EU AI Advisers Warn Europe Is Cooked Without Action

Europe’s AI problem is not overregulation but underinvestment in compute, capital, cloud, and industrial scale.

When EU AI advisers warn Europe could be ‘cooked’ without action, my first reaction wasn’t outrage. It was the much worse feeling: yeah, fair.

I’ve spent years defending Europe from the lazy American take that Brussels is just a museum with better bread and more paperwork. Put me at a dinner in Brooklyn and I’ll do the whole speech. Single Market. Airbus. ASML. Erasmus. Industrial depth. I become the human version of an EU flag with espresso.

But this time I can’t do the patriotic routine, because the warning lands. Hard.

If another country can decide who gets access to frontier models, who gets the chips, who gets the cloud, who gets the compute, and who gets a polite buona fortuna after export controls kick in, this is not a normal tech gap. It’s dependency. Clean haircut, nice blazer, still dependency.

That’s why I don’t buy the lazy line that Europe regulated itself into irrelevance. Europe is not failing because it cares about safety. Europe is failing because it never matched regulation with industrial muscle. We got very good at writing rules for technology we still rent from other people. As an Italian who genuinely loves the EU maybe more than is healthy, that one hurts.

The scary part is how obvious this sounds now

What struck me about the “cooked” line wasn’t just the phrasing. It was that nobody serious even blinked.

People in Brussels usually speak in diplomatic oatmeal. “Frameworks.” “Stakeholders.” “Pathways.” Verbs so soft they could be used as pillows. Then suddenly, according to Euractiv’s reporting, EU AI advisers are basically saying Europe needs drastic action on compute, infrastructure, capital, and deployment or it gets left behind. That’s not normal Brussels language. That’s founder-after-seeing-the-burn-rate language.

Good.

Because the polite fiction has finally collapsed. Regulation does not create European AI competitiveness on its own. A board is not a product. A forum is not a model. A glossy strategy PDF is not compute, no matter how expensive the consultant was.

The Commission’s own documents basically admit this if you read past the polished language. In the 11 June 2026 AI Board meeting, the Commission discussed the Tech Sovereignty Package with a focus on the Cloud and AI Development Act and its role in strengthening AI innovation in Europe. “Geoeconomic landscape” is Brussels-speak for: okay, this is real now.

At that same meeting, the Commission also presented the new Scientific Panel and AI Act Advisory Forum. Fine. Useful, even. I’m not anti-governance. I run companies. I enjoy not being sued. But governance is not capacity. You can have the cleanest rulebook in the world and still be a tenant.

I know this feeling too well from startup life. Everyone has a roadmap. Everyone has a Notion doc. Three calls, six priorities, one cute deck. Everybody feels productive. Then the month ends and nothing shipped. Europe has that energy sometimes. We confuse motion with traction. My nonna had a cleaner phrase for it: molto fumo, poco arrosto.

That’s the real warning. Not that Europe is behind. That even insiders now seem to understand the old model — regulate first, scale later — has hit a wall.

The “AI kill switch” should have embarrassed all of us

The most alarming part of this whole debate is not that Europe fell behind in AI. Countries fall behind all the time. The alarming part is that Europe got a live demo of what dependency looks like when access itself becomes a weapon.

A 30 June 2026 Euronews op-ed called the US decision to withhold access to Anthropic’s Mythos for non-US nationals the first real “AI kill switch” moment. That phrase sounds dramatic until you think about what it means. Not price competition. Not a better product winning. Access control. Someone else decides whether you get to play at all.

That is a very different problem.

According to the same Euronews piece, an estimated 88% of the world’s population could lose access to American frontier models overnight if Washington chose to. Eighty-eight percent. That’s not market dominance. That’s geopolitical leverage with an API.

CEPS adds the uglier details. In its piece on regulating the AI Europe isn’t building, it notes that Anthropic initially shared Mythos through Project Glasswing with a small club that included Amazon, Apple, Nvidia, and ENISA. Then, after a reported jailbreak, the US government ordered both Mythos and the public-facing Fable 5 pulled offline worldwide three days later.

Worldwide.

If you still think dependency is an abstract concern after that, I honestly don’t know what to tell you. Maybe also bring back your WeWork tote bag while you’re at it.

The capability side is not exactly calming either. CEPS cites the UK AI Security Institute finding that Mythos Preview completed expert-level cybersecurity tasks more than 70% of the time. That is not a toy chatbot helping a cousin write cover letters. That is strategic capability.

This is the moment where EU AI sovereignty stops sounding like a French policy paper and starts sounding like basic statecraft. If the platform is American, the model is American, the cloud is American, and the legal choke point is American, Europe is not a peer. It’s a customer with a nice accent.

And yes, I say this as someone who lives in America and likes a lot about it. The scale here is intoxicating. The ambition is real. In San Francisco you can feel the money in the air like humidity. Last month I walked past a billboard for an AI startup that had probably raised more money than some EU programs deploy in years, and I had that very specific immigrant-founder feeling: admiration mixed with dread.

You can respect the machine and still not want your home continent permanently underneath it.

