EU Opens €30 Billion AI Megafactory Tender to Close the Gap on US and China

EU's €30 Billion AI Megafactory Plan: What Was Announced

The European Commission has launched a tender for seven artificial intelligence megafactories — large data centres that combine supercomputers with the infrastructure needed to train advanced AI models. The projects are worth €30 billion in total: the EU budget will contribute €5 billion, host member states will match that with €5 billion, and the private sector is expected to cover roughly two-thirds of the cost.

The plan is an expanded version of an initiative Commission President Ursula von der Leyen first presented at the Paris AI summit earlier this year, when it covered five centres and €20 billion. After strong interest from member states, it has grown to seven projects. Four smaller centres will house between 25,000 and 75,000 specialist AI chips each, while three larger ones will be equipped with 40,000 to 100,000 chips.

Winners are expected to be announced early in 2027 and will have 18 months to complete construction, so the first megafactories could begin operating around mid-2028. Germany, Italy, Spain, Portugal and Greece are candidates to host the largest facilities; France, Poland, Finland, Denmark and the Czech Republic are competing for the smaller ones. Croatia, Estonia, Hungary, Ireland, Latvia, Lithuania, Slovakia and Sweden are so far involved only in consortia led by other countries and are not named as hosts. The Commission says projects with the strongest energy-efficiency and sustainability standards will be favoured.

The initiative is a response to Europe's weak position in frontier AI. Stanford University data cited in the report puts the US share of advanced AI model development at roughly 60 percent, China's at about 30 percent and Europe's at less than 5 percent. Brussels regards the existing network of 19 EU co-funded AI centres, in place since 2022, as too small to support a competitive European AI ecosystem.

Why Brussels Needs Seven AI Megafactories — and Still Buys US Chips

What the €30 Billion Structure Really Means

The headline number is less important than how the money is split. Only €10 billion — €5 billion from the EU budget and €5 billion from host states — is committed public money. The remaining €20 billion depends on private investors. The tender is therefore also a test of investor appetite for European AI infrastructure. If private capital does not arrive in the expected volumes, the programme will be smaller in practice than its political framing suggests. This is an interpretation of the funding structure; the underlying figures are confirmed in the report.

Europe's Compute Gap and Its Continued Dependence on US Chips

The rationale is straightforward: European developers cannot train frontier models competitively without domestic compute capacity, and the existing 19-centre network is too small to meet demand that the Commission already describes as outstripping supply. But the first phase will not close Europe's strategic technology gap. Brussels admits Europe has no domestic production of the most advanced AI processors, and the megafactories will initially run on chips from Nvidia and AMD. MEPs and analysts argue this could deepen dependence on US technology. Both readings can be true at once: Europe gains its own compute infrastructure while its silicon supply chain stays American for years. The Commission says European chip solutions will be phased in later, which is a goal rather than a current capability.

Who Gains and Who Loses in the Hosting Race

The immediate winners are the candidate host states. Germany, Italy, Spain, Portugal and Greece are in the running for the three large centres; France, Poland, Finland, Denmark and the Czech Republic for the four smaller ones. They get anchor infrastructure, construction work and a stronger base for attracting AI investment and talent. Nvidia and AMD benefit commercially whatever the outcome, because the first-phase facilities will be built around their processors. States that appear only in consortia — including Croatia, Ireland, Hungary and the Baltic countries — miss the direct economic pull of hosting. European AI startups gain future capacity, but not until the new centres operate, which will not happen before roughly mid-2028.

The Three Risks That Could Slow the Schedule

The first risk is budgetary. €4 billion of the Commission's €5 billion is earmarked for the next multiannual EU budget, which member states are still negotiating. The Commission says it has alternative scenarios, but the money is not yet guaranteed. The second risk is energy and water. AI megafactories will consume large amounts of electricity and cooling water, and the criteria for awarding the tender are still being shaped. Experts are demanding requirements for new low-carbon power, a ban on diesel generators and wider use of water-cooling systems — conditions that, if adopted, will raise the cost and complexity of bids. The third risk is delivery. An 18-month construction period after an early-2027 award is demanding for projects of this scale, and construction slippage is the most common cause of delay in large data-centre programmes.

Who Should Act on the EU AI Megafactory Tender — and When

The next critical phase runs from now until the awards are announced in early 2027. The concrete moves depend on who is acting:

• Bidding member states and consortia: With the smaller centres sized at 25,000–75,000 AI chips and the larger ones at 40,000–100,000, bids should be locking in electricity supply, water-cooling capacity and site permits now. The Commission has already said sustainability standards will be a selection priority, and expert proposals for binding low-carbon power requirements could become award conditions.
• Private investors and infrastructure funds: The plan assumes about €20 billion of private money in a €30 billion programme, so the funding gap is real. Before committing, investors should price in the unresolved €4 billion EU budget contribution; the Commission has alternatives, but no final budget decision has been made.
• European AI companies and research labs: New capacity will not be online before roughly mid-2028. Until then, European demand for compute already exceeds supply, so near-term training plans should still be built around existing national centres and cloud capacity rather than the megafactories.
• Energy and cooling suppliers: Selection criteria will reward efficiency, and expert groups are pressing for new low-carbon sources, no diesel generators and greater use of water cooling. Suppliers that can document such solutions are better positioned for contracts at whichever sites win.