Europe Opened a €10 Billion AI Gigafactory Tender. Good Luck With Electricity.

Europe has looked at the American AI race, looked at China’s data-center buildout, and responded in the most European way possible: by opening a tender.

On July 30, the European Union launched a call for up to seven AI gigafactories, backed by as much as €10 billion in EU and national funding. Brussels hopes that public money will pull in at least another €20 billion from private investors, because nothing says “strategic autonomy” like asking venture capital to co-sign the industrial policy.

The European Commission’s announcement describes a serious infrastructure program: giant facilities combining advanced AI processors, software, cloud systems, high-speed networking, and energy-efficient data centers. The goal is to give European startups, universities, manufacturers, public agencies, and scale-ups access to enough compute to train, fine-tune, and run advanced models without sending every sensitive workload through an American hyperscaler.

That is a real problem. It is also an unusually expensive way to discover that electricity remains physical.

Europe Wants Its Own AI Plumbing

“Gigafactory” is doing a lot of emotional work here. This is not seven robot factories stamping out cheerful household androids. An AI gigafactory is a huge computing complex: processors, storage, cooling, networking, secure access, and the software layer that turns a warehouse full of chips into something researchers can actually use.

The Commission says each project will be built through consortia or special-purpose vehicles that can include companies, governments, investors, and other partners. Eighteen member states have signed the joint procurement agreement with EuroHPC, the EU body that coordinates high-performance computing. The selected sites are expected to be capable of handling the complete lifecycle of frontier AI work, from training to inference.

In plain English: Europe wants enough local computing that a startup can train a serious model, a hospital can run sensitive inference, and an industrial company can experiment with AI without first booking time on the global cloud equivalent of a hotel where the minibar charges per token.

The infrastructure argument is strong. Europe’s current AI capacity is spread across 19 AI data centers, and the new gigafactories are intended to more than double the bloc’s available compute. The facilities are also meant to create predictable demand for European semiconductors and give local hardware companies a fighting chance to sell into a market larger than a university lab.

That is why this story matters. It is not a chatbot launch. It is an attempt to move the strategic center of AI from the model demo to the boring physical layer underneath it. As Broadcom, Apollo, and Blackstone turning compute into a finance product made clear, the industry has already reached the stage where AI infrastructure is a financial instrument with a cooling system attached.

The Tender Is Specific, Which Is Refreshing

The Commission’s plan has more detail than the average “strategic partnership” press release. The procurement supports up to seven projects across two lots and two development phases.

• The first lot supports up to four projects, with each eligible for up to €100 million in the first phase and up to €400 million more in the second.
• The second lot supports up to three projects, with up to €200 million available initially and up to €800 million more later.
• Each site must deploy at least as many advanced AI processors as Europe’s most powerful current AI factory in the first phase, then scale to multiples of that capacity.
• The public side will buy compute access from the selected facilities, rather than merely tossing grant money over a wall and hoping a business model appears.

The tender closes November 12, 2026. Award decisions are expected in early 2027, construction is supposed to begin that year, and the selected facilities are expected to start operating within a maximum of 18 months after the contracts are signed.

Sovereignty Is Useful Until the Power Company Calls

The EU’s strategic case is easy to understand. American companies dominate cloud infrastructure, and the largest frontier models are trained on American or Chinese compute stacks. European businesses and public authorities are already dependent on foreign providers for critical workloads. That dependence creates obvious concerns around data access, service continuity, export controls, pricing, and the unpleasant possibility that geopolitics can turn your API endpoint into a diplomatic bargaining chip.

The Associated Press reported the same-day announcement with the blunt arithmetic: €10 billion in public funding, a hoped-for €20 billion in private investment, and facilities planned around at least 100,000 cutting-edge AI chips. The Commission says the sites will follow European rules on data protection, safety, security, and ethics.

All of that is sensible. It is also not magic sovereignty dust.

Europe can own the building and still depend on American chip designers, Taiwanese foundries, American software stacks, foreign cloud tooling, imported transformers, imported memory, and an international supply chain that does not care how sternly the procurement documents say “strategic autonomy.” The Commission says projects may source hardware from Europe or like-minded countries, and it has signed letters of intent with AMD, Nvidia, and Qualcomm. That is practical. It is not the same as producing the whole stack at home.

