Peter Thiel thinks Europe is losing the AI race.
He might be right.
The usual explanation goes something like this:
America innovates. China builds. Europe regulates.
Nice line.
Also way too simple.
Because if Brussels deleted half its AI rulebook tomorrow morning, Europe would not suddenly wake up with Nvidia, OpenAI, AWS and another trillion dollars of growth capital.
Its deeper problem is this:
Europe increasingly uses technology that somebody else finances, somebody else owns and somebody else can switch off.
That’s not an AI problem.
That’s a power problem.
Europe actually has plenty of AI.
Researchers? Yes.
Engineers? Yes.
World-class industrial companies? Absolutely.
Customers? A gigantic market.
And AI adoption is climbing.
In 2025, around 20% of EU enterprises with at least ten employees in the surveyed sectors reported using AI, up from 13.5% a year earlier.
So Europe isn’t sitting in a candle-lit room refusing to use ChatGPT.
The problem comes one layer deeper.
Who owns the models?
Who owns the cloud?
Who finances the companies?
Who owns the customer relationship?
And when a European company becomes really, really good...
who eventually owns that company?
That’s where the picture starts getting ugly.
An EIB study found that, after ten years, EU scale-ups had raised roughly half as much capital as comparable San Francisco firms. European providers’ share of Europe’s own cloud infrastructure market also fell from 29% in 2017 to roughly 15% by 2022 and stayed around that level through mid-2025.
Read that again.
Europe can become one of AI’s biggest customers...
while becoming weaker inside AI’s economic stack.
Both can happen at the same time.
Imagine you’re the CEO of a German manufacturer.
You want the best AI system for your factories.
Do you buy the mediocre European product because Thierry Breton once gave a speech about strategic autonomy?
Of course not.
You buy the best system.
Maybe that means an American model running on an American cloud using American chips.
That might be exactly the correct decision for your shareholders.
Now multiply that decision across 100,000 companies.
Every individual business becomes more productive.
But collectively, Europe sends revenue, usage data, developer ecosystems and bargaining power toward foreign platforms.
The rational decision for the company can become the strategic weakness of the continent.
That’s the trap.
And yelling “BUY EUROPEAN” doesn’t fix it either.
Forcing businesses to use worse products is basically economic self-harm wearing a sovereignty costume.
The real question is harder:
Who pays while the European alternative becomes good enough?
Investors?
Governments?
Customers?
Probably some combination of all three.
This is why the regulation debate annoys me.
Regulation matters.
Bad regulation can absolutely slow companies down.
A startup with 30 employees feels a compliance burden differently from Microsoft.
And Europe has mastered the bureaucratic art of turning one rule into seven forms, three agencies and a PDF that looks like it was designed during the Austro-Hungarian Empire.
But regulation cannot explain everything.
Europe’s cloud ownership problem started well before the generative-AI boom and before today’s AI-specific rules.
Remove the AI Act tomorrow and Europe still needs:
Capital.
Power.
Data centres.
Customers willing to buy from challengers.
Experienced operators.
Liquid capital markets.
And companies capable of scaling across borders without treating every European country like a new expansion campaign.
That’s the boring answer.
Which usually means it’s closer to the truth.
And capital does more than fund companies.
It funds mistakes.
This is wildly underrated.
AI companies don’t discover the perfect product on Tuesday and IPO on Friday.
They spend years doing stupid experiments.
Hiring the wrong people.
Burning compute.
Changing product direction.
Losing customers.
Rebuilding infrastructure.
Trying again.
American venture capital doesn’t just finance success.
It finances the right to be wrong for longer.
That creates learning.
Then those founders become investors.
Employees become founders.
Executives become board members.
Customers become early adopters of the next company.
The money compounds into an ecosystem.
If your companies continually relocate, sell early or raise from foreign capital because domestic expansion financing is thin, you don’t just lose ownership of one company.
You lose pieces of the next ten companies too.
That’s the flywheel Europe should worry about.
So should Europe build its own OpenAI?
Maybe.
But trying to duplicate every layer of the American AI stack would be brutally expensive.
Models.
Cloud.
Chips.
Data centres.
Developer platforms.
Applications.
