Europe’s plan to spend billions on its own AI infrastructure sounds like a dream. On paper, it’s about "digital sovereignty" and keeping pace with Silicon Valley. But look closer at the EuroHPC Joint Undertaking and the recent pushes for a massive "AI Factories" network. You’ll see a project that might actually suffocate the very startups it aims to help. If you're a founder or an investor in the EU, you shouldn't just celebrate this influx of cash. You should be worried about where it's going and who actually gets to use it.
The European Commission is doubling down on a strategy that prioritizes hardware over talent. They’re betting that by building more supercomputers and physical data centers, they can force an AI revolution. It's a top-down approach in an industry that moves horizontally. Big tech in the US didn't win because the government built them a server farm. They won because they had the flexibility to pivot every six months. The EU's multi-billion splurge is rigid, slow, and critics are starting to shout about it.
Why the EU AI investment strategy is failing startups
The core problem isn't the money. It's the bureaucracy attached to every cent. To get access to these high-performance computing (HPC) resources, a small team in Berlin or Paris has to jump through hoops that would make a marathon runner dizzy.
I’ve talked to founders who tried to use EuroHPC resources. They spent weeks filling out forms only to be told the wait times were months long. By the time they get the compute power, the model they wanted to train is already obsolete. In the AI world, six months is a lifetime. OpenAI or Anthropic don't wait for a committee to approve their next GPU cluster. They just scale.
Most of this multi-billion euro plan is focused on "sovereignty." The idea is that Europe shouldn't depend on Nvidia chips or AWS clouds. That’s noble. It’s also nearly impossible right now. By trying to build a localized version of everything, the EU is spreading its resources too thin. Instead of being world-class at one thing, they're becoming mediocre at five things.
The critics aren't just lobbyists for American tech. They are European economists who see a repeat of the Concorde fallacy. We’re throwing good money after bad because we’ve already committed to the idea of a "European Cloud."
The hidden costs of the AI Factories initiative
The recent push for "AI Factories" is the latest shiny object in Brussels. The plan involves upgrading existing supercomputers specifically for AI training. Sounds great, right?
Here is what they don't tell you. These "factories" are often located in places that make sense politically, not technically. Data centers need massive amounts of cheap, green energy and cooling. When you pick a site based on which member state needs a win, you end up with inefficient infrastructure.
The GPU shortage and the procurement trap
Europe is trying to buy its way out of a shortage that money alone can't fix. The demand for H100s and B200s is global. When the EU places a multi-billion euro order, they're still at the mercy of a supply chain they don't control.
- Slow procurement cycles: EU public tender rules mean it takes years to finalize a purchase.
- Hardware aging: By the time the machines are plugged in, they're often a generation behind.
- Maintenance gaps: Building the center is easy; finding the specialized staff to run it at 99.9% uptime is the real challenge.
European taxpayers are essentially subsidizing the purchase of American silicon to put into buildings that are too slow to compete with private clouds. It’s a circular logic that ends with us owning a lot of very expensive, very quiet hardware.
Innovation happens in the code not the concrete
If you want to understand why Europe is lagging, don't look at the number of flops in our supercomputers. Look at the regulatory burden. The AI Act, while well-intentioned for safety, has created a climate of fear.
Investors aren't pouring money into EU AI companies because they're worried about the hardware. They’re worried about the legal fees. While the US and China are "breaking things" and moving fast, Europe is busy writing the instruction manual for a machine that hasn't been built yet.
The splurge on infrastructure is a distraction. It’s easier for a politician to cut a ribbon on a new data center than it is to fix the fragmented venture capital market. We don't have a "compute" gap as much as we have an "exit" gap. When a European startup hits a certain size, they often move to the US to go public or get acquired. No amount of supercomputing power in Finland is going to stop that.
What a better AI plan would actually look like
If we actually wanted to help European tech, we'd stop building monuments to "sovereignty" and start making it easier to fail.
We should be subsidizing the use of existing clouds—wherever they are—rather than building our own proprietary ones that nobody wants to use. Give a startup a €500,000 credit for any cloud provider they choose. That is an immediate boost to their runway. It lets them compete today.
We also need to talk about the "brain drain." We're training some of the best computer scientists in the world at ETH Zurich, Oxford, and TU Delft. Then they move to California because the salaries are triple and the red tape is non-existent.
Money spent on "AI infrastructure" should be redirected into massive, no-strings-attached grants for researchers. Let them keep their IP. Let them start companies without taking a decade to spin out of a university.
The risk of a European AI bubble
There is a real danger that this multi-billion euro splurge creates a false economy. If the only reason a company exists is because it gets free time on a government supercomputer, it isn't a real company. It’s a zombie.
When the subsidies dry up, these companies will collapse. We’ve seen this before in the solar industry and in previous "tech champion" initiatives. Real innovation is driven by market demand, not by government committees deciding which "pillars of excellence" deserve a budget.
The EU is effectively trying to pick winners. History shows that governments are terrible at that. They usually pick the companies with the best lobbyists, not the best code.
The path forward for European tech founders
If you're building in Europe right now, don't wait for the government's "AI Factory" to open. It’s a trap. By the time you get your login credentials, the market will have moved on.
Focus on these three things instead:
- Optimize for speed: Use the most accessible compute you can find, even if it’s more expensive in the short term. Time to market is your only real advantage.
- Build for a global market: Don't just solve "European" problems. If your software only works within the EU's specific regulatory framework, you’ve capped your growth.
- Ignore the hype: The "sovereign AI" movement is mostly political. Your customers don't care if their LLM was trained on a French supercomputer or a server in Virginia. They care if it works.
Europe has the talent to lead in AI. We have the data, particularly in industrial and medical fields. But we won't win by out-spending the US on hardware. We’ll win by being smarter, leaner, and more aggressive.
Stop looking at the multi-billion euro headlines as a sign of success. They're a sign of a government trying to buy its way out of a problem it created with over-regulation. The real growth will happen in the cracks of the system, away from the state-funded "factories." Use the tools that are available now, stay agile, and don't let the promise of "free" government compute slow your development cycle.
