Europes EUR30B Bet Against Big Tech

The numbers do not line up the way Europe would like them to, and that is the whole story in one sentence.

On July 30, the European Commission opened bidding for up to seven so-called AI gigafactories. The plan is simple on paper: pool 10 billion euros of public money from the EU and its member states, attract at least 20 billion in private investment, and build computing centers that give European startups, universities, and government labs the firepower to train frontier AI models. Eighteen member states have signed on, including Germany and France. Letters of intent are already in place with AMD, Nvidia, and Qualcomm for chip supply. Applications close November 12, 2026. Construction begins 2027.

Now the comparison that makes the whole exercise read like a startup pitch at a Fortune 500 board meeting.

The three largest US tech companies alone plan to spend more than 600 billion dollars on data center infrastructure this year. That is one year, not a multi-year program. Amazon just completed a 50 billion dollar investment in OpenAI, more than the entire public funding side of the EU plan in a single corporate deal. Microsoft, Google, and Amazon are individually outspending the combined public-private AI infrastructure budget of an entire continent of 450 million people.

The EU calls this the AI Continent strategy. The numbers call it something closer to a declaration of independence that no one on the other side of the Atlantic has noticed.

The spending gap in context

To understand why 30 billion euros is simultaneously a large number and a small one, consider the math from the other direction. Amazon plans to spend roughly 100 billion dollars on capital expenditures this year alone, the vast majority of it on AWS data centers and AI compute. Microsoft is on a similar trajectory. Google's capex is approaching 75 billion. When the European Commission says it wants to attract 30 billion euros of private money into AI infrastructure over the full life of the program, that is roughly what one US hyperscaler spends on compute in a quarter.

Alt text

This is not an apples-to-oranges comparison. Both sides are trying to do the same thing: build the physical capacity to train and run large AI models. The difference is that in the United States, that capacity is being built by three private companies that each operate at a scale exceeding most national economies. In Europe, it is being built by a consortium of 27 governments, a handful of chipmakers, and whatever private capital the promise of sovereignty can attract.

Eighteen countries versus three companies

There is a structural irony here that is hard to miss. The EU's AI gigafactory program involves 18 member states coordinating on funding, regulation, and procurement. It has a formal application process, a November deadline, and a multi-year construction timeline. It signed memoranda of understanding with three chip companies to secure supply.

Across the Atlantic, three companies (Amazon, Microsoft, and Google) are building the same thing without any of that apparatus. They do not need government applications or multinational consortiums. They issue earnings guidance, their stock prices rise or fall, and the money appears. When they need chips, they call Nvidia and TSMC directly. They are not hoping to attract private investment; they are the private investment.

The result is that European AI compute capacity depends on a political process, while American AI compute capacity depends on a market process. One requires consensus among 18 governments. The other requires Jeff Bezos, Satya Nadella, and Sundar Pichai to stay bullish on AI. So far, all three seem very bullish.

What EUR30B actually buys

The 30 billion euro figure covers construction, hardware, energy, and operations for up to seven facilities. Each gigafactory would be designed to host thousands of accelerators: the Nvidia H200 and B200 GPUs, AMD Instinct MI series, and whatever Qualcomm brings to the table through the letter of intent.

For a European startup training a competitive foundation model, that capacity is currently either unavailable or priced at a premium that makes it cheaper to rent compute from an American cloud provider. The gigafactories aim to fix that by offering subsidized compute access to European entities. The hope is that lower infrastructure costs let European AI labs keep their intellectual property and training data on European soil, which is the real prize behind the sovereignty framing.

But 30 billion euros of compute capacity is still a rounding error in the global AI infrastructure market. A single 100,000-GPU cluster costs somewhere between 3 and 5 billion dollars to build and run for a year. Seven of those eat most of the budget before you account for power, cooling, staff, or the fact that GPU generations turn over every 18 months. The EU plan is not large enough to match US capacity. It is large enough to keep a few anchor tenants alive.

The uncomfortable question

The EU's AI gigafactories are a serious policy response to a real problem. Without domestic compute infrastructure, European AI startups will remain tenants on American cloud platforms, subject to American export controls, American pricing, and American terms of service. Sovereignty as a concept is not wrong.

The uncomfortable question is whether 30 billion euros is enough to buy independence or just enough to buy the appearance of it. The gap between what the EU is spending and what the market is spending is not narrowing. It is widening, and it will have widened further by the time the first gigafactory breaks ground in 2027.

Europes EUR30B Bet Against Big Tech

The spending gap in context

Eighteen countries versus three companies

What EUR30B actually buys

The uncomfortable question

Sources

Want to learn more?

Related Articles