Energy infrastructure.
Europe doesn’t need autarky.
It needs options.
There’s a massive difference.
To me, technological sovereignty doesn’t mean:
Everything must be European.
It means:
If your supplier screws you, can you leave?
Can you move your data?
Can you switch models?
Can critical systems keep operating?
Can you test the claims your vendor makes?
Can you negotiate pricing instead of accepting whatever appears in the renewal email?
That’s power.
A European supplier can lock you in too.
An American supplier can sometimes be perfectly replaceable.
So nationality is only part of the equation.
The variables are concentration, substitutability and switching cost.
This creates an interesting strategy for Europe.
Don’t try to win every layer.
Win layers where Europe already has leverage.
Industrial AI.
Manufacturing.
Energy.
Pharma.
Science.
Robotics.
Defense.
Infrastructure.
Europe owns extraordinary domain expertise and some of the world’s most valuable industrial customer relationships.
So perhaps the play isn’t:
Build Europe’s ChatGPT.
Maybe it’s:
Build the AI systems that run the world’s factories, grids, laboratories and machines.
But even here there is a catch.
If your “European AI champion” is nothing more than a nice interface sitting on top of someone else’s model, someone else’s cloud and someone else’s distribution...
you haven’t built sovereignty.
You’ve built a reseller.
The question becomes:
What do you own that your supplier cannot easily copy?
The customer?
The data rights?
The workflow?
The proprietary evaluations?
The domain knowledge?
The distribution?
If the answer is “none of the above,” your moat is basically PowerPoint.
Which brings me back to Peter Thiel.
Europe may indeed be falling behind.
But “behind” needs a definition.
Behind in frontier models?
Probably one conversation.
Behind in capital formation?
Another.
Behind in cloud ownership?
Another.
Behind in AI adoption?
Different again.
Behind in productivity?
That’s ultimately the one that matters economically.
And this is where Europe has to make a choice.
Not between AI and regulation.
That’s Twitter-brain framing.
The actual choices are uglier:
How much dependence is acceptable?
Which technologies are important enough to subsidise?
How much more should Europe pay for resilience?
When should governments back domestic challengers?
When should they let them fail?
And which parts of the AI stack are simply not worth replicating?
Those aren’t technology questions.
They’re capital-allocation questions.
They’re industrial-policy questions.
They’re power questions.
And the scoreboard shouldn’t be how many billions Brussels announces at the next summit.
It should be painfully simple:
Did productivity rise?
Did European companies gain customers?
Did infrastructure actually get built and used?
Can critical systems switch suppliers?
And did taxpayers get something valuable for the money?
Because announcing €100 billion is easy.
Building bargaining power is much harder.
The AI race is usually described like Formula 1.
America is ahead.
China is chasing.
Europe is somewhere behind writing the safety manual.
I think that’s the wrong metaphor.
AI is becoming more like the global energy system.
You don’t need to own every oil field.
But you probably don’t want one country controlling your fuel, your refinery, your pipelines and your emergency reserves either.
Europe doesn’t need to own everything.
It needs enough capability, enough competition and enough alternatives that nobody else owns it.
That’s the real race.
And unlike a benchmark leaderboard, there is no single winner.
There are just countries with leverage...
and countries paying rent to the ones who have it.
— Full Stack Capitalist
I find this text very valuable because it shifts the AI debate away from: "Is Europe over-regulating?" to "Where does Europe possess genuine bargaining power?"
Europe should not strive for costly autarky or feel compelled to replicate every layer of the technology stack. Instead, it needs to build targeted, interchangeable options: open and portable data and model architectures, European cloud capacities for critical applications and significantly more patient growth capital.
For me the most compelling focus lies in industrial AI: factories, energy, pharmaceuticals, science, robotics, defense, and critical infrastructure. In these areas, Europe boasts not only strong research capabilities but also real customers, complex workflows, proprietary data and domain expertise.
A European interface running on a US model in a US cloud does not constitute sovereignty. What matters is who controls data rights, workflows, integration, evaluation, and customer relationships.
The political benchmark should be pragmatic: Is productivity rising? Are European re critical systems becoming truly portable? And is a sufficient share of value creation remaining in Europe?