Sorry, but this is not mainly an AI Act problem

Here’s my hot take, and I’m pretty comfortable with it: if Europe is cooked on AI, it’s not mainly because of the AI Act.

It’s because Europe underbuilt.

That CEPS title says the whole thing in one line: “Why we have six months to regulate the AI that Europe isn’t building.” Brutal. Accurate. Slightly humiliating.

I’m not dismissing the AI Act. Rules for general-purpose AI, transparency, systemic-risk obligations — yes, these things matter. But the AI Act has become a very convenient scapegoat for deeper structural weakness: fragmented capital markets, slow scaling, procurement inertia, weak deployment pathways, and Europe’s eternal talent for turning cross-border execution into a cultural debate.

The capability gap is also moving too fast for the old excuses. CEPS cites the 2026 Stanford AI Index showing the top closed model leading the top open model by just 3.3% on the Arena Leaderboard. That’s tiny. The moat is thinning. If Europe still isn’t building seriously while capability diffuses this fast, the problem is not that compliance lawyers exist. The problem is that we still do not finance and scale like a continent that understands the urgency.

Same story with Stanford’s Cybench, which CEPS also cites: AI agents’ cybersecurity task success rates jumped from 5% to 96% in about two years.

Five to ninety-six.

In startup terms, that’s not “the market is moving.” That’s “you looked away for one quarter and now the game is different.”

This is where a lot of anti-Brussels commentary gets lazy. Europe did not choose safety instead of innovation, like it was a neat philosophical trade. Europe treated innovation policy like a side dish. Nice to have. Sprinkle parsley. Mention startups in a speech. Then go back to national carve-outs and procedural trench warfare like the iPhone still hasn’t happened.

Science|Business got closer to the adult version of the problem when it said European AI sovereignty will require “ugly trade-offs.” Exactly. Not vibes. Not slogans. Trade-offs. If you want power, you have to prioritize. If you want autonomy, you have to concentrate resources. If you want European AI competitiveness, you need scale.

Europe loves debating the menu because choosing the kitchen is politically painful. I say “we” because I’m fully part of this pathology. I grew up in Italy. I know what facciamo sistema sounds like. I’ve heard “we need coordination” in rooms where nobody was willing to give up a millimeter of local control. It’s one of our continental toxic traits.

Europe doesn’t need another moonshot speech. It needs money

This is where my patience starts to evaporate.

Europe keeps announcing billion-euro plans in a world that has already moved into half-trillion territory. That’s not a small mismatch. That’s the whole plot.

CEPS, in its piece on Stargate and the fight for AI supremacy, describes the US Stargate Project as a $500 billion effort, with an initial $100 billion ramping to half a trillion over four years. The announcement featured President Trump with the CEOs of OpenAI, SoftBank, and Oracle. Say what you want about the politics — and trust me, I often do — but the symbolism was brutally clear: state power, private capital, industrial speed, all lined up in the same frame.

That alignment matters more than Europeans like to admit. CEPS explicitly notes that launching Stargate at the start of Trump’s presidency signaled strong political backing and an explicit alliance with Big Tech. I’m not saying Europe should copy America in every respect. America has a gift for overdoing things until they catch fire. But in strategic sectors, alignment beats fragmentation every single time.

Euronews put it even more bluntly: “Trillions are more realistic than billions” for the AI race.

That sentence should give every European finance ministry heartburn.

We are still emotionally attached to sounding ambitious with numbers that no longer match the battlefield. It’s like showing up to a Formula 1 race with a very elegant bicycle and expecting applause for sustainability.

The Euronews piece argues that the Commission should convene an emergency summit with leading European businesses. Yes. Absolutely. But not one of those polite Brussels gatherings where everyone gets coffee, a lanyard, and a declaration full of verbs in the passive voice. A real summit. Names. Numbers. Deadlines. Capital commitments. Consequences if nothing happens.

And yes, this is where I go full federalist and annoy people.

No individual member state can finance or coordinate this alone. Not France, despite the grandeur. Not Germany, despite the industrial base. Certainly not Italy, and I say that with love and a passport. This is exactly what the EU is for: pooling scale when fragmented nation-states become strategically unserious.

If you still think AI infrastructure should mostly be handled country by country, you are bringing a Vespa to a Formula 1 race and insisting it builds character.

The same Euronews piece calls for completing the Capital Markets Union and the Single Digital Market in fast-track mode. Good. Finally. We’ve been talking about Capital Markets Union AI implications for so long that the acronym feels older than some founders.

Europe does not lack money. Europe lacks mechanisms to turn European savings into European power. That’s the difference.

A group of EU advisers discussing AI regulations, highlighting urgency and collaboration for Europe's future in technology.

Compute matters. But empty AI gigafactories won’t save Europe

I’m fully on board with building more European AI infrastructure. Europe needs compute. Europe needs cloud capacity. Europe needs serious data-center muscle. No debate there.

But if the plan becomes “build giant facilities and hope talent magically reorganizes itself around them,” then mamma mia, we are about to build some extremely expensive empty cathedrals.

CEPS’s Goldilocks analysis of the EU gigafactory strategy is smart precisely because it avoids infrastructure fetish. According to CEPS, €10 billion has gone into AI factories and €20 billion in public-private funding has been committed to AI gigafactories Europe. Those are real numbers. This is not nothing. But hardware by itself solves very little if the surrounding system is clumsy.

