Europe and AI: The Continent That Quietly Quit the Frontier

Transcript

Sam: Okay, here's a number that stopped me cold. For every hundred advanced AI chips running in the world right now, about seventy-five sit in the United States.

Alex: And how many sit in Europe? The whole continent?

Sam: Go on, guess.

Alex: I know the answer and it still sounds like a typo. Five. Roughly five.

Sam: Five. Which is why the people actually building AI in Europe have quietly stopped trying to win a race their own politicians keep promising they'll win.

Alex: Europe didn't lose the frontier last week. It already lost it. And the interesting part is what it decided to do instead. Welcome back to Dan's AI Intel — the show where we try to make sense of the fastest, most consequential shift any of us is likely to live through, one story at a time.

Sam: I'm Sam, here with Alex, and today we're doing something a bit different. Everyone tells the AI story as America versus China, two giants slugging it out. There's a third character who's barely in the script — and that absence is the whole point.

Alex: Europe. The continent that wrote the rulebook for the entire internet age. Home to some of the most valuable industrial companies on earth. And in the defining technology of the century, it's basically a spectator.

Sam: So the news peg is that Europe keeps announcing it's going to be, quote, "the first AI Continent." Ursula von der Leyen up on a stage, ten billion euros for AI gigafactories, the works.

Alex: But the question I couldn't put down isn't "is Europe behind." Of course it's behind. The real question is sharper and stranger: when you can't win the race, what's the smartest thing to do — and has Europe actually figured that out, or is it still pretending?

Sam: We're going to run that lens over the compute gap, over the money, over Europe's one big AI hope — a company called Mistral — over defence, and over the moment Europe started quietly dismantling its own most famous law. And there's a twist near the end about where Europe's real contribution to AI is going, that genuinely reframes the whole thing.

Alex: If you've been enjoying the show, do one quick thing before we start — follow us in whatever app you're in right now. It's free, and it's the single biggest thing that helps a small show like this keep going. Right. Let's start with the number.

Sam: So you said seventy-five percent America, about five percent Europe. Where's that number even from — is that a fair way to measure it?

Alex: It's the fairest one we've got, and you can cross-check it. Measure it a totally different way — by data-centre electricity, the physical power the machines actually draw — and in 2024 the split was roughly forty-five percent United States, twenty-five percent China, fifteen percent Europe. Different yardstick, same story. The frontier of AI is a machine that mostly runs on American soil, on American, and increasingly Gulf, money.

Sam: Okay, but here's my instinct, and tell me if it's naive. If the gap is money — throw money at it. Europe's rich. Why can't it just spend its way back in?

Alex: Because a gap this size isn't a spending problem, it's a physics-and-plumbing problem. Think about what one of these AI data centres actually needs. Land. Gigawatts of electricity. A grid connection. The most advanced chips on the planet, which are made in one place, Taiwan, from designs owned in one place, America. Europe is disadvantaged on nearly all of it.

Sam: Disadvantaged how, concretely?

Alex: Its data centres run an estimated twenty to thirty percent more expensively than America's — higher power prices, slower permitting. And in the best European markets, getting a big new grid connection can take up to a decade.

Sam: A decade. To plug in.

Alex: To plug in. And here's the trap. McKinsey reckons European data-centre demand roughly triples by 2030 — from about ten gigawatts of computing load to thirty-five. Which sounds like a boom, until you realise America is building out at a pace measured in hundreds of billions of dollars a year. You can triple your output and still fall further behind, if the other guy is going up tenfold.

Sam: So it's like sprinting faster and still losing the race, because you started half a lap back and the other runner's on a motorbike.

Alex: That's exactly it. The scarce thing was never euros. It was chips, and power, and the kind of capital markets that finance both at enormous scale in a matter of months. Europe is short on all three.

Sam: Alright, so let's actually talk about the money, because the headline number everyone laughed at was that ten billion. Von der Leyen waves ten billion euros and the AI world basically pats her on the head.

Alex: And the ten billion is genuinely misunderstood — in Europe's favour, weirdly. It's not Europe's answer to a frontier lab. It's a seed. It's public money designed to lure in private money to build gigafactories — big chip clusters. Ten billion of public cash structured to unlock at least twenty billion more in private, so call it thirty billion, with letters of intent already signed with American chipmakers — AMD, Nvidia, Qualcomm.

Sam: So the ten billion isn't the real number.

Alex: Not even close. The real commitment is a programme called InvestAI — two hundred billion euros across the bloc. Fifty billion of public money meant to lever in a hundred and fifty billion private. And about twenty billion of it is carved out specifically for those gigafactories — each one imagined to house roughly a hundred thousand of the next generation of chips. That's the actual scale of European ambition.

Sam: Okay, so it's not a token gesture. There's a real plan with real hardware in it.

Alex: There's a real plan. This is the thing I want to be fair about — Europe is not being lazy or cheap here. The ambition is genuine and the structure is sensible. The problem isn't the will. The problem is what that will is up against.

Sam: Okay, two hundred billion. Now that is a serious number. So why does everyone still say Europe's losing?

Alex: Because of what that two hundred billion is standing next to. Mario Draghi — former head of the European Central Bank — wrote this landmark report on European competitiveness. And he laid the gap bare. In 2023, AI investment into the EU was about eight billion dollars. Into the United States? Sixty-eight billion. Europe was putting in roughly four cents on America's dollar.

Sam: Four cents. Per dollar.

