EU Pledges €10 Billion for Seven AI Gigafactories — What It Really Means

The EU Just Put €10 Billion of Taxpayer Money Into AI Data Centers — Let Me Tell You What That Actually Means

Let me tell you something that's been bouncing around my head since yesterday's announcement out of Brussels. The European Commission formally opened bidding for up to seven AI gigafactories — giant AI data centers backed by €10 billion in public money. And depending on who you ask, this is either the smartest industrial policy Europe has run in a generation, or a slow-moving train that's going to arrive after the station stops mattering.

I've been running hosting infrastructure for over a decade, and I've watched the AI buildout from the cheap seats the whole way through. I've written eleven days straight about hyperscaler capex, credit crises, and the power grid. But this one is different. This is the first time a major government bloc has decided to become a data center operator — with your tax money, not a VC's — and I've got some thoughts.

The Announcement — What Brussels Is Actually Building

Here's the shape of it. On July 30, the European Commission opened a call for proposals for up to seven AI gigafactories across the bloc. The public money: up to €10 billion ($11.5 billion) in combined EU and national funding. The expectation: at least €20 billion more in private investment, for a total potential pot north of €30 billion. Each facility is meant to host at least 100,000 advanced AI processors — roughly four times the scale of Europe's existing data center facilities.

The target users aren't hyperscalers. They're start-ups, scale-ups, SMEs, industrial companies, universities, and public authorities — the groups that can't get GPU capacity because they're at the back of a queue behind people writing nine-figure checks. The facilities are supposed to combine advanced AI processors, cloud and software systems, high-speed connectivity, and energy-efficient data centers. EU tech chief Henna Virkkunen called it "a strategic necessity for Europe."

And it's not starting from zero. This sits on top of the 19 AI Factories already scattered across EU member states, part of the broader AI Continent Action Plan that targets tripling EU data center capacity in five to seven years. The seven gigafactories replace an earlier plan for five, after strong interest from member states. France, Germany, Italy, Spain, the Netherlands, Finland, and Portugal are the co-financiers.

Why Europe Feels It Has No Choice

You have to understand the desperation behind this before you mock the numbers. Because the numbers behind the numbers are brutal.

The Stanford AI Index found that the United States produced more than half of all significant AI models in 2024. Europe produced 6%. Oxford research cited by CEPS puts American and Chinese companies in control of more than 90% of the world's AI-specialized data centers. And on venture capital — the fuel that actually builds this stuff — CEPS says 66% of global AI startup investment between 2023 and mid-2025 went to the US. Europe got 12%.

Europe's share of global data center capacity has fallen from over 25% in 2015 to around 15% in 2024, and its data center market grew at roughly half the worldwide average over that period. Every frontier model Europe runs — every sovereign AI application, every regulated industry deployment — is renting compute from companies headquartered in California, paid in dollars, governed by a legal system Europe doesn't control. When you're a policymaker looking at that, and you're watching the US pour $700 billion-plus into AI capex this year while China builds state compute at scale, sitting on your hands stops being an option.

The Sovereignty Catch — American Chips and the CUDA Blind Spot

But here's the part that should make any infrastructure guy smirk. The Commission has signed letters of intent with AMD, Nvidia, and Qualcomm to supply the chips. Read that again. Europe is spending €10 billion of public money on sovereign AI infrastructure, and the three chip suppliers it signed up are all American companies.

CEPS — the think tank that's been the EU's closest advisor on this — literally calls it the sovereignty blind spot. Every planned gigafactory will run on Nvidia silicon. Training frontier models on Nvidia GPUs requires CUDA, Nvidia's proprietary software stack, which controls what's compatible and what isn't. You can build the building, wire the grid, sign the sovereignty documents — and the most important piece of intellectual property in the room is still owned by a company in Santa Clara. It's like buying the port but not the ships, as one analyst put it. The gigafactories solve where the compute lives and who operates it. They don't touch who designs the chips or who sets the export rules.

That's not nothing — moving compute inside EU jurisdiction matters for GDPR, for AI Act compliance, for CLOUD Act exposure, for data residency requirements that European enterprises actually have to meet. But let's be honest about what it is: sovereignty at the facility level, dependency at the silicon level.

The Real Bottleneck Nobody in Brussels Is Advertising — the Grid

And then there's the part I actually know something about, because it's the same wall I've been watching hyperscalers run into for a year: power. The IEA has been screaming this from the rooftops. Grid connection wait times across the EU run from two to ten years depending on the country. In the FLAP-D hubs — Frankfurt, London, Amsterdam, Paris, Dublin — developers face queues averaging seven to ten years. Dublin and Amsterdam have already paused new projects because the grid simply cannot take them.

