The E.U.’s AI Drive Undermines Its ​Own Chip Strategy

Europe is in the middle of a massive identity crisis. They want to be the world’s AI hub, but they’re realizing that building the “factories of the future” requires buying tools from everyone but themselves. As a founder, I see this all the time: a grand vision that hasn't accounted for the basic supply chain reality. The European Union is currently pushing a massive expansion of AI infrastructure, yet they only produce about 10% of the world’s semiconductors. It’s a math problem that doesn’t add up.

The Sovereignty Paradox

The E.U. recently realized that their original Chips Act from 2023 was a bit of a dud. It focused on supply—trying to get companies to build plants in Europe—without making sure anyone actually wanted to buy what those plants produced. Now, we’re looking at Chips Act 2.0. The goal is to stimulate demand, using public money and “AI factories” to create a market for chips. The problem? To build those AI factories today, you have to buy from Nvidia.

This is what I call the demand trap. If you build nineteen AI factories and seven gigafactories, you need hundreds of thousands of high-end GPUs. Europe doesn’t make them. So, in an effort to become “technologically sovereign,” the E.U. is about to write a massive check to Silicon Valley. It’s a loop that reinforces the very dependency they say they want to escape.

The Nvidia Dependency Trap

Let’s look at the numbers for a second, because they’re staggering. A single AI factory needs about 25,000 advanced chips. A gigafactory needs 100,000. Right now, almost all of that demand is being funneled toward Nvidia. It’s not just the hardware, either. It’s the software layer—CUDA. When you build your entire infrastructure on a proprietary stack owned by a single U.S. company, you aren’t sovereign. You’re a high-paying tenant.

• Mistral, the French AI darling, is using nearly 14,000 Nvidia GPUs.
• Deutsche Telekom is building its industrial cloud with 10,000 Nvidia Blackwell chips.
• Microsoft’s project in Portugal is scaling up to 66,000 units.

Every time a European entity “scales up,” they entrench Nvidia’s ecosystem further. For a builder, this is a warning: infrastructure is only as independent as the stack it runs on. If your “sovereign” cloud is just a wrapper for U.S. IP, you’re still subject to their export controls, their pricing, and their whims.

A Fragmented Value Chain

The E.U. likes to point to ASML as their saving grace. Yes, the Dutch own the lithography machines that make modern life possible. That’s a huge lever. But a lever isn’t a full machine. Europe is almost non-existent in the “back end” of the process—packaging and testing. They hold about 4% of that market. They also lack the raw materials, which are largely controlled by China.

Economists are now warning that Europe is caught between two fires. On one side, they depend on U.S. design and intellectual property. On the other, they depend on Asian manufacturing and materials. Moving a data center to Paris doesn’t fix the fact that the silicon inside it was designed in Santa Clara and etched in Hsinchu.

The objective is not to expel foreign suppliers, but to avoid excessive dependence on a single country, company, or technology.

That quote from economist Toni Roldán-Monés hits the nail on the head, but it’s easier said than done. Diversification sounds great in a policy paper, but in the trenches of building a startup or a data center, you go with what works. Right now, what works is the U.S.-Asia pipeline.

What This Means for Builders

If you’re building in the crypto or AI space, this geopolitical tug-of-war matters. It tells us that “sovereignty” is becoming a marketing term rather than a technical reality. If you are building on top of European “sovereign clouds,” you need to look under the hood. Are you just using a localized version of the same centralized tech? If so, your regulatory risk hasn’t actually changed.

For the E.U. to actually win this, they need to stop trying to compete on Nvidia’s home turf and start focusing on the next generation of architecture. Trying to catch up to the 20% global market share target by 2030 looks like a pipe dream when current projections show them hovering around 11%. You can't subsidize your way out of a decade of underinvestment in manufacturing.

The Takeaway

True sovereignty isn't about where the building is located; it's about who owns the blueprints and the tools. Europe’s Chips Act 2.0 is a desperate attempt to create a market, but unless they can foster domestic designers who don’t rely on the CUDA ecosystem, they are just building a very expensive home for American technology. For founders, the move is to stay agile and multi-provider. Don't let a government's “sovereign” branding lull you into a false sense of security. The hardware bottleneck is real, and it isn't going away anytime soon.

Read the original at IEEE Spectrum →