required, high-density cooling design, security, operations, and long-term commercial utilisation.
Earlier market interest was substantial. The Commission has said an informal expression-of-interest process produced 77 proposals across 16 member states and 60 sites. That response showed appetite from technology companies, operators, telecoms groups, power suppliers, and financial investors, but it did not prove that projects could be financed, permitted, energised, and delivered at the implied scale.
A 100,000-chip facility is not just a data centre. It is a power project, a semiconductor procurement exercise, a supply-chain commitment, and a sovereign infrastructure wager. Public money can de-risk some of that, but it cannot create grid capacity, shorten every permitting path, or guarantee enough long-term demand to keep the asset fully used.
By opening with smaller facilities, the EU may also avoid a procurement that only a small number of consortia could credibly answer. Four 25,000-GPU sites and three 40,000-processor sites would still represent a major infrastructure commitment if they proceed, while giving the Commission and EuroHPC more deployment models to compare across different power systems, host countries, construction routes, and governance structures.
The sovereignty test moves on site
Europe wants more control over the computing layer behind frontier AI, scientific modelling, industrial automation, public-sector applications, and defence-adjacent research. The infrastructure question is whether that control is best built through a few huge sites, a more distributed network of large facilities, or a hybrid model combining EuroHPC systems, commercial neocloud capacity, and private hyperscale investment.
European demand for AI compute is growing while data centre development faces tougher scrutiny over land, electricity, water, heat rejection, and emissions reporting. Gigafactory sponsors will need to show that they can secure processors and place them inside sites with credible power paths and resilient operating models.
The strongest bids are likely to be those that can demonstrate available or near-term power, a realistic build programme, supply-chain control, bankable customers, and a subsidy structure that does not leave the public sector underwriting speculative capacity. A trimmed first phase would not remove the ambition. It would test whether Europe can turn that ambition into energised, cooled, insured, and commercially used infrastructure.