Then there is electricity. AI data centers need land, grid connections, cooling water, transmission upgrades, and the sort of power contracts that make local politicians suddenly very interested in economic development. European electricity prices are already higher than those in the United States and China in many markets. A gigafactory that cannot get reliable power is a very large monument to the concept of compute.

We have already watched Meta turn rural Louisiana into a $50 billion AI extension cord, complete with the usual collision between national ambition and local infrastructure. Europe is about to run the same experiment across multiple countries, languages, grid operators, planning regimes, and municipal arguments about who gets the substations.

The Business Case Has Entered the Chat

The most interesting question is not whether Europe needs more compute. It does. The question is whether seven facilities can become useful businesses instead of state-sponsored capacity waiting for a flagship customer.

Industry skepticism is not new. In 2025, Science|Business reported concerns about who would actually use the gigafactories, who would maintain them, and whether European companies had enough demand or capital to train models at the scale these facilities imply. Those questions become more urgent when the public commitment reaches €10 billion and the hoped-for private match doubles it.

There are several plausible customers: Mistral and other European model companies, pharmaceutical and automotive firms, public-sector research, defense, climate modeling, robotics, and thousands of smaller companies that currently cannot afford frontier-scale experiments. The facilities could also support inference, which is the part of AI where models answer users and generate bills rather than merely appearing in benchmark charts.

But demand is not the same as ambition. A startup may want access to 100,000 processors in the same way a teenager wants a private jet: abstractly, enthusiastically, and without a finalized operating budget. The public procurement model helps by guaranteeing some usage, but the facilities still need to fill capacity, maintain hardware, manage security, and keep prices competitive with clouds that already know how to do this at scale.

The European answer is that the gigafactories are strategic infrastructure, not ordinary commercial data centers. Fine. Strategic infrastructure can be worthwhile even when it is not immediately profitable. But it still needs a use case, or Europe will have achieved sovereign access to a very expensive empty room.

Seven Megaprojects and a Model to Be Named Later

Europe is not trying to win by cloning the American hyperscaler playbook exactly. The pitch is more cooperative: shared access for startups, researchers, industry, and public bodies; infrastructure aligned with European rules; and a chance for local companies to build on top of the compute instead of renting the entire future from abroad.

That is a credible niche. Europe has deep industrial companies, research institutions, engineering talent, and a long list of sectors where AI needs to operate under real constraints. It does not need to produce the world’s most emotionally fluent chatbot to make good use of serious compute. It needs models that help design batteries, optimize factories, discover materials, interpret medical data, and run public services without treating privacy law as a suggestion.

The risk is that “European AI” becomes a branding layer over imported hardware and a few government-friendly demos. We have seen how quickly sovereignty language can turn into a logo, a summit, and a slide showing a robot standing in a wheat field. The plumbing is the point. If Europe funds capacity but does not build the software, chips, talent pipelines, procurement habits, and commercial customers around it, the gigafactories will be strategically independent in the same way a hotel is independent when it imports the furniture, the food, and the guests.

There is a useful precedent in the local-AI movement: Ollama turning a laptop into a small AI data center showed why control over where inference happens can matter. Europe is now attempting the industrial-scale version, with fewer laptops and significantly more permitting.

Verdict: A Real Bet, Not Yet a Real Factory

Europe’s AI gigafactory tender is a meaningful shift, not empty theater. It acknowledges that AI power is physical, strategic, and increasingly infrastructural. It gives smaller companies and public institutions a path to compute access. It may also create the demand needed for European AI hardware and industrial applications to develop beyond the prototype stage.

It is still a risky bet. Seven facilities will not erase the gap with the United States or China, especially if they arrive late, cost more to run, or struggle to find customers. The €10 billion is a beginning, not a finish line. The €20 billion private match is a forecast, not money in the bank. And the 18-month operating target is the kind of sentence that looks brave until a transformer, permit, chip shipment, or budget committee says otherwise.

So I am cautiously impressed. Europe is finally arguing about the thing underneath the AI argument: who owns the compute, who gets access, who pays for the power, and what happens when the models become part of the industrial base.

That is progress. Just remember that a tender is not a data center, a data center is not a business, and a business is not automatically a reason to train another model that writes the word “revolution” 400 times per hour.