Alex: Four cents. And even China was double Europe, at fifteen billion.

Sam: But hang on — that's a single year. The two hundred billion is the multi-year war chest, right? So isn't that comparing one year of American spending to Europe's whole plan?

Alex: That is exactly the right question, and here's the gut-punch answer. Take Europe's entire flagship two-hundred-billion programme — the whole continent, spread over years, twenty-seven countries. It sits at roughly the same level as what a single American AI company now commits to compute. Anthropic alone has signed on the order of two hundred billion dollars with Google, over five years, just for the computing power.

Sam: Wait. One company's cloud bill... equals the entire continent's flagship AI programme?

Alex: One firm, against a continent. At par. And that's the comparison that should actually sting — not the ten-billion press conference. It's discovering that your grand continental mobilisation is, roughly, one San Francisco company's electricity-and-chips contract.

Sam: You know, this is the second time we've bumped into that dynamic — money moving at a scale that stops looking like normal corporate spending and starts looking geopolitical. Quick aside for anyone who wants to go deeper: we did a whole episode on exactly this, how AI turned into a capital game in the summer of 2026 — that's episode twenty-two, from a couple of months back.

Alex: And the lesson from all of it is the same. Spending is downstream of a market that can form frontier-scale capital almost overnight. America has that market. Europe, for all its wealth, does not.

Sam: Why not, though? Europe has huge banks, huge pension funds, trillions in savings. Where does all that money go?

Alex: It stays home, and it stays cautious, and above all it stays national. There's no single deep pool of European risk capital the way there's one in America — it's twenty-seven smaller ponds, each with its own rules. So a company that needs ten billion in a hurry can raise it in San Francisco in an afternoon and would spend a year stitching it together in Europe. The money exists. The plumbing to move it fast, at scale, does not.

Sam: Right, so if the money's outgunned and the chips are American, let me play optimist. Surely Europe has at least one real contender. A champion. There's always a champion.

Alex: There is, and it's the emotional centre of this whole story. A Paris lab called Mistral. If Europe has a horse in the frontier race, it's this one.

Sam: I've heard the name. Give me the honest read — is it actually good, or is it good "for Europe"?

Alex: It's genuinely good, full stop. And the smartest thing about it is what it's decided not to do. Mistral is not trying to beat the biggest American models head-on. And the numbers behind it are real. Annualised revenue went from around sixteen million dollars at the end of 2024 to more than four hundred million by early 2026.

Sam: Sixteen to four hundred. That's — what, twenty-five times, in a year and a bit?

Alex: About twenty-five times. And the CEO, Arthur Mensch, is targeting a billion euros by the end of 2026. And look at who's buying: ASML, HSBC, BMW, the shipping giant CMA CGM. Blue-chip European industry.

Sam: So what are those companies actually buying? Because they could just buy from an American lab, presumably cheaper.

Alex: They're buying something the American labs sell less willingly. Private, on-premise deployments. Custom models trained on their own proprietary data. Hosted in Europe, under European law. They're not buying the highest benchmark score. They're buying control.

Sam: Okay, so far this sounds like a success story. Where's the catch? Because I can hear one coming.

Alex: The catch is in the valuation, and what it tells you. Mistral raised one-point-seven billion euros back in September 2025, at an eleven-point-seven billion valuation — led, tellingly, by ASML, a European industrial giant betting on a European lab. And by mid-2026 it was in talks to raise around three billion more, at close to a twenty-billion-euro valuation. Nearly doubling in under a year. Samsung reportedly taking about a billion of that.

Sam: So the champion's getting richer. That's... good? That's the optimistic read, surely — twenty billion euros, doubling, big names piling in.

Alex: It is good. But now hold that twenty billion up next to the frontier. Anthropic is raising at around three hundred and fifty billion dollars. OpenAI, larger still.

Sam: Oh. So Europe's great hope, at twenty billion, is worth about... a fifteenth of a single American lab.

Alex: Right. And that's the moment the word "champion" comes into focus. Mistral isn't a frontier rival. It's a hedge. Think of it like insurance. If you're a European bank, or a carmaker, or a government ministry, and your most sensitive AI is riding entirely on a company in San Francisco — that's a risk. Mistral is the policy you buy against that risk. Something you host yourself, under your own law, that no one abroad can switch off.

Sam: So it's not "we'll build the best AI." It's "we'll build the AI you're allowed to keep."

Alex: That's a lovely way to put it, and it's exactly right — a real, defensible, growing business. Just not the thing the politicians are implying when they point at it and say "look, we have a frontier lab too." It isn't one, and it isn't trying to be. But here's the twist that undercuts the whole sovereignty story from the inside. To build the compute its independence needs, Mistral is putting up its own data centres. And the second you look at what those data centres are made of — the dependence walks right back in the front door.

Sam: Oh no. What are they made of?

Alex: Its first big site — forty megawatts, in Essonne, going live in 2026 — is filled with roughly eighteen thousand Nvidia Blackwell GPUs. American chips. Financed by an eight-hundred-and-thirty-million-dollar debt round. And the far bigger ambition, a 1.4-gigawatt campus near Paris, billed as the largest in Europe? It's a joint venture. Nvidia supplies the silicon. France's state bank chips in. And a big chunk of the money comes from MGX — a hundred-billion-dollar state investment fund from Abu Dhabi.

Sam: Let me just say that back to you, because it's almost funny. Europe's flagship sovereign-AI infrastructure... