On July 30, 2026, the European Commission formally opened its call for proposals for AI gigafactories, inviting companies to bid to build as many as seven large AI data centers across Europe with public backing. The numbers attached are deliberately large: up to €10 billion in combined EU and national funding, which the Commission expects to draw in at least €20 billion more in private investment, for a total of more than €30 billion. Each site is meant to house at least 100,000 advanced AI chips, roughly four times the scale of Europe's current data-center facilities. The stated goal is straightforward and geopolitical: give European startups, research institutions and industry access to frontier-scale compute at home, and narrow the gap with the United States and China. This piece walks through the structure of the plan, the timeline, how it compares to what the rest of the world is spending, and the genuine questions the announcement leaves open.
The structure: public money as a magnet
The most important feature of the plan is that the public funding is the smaller half of it. The €10 billion from the EU and member states is not meant to build the gigafactories directly; it is meant to de-risk them enough that private capital supplies the majority of the money.
Public money is the smaller share of a EUR 30 billion target
The EUR 10 billion in public funding is designed to pull in at least twice as much private capital, an industry-led model rather than a state-built one.
The money is spread across up to seven sites, each expected to host at least 100,000 advanced AI chips, roughly four times the scale of current EU data centres.
This is an industry-led model rather than a state-built one, and the distinction matters for how to read the headline figure. The Commission is not promising to construct seven data centers. It is offering to co-fund proposals that private operators bring forward, with the public share acting as a catalyst. That approach has the advantage of leverage: €10 billion that mobilizes €20 billion more does more than €10 billion spent alone. It also carries the risk inherent in any matching-fund scheme, which is that the private capital shows up only where the economics already worked, and stays away from the harder, more strategically motivated builds the public money was meant to enable.
The timeline: years, in a field that moves in months
The schedule the Commission published is orderly and, by the standards of large public infrastructure, reasonably quick. It is also slow relative to the technology it is meant to serve.
A schedule measured in years, not quarters
Even on the stated timeline, the first EU gigafactory compute does not come online until well into 2028, a long horizon in a field where model generations turn over in months.
Call for proposals formally opens.
Application window closes.
Funding decisions, then framework agreements and contracts signed.
Construction begins on selected sites.
Selected gigafactories expected to become operational.
The call closes on November 12, 2026, with funding decisions expected in early 2027, framework agreements signed shortly after, and construction beginning during 2027. Selected gigafactories are then expected to become operational within 18 months of contract signature. Stacked end to end, that puts the first EU gigafactory compute coming online no earlier than well into 2028. In most infrastructure categories, a build that reaches operation inside three years of the initial call would count as fast. In frontier AI, where model generations have been turning over in a matter of months and the leading labs are contracting for capacity years in advance, a 2028 delivery date is a long horizon. The plan is best understood as an attempt to build durable, sovereign capacity for the back half of the decade, not to close the compute gap that exists right now.
The comparison Europe cannot avoid
Any assessment of the gigafactory plan runs into the same wall: the sums involved, while large for a European public program, are modest against what the largest private players are spending. The world's biggest cloud providers are collectively on pace for well over $700 billion in AI-related capital expenditure in 2026, according to industry compilations and company guidance, up sharply from around $410 billion in 2025. Alphabet alone raised its full-year 2026 capital-expenditure guidance to $195 billion to $205 billion. Meta guided to as much as $145 billion, and Microsoft to roughly $190 billion.
Set against those figures, a €30 billion total spread across up to seven sites and several years is a serious commitment to sovereignty rather than an attempt to match hyperscaler scale, and it is worth being clear-eyed about which of those two things it is. Europe is not trying to out-spend the American cloud giants; on these numbers it plainly cannot. It is trying to ensure that European researchers, companies and startups are not entirely dependent on compute controlled by non-European firms, subject to non-European export rules and pricing. Whether €30 billion is enough to achieve even that narrower goal is the plan's central open question.
Europe is not trying to out-spend the American cloud giants. On these numbers it cannot. It is trying not to depend entirely on their compute.
The dependency the plan does not solve
There is a second constraint the announcement does not resolve, and it sits inside the gigafactories themselves. A data center stocked with 100,000 advanced AI chips is only sovereign in the sense of where the building stands and who operates it. The chips that fill it are overwhelmingly designed by a small number of non-European companies, and the most capable of them are subject to export controls set outside Europe. Building the halls does not, by itself, change who makes the silicon inside them or who sets the rules on shipping it.
This is not a reason to dismiss the plan; access to compute capacity is valuable even when the underlying hardware is imported, just as a country can benefit from a busy port without owning the ships. But it does define the ceiling on what sovereign AI means here. The gigafactory program addresses the where and the who-operates of European compute. It leaves the harder questions of chip design and supply chain, which run through Taiwan, the United States and a handful of specialized suppliers, largely untouched. A fuller sovereignty story would need to pair this with progress on domestic chip capability, which is a slower and far more expensive undertaking.
A notable shift in posture
Read against the last few years of European AI policy, the gigafactory call marks a shift in emphasis worth naming. Europe's most visible AI moves have been regulatory, most prominently the AI Act, whose transparency obligations for general-purpose models are themselves phasing in around this period. The dominant external narrative had been that Europe writes rules while the United States and China build capacity. Opening a €30 billion infrastructure initiative is a deliberate counter to that narrative, an attempt to pair the regulatory posture with an industrial one. The plan does not resolve the tension between the two, and there is a live debate about whether Europe's regulatory load and its infrastructure ambitions pull in the same direction. But it does signal that European institutions have concluded compute capacity is now a strategic asset that policy has to actively build, not only govern.
Why portability is the practical hedge
For the European organizations the plan is ultimately meant to serve, the near-term reality is that frontier compute and the strongest models will remain distributed across many providers and jurisdictions for years, whatever gets built by 2028. That makes flexibility the practical form of sovereignty available today. A team that can move its work across models and providers, rather than being locked into one vendor's stack, keeps its options open as capacity and rules shift underneath it. This is the same model-agnostic principle Metir AI is built on: giving teams access to leading AI systems through one workspace without binding them to a single provider, so that whichever compute wins out, in Europe or elsewhere, the people using it are not stranded by a bet on the wrong platform.
The takeaway
The EU's AI gigafactory initiative is a genuine and substantial commitment, and it is also a limited one, and both of those things are true at once. It is substantial in that €30 billion of coordinated public and private investment aimed at sovereign compute is a real change from a policy stance long characterized as regulation-first. It is limited in that the sums are small next to private hyperscaler spending, the first capacity is years away, and the deepest dependency, on non-European chips, is left in place. The most accurate reading is that Europe is buying insurance against total compute dependence rather than a ticket to compute parity. Whether that insurance is adequately sized is a question the bids that arrive by November, and the buildout that follows, will begin to answer.
Sources:
- EU Pledges €10 Billion in Public Funding for New AI Data Centers | Bloomberg
- EU lays out $11.4 billion for 7 AI gigafactories as it aims to catch up with US and China | Washington Post
- EU launches €10bn drive to build AI Gigafactories across Europe | Innovation News Network
- EU pulls trigger on €30B AI gigafactory initiative | Mobile World Live
- EU Opens Bids for Seven AI Super-Hubs To Break US and China Monopoly | IBTimes UK
- Alphabet Q2 2026 earnings takeaways: Google Cloud revenue jumps 82%, stock sinks on capex hike | CNBC
- Hyperscaler CapEx Hits $690B in 2026 | Introl
Image credits
Header image: the Berlaymont building in Brussels, headquarters of the European Commission, photographed by EmDee via Wikimedia Commons, licensed under CC BY-SA 4.0. In-body photograph: the Berlaymont entrance at night by Radek Kucharski via Wikimedia Commons, licensed under CC BY 2.0. The images depict the European Commission headquarters rather than the specific gigafactory sites, none of which have been selected or built.