European industry pays roughly double the US electricity rate, according to ACER data. Direct grid congestion costs hit €4.3 billion in 2024. And the IEA's modelling says that even though Europe's data center project pipeline equals 130% of installed capacity today, actual installed capacity will grow only about 70% by 2030 — because grid constraints and permitting delays eat the rest. The Commission wants to triple capacity in five to seven years. The grid can't even deliver a doubling by 2030.

So the seven gigafactories — which must be "energy-efficient" by tender requirement — are going to land in a region where the electricity market is the real sovereign. Spain and Finland keep getting named as the emerging hubs, and that's no accident: cheaper power, more headroom, fewer queues. Watch where the private consortia actually choose to build, because that will tell you more than any press release from Brussels.

The Timeline Reality Check — 18 Months Is a Long Time in AI

Now the part that makes me laugh, and not in a good way. Applications close November 12. Award decisions come early 2027. Construction starts during 2027. And selected gigafactories are expected to become operational within 18 months of contract signing. Stack it up, and the first EU gigafactory compute doesn't come online until well into 2028.

In normal infrastructure terms, three years from call to operation is fast. In frontier AI terms, it's geological. Model generations turn over in months. The leading labs are contracting compute years in advance. Meta built tent data centers in Ohio in three months with off-grid gas turbines because it couldn't wait for conventional builds. The EU is running a procurement cycle that would be slow for a highway project, pointed at the fastest-moving technology in human history.

And the early enthusiasm has already cooled once. The bidding was originally scheduled for May, got pushed to July, and early interest reportedly narrowed from around 70 companies to roughly 10 expected bidders. Only two of the original five facilities were projected to qualify for subsidies before 2028. Meanwhile, Mistral — Europe's flagship AI lab — isn't waiting: it announced a $1.2 billion data center investment in Sweden and raised more for a facility near Paris. Nebius is building 310 megawatts in Finland and 240 in France on its own. Even SoftBank's France plans reportedly dwarf the entire gigafactory program. The market isn't waiting for Brussels. It never does.

What This Actually Means for Independent Hosting Providers

Okay. So what do you do with this if you're running a hosting business, a colo operation, or an infrastructure shop with European customers? Here's my take, and I'm not hedging.

First, don't compete with subsidized compute — serve it. The gigafactories are aimed at frontier training workloads. That's not your market. Your market is the latency-sensitive, compliance-heavy, everyday production workload — and every European business that gets into AI via these facilities is going to need a place to run the models in production. That's you. Position yourself as the production layer under Europe's training layer.

Second, watch the secondary markets. Spain and Finland are the names being whispered for gigafactory sites. When public money flows into a region, power contracts, land, and colo space in that region get more expensive and more scarce. If you're thinking about European expansion, the time to lock contracts is before the awards are announced in early 2027 — not after.

Third, lead with compliance. The tender requires everything to meet EU data protection, safety, security, and ethics standards, and the AI Act's transparency obligations start phasing in around now. American hyperscalers are spending billions complying with EU rules under protest. You can make EU compliance your selling point, not your cost center. European enterprises are terrified of both US CLOUD Act exposure and GDPR fines — a provider who speaks fluent EU regulation has an unfair advantage.

Fourth, price for power — because it's about to get more volatile. If European grids are already congested at €4.3 billion in annual congestion costs, and the gigafactories land on top of that load, industrial electricity pricing gets uglier before it gets better. Lock your power contracts, model your pricing against 10-20% energy cost increases, and pass through honestly. The providers who price power correctly will survive the shakeout. The ones who eat it to win deals will be gone by 2028 — right around when the first gigafactory turns on.

The Bottom Line

I'm not here to mock the EU. Public money as a magnet for private capital is the right architecture — €10 billion pulling in €20 billion does more than €10 billion spent alone. And a bloc that writes AI rules while importing 100% of its compute is making a legitimate strategic bet that owning the buildings is better than renting everything. It is. Barely.

But here's the truth bomb. Government money doesn't fix physics. The grid is the constraint, the timeline is the constraint, and the silicon is the constraint. What Brussels has actually done is placed a €30 billion bet that it can buy its way back into the AI game by 2028 — while the market, in the form of Mistral, Nebius, and every private investor with a spreadsheet, is already moving on its own. The gigafactories will probably get built. The question is whether they'll matter when they arrive, or whether they'll be beautiful, compliant, energy-efficient monuments to a race Europe was already losing when the tender opened.

I know which way I'd bet. But that's why I run servers instead of writing policy.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Reporting is based on sources cited in the article.