The placement issue is a perfect example. CEPS warns that AI factories are often being located near existing EuroHPC sites and energy-efficient infrastructure instead of near established AI talent hubs. I understand the logic. Land, power, cooling, water, supercomputing assets. All sensible. But sensible for what? A ribbon-cutting? Or a founder trying to train models without relocating half a team to the middle of nowhere because a committee liked the grid connection?

Their best line is devastating because it’s true: “Talent won’t go to the factories — the factories must go to the talent.”

Print it. Frame it. Hang it in every ministry.

Because yes, training can be done remotely, which is why CEPS argues for a federated network with seamless remote access. That’s exactly the right instinct. Sovereignty is not just owning hardware. It’s making sure European researchers and startups can use it without ten forms, three national intermediaries, and a six-month waiting period that ends with “unfortunately…”

I’ve felt a softer version of this myself. A couple of years ago I tried to get access to a European public innovation program for a startup experiment. The process was so administratively baroque that by the time an answer arrived, the product had changed twice and one investor had already ghosted us. I remember feeling not even angry. Just small. Like I was asking a civilization that claims to back innovation to please acknowledge that time exists.

That’s the danger here. Europe loves geographic balance, and I get why. Cohesion matters. Political buy-in matters. But there are moments when spreading projects around so everyone gets a slice stops being solidarity and starts being dilution.

Cute for regional policy. Dangerous for frontier infrastructure.

If we want EU AI sovereignty, compute has to be designed for use, not just ownership. Access. Federation. Talent density. Energy realism. Deployment pathways. Otherwise we’ll have lovely industrial monuments and not enough companies worth powering.

Sovereignty means picking winners. Fast.

This is the sentence Brussels hates, so naturally I’m going to say it.

Serious sovereignty means preference.

It means concentration. It means deciding that some sectors, firms, clusters, and infrastructure layers matter more than others, then backing them hard enough that the decision actually matters. You cannot maximize cohesion, local distribution, political comfort, strategic independence, fiscal caution, open competition, and speed all at once. At some point, somebody has to choose.

That’s what Science|Business means with “ugly trade-offs.” Not some moral failure. Just reality. Power is finite.

To be fair, the Commission finally seems to understand the shape of the challenge. Its tech sovereignty package includes Chips Act 2.0, the Cloud and AI Development Act, an Open Source Strategy, and a Strategic Roadmap for Digitalisation and AI in Energy. That is a real stack. Not just one lonely law trying to cosplay as industrial policy.

The Commission also says these measures are meant to support Europe’s ambition to become an “AI continent” and reduce “structural dependencies.” Good. That is exactly the right diagnosis. Structural dependency is the disease. Everything else is symptom management.

Ursula von der Leyen said in the Commission release: “We cannot afford to depend on others for the technologies that keep our hospitals running, our energy grids stable and our services secure.”

We cannot afford to depend on others for the technologies that keep our hospitals running, our energy grids stable and our services secure.

She followed it with the line that matters even more: “Europe has the talent, the research excellence, the industrial base and the Single Market. Together, we must turn these strengths into technological sovereignty.”

Europe has the talent, the research excellence, the industrial base and the Single Market. Together, we must turn these strengths into technological sovereignty.

The key word there is together.

This is where my pro-European bias is not subtle at all. If every country gets a veto, every country gets decline. Sorry, but yes, I mean that. The EU was not invented so that 27 governments could take turns explaining why common action is too hard until every strategic layer is controlled from somewhere else.

And I’m not arguing for centralization because centralization is chic. I’m arguing for adulthood. If cloud matters, centralize procurement power. If compute matters, prioritize cross-border access over national vanity projects. If open-source resilience matters, fund it at actual scale. If a handful of research clusters can plausibly become European champions, back them like adults, not like a committee trying not to offend anyone. If energy is the bottleneck, stop pretending digital policy and energy policy live on different planets.

I want the pro-EU camp to be more honest about this. Loving Europe does not mean defending every procedural reflex. It means wanting the Union to become capable of doing what the moment requires. Sometimes that means saying no to localism, no to duplication, no to performative balance, and yes to strategic concentration.

Because “strategic autonomy” without strategic preference is just branding.

My actual fear is not that Europe is too late. It’s that Europe still thinks it can stay comfortable.

The AI era is going to be brutally unfair to places that confuse market size with power. Europe has the Single Market. Europe has talent. Europe has industrial depth. Europe now even has serious people saying the right things out loud. Bene. Wonderful. Applause.

Now comes the grown-up question: what are we actually willing to do together that none of our member states can do alone?

Because if the answer is still “not much, but with excellent regulatory language,” then yes — EU AI advisers warn Europe could be ‘cooked’ without action, and we’ll deserve the headline.

If the answer is federal scale, shared capital, shared compute, and actual European champions, then maybe this whole “cooked” moment will be remembered as the slap that finally woke us up.

Europe still has time. Not infinite time. Real time.

And in startup land, that’s the difference between a comeback and a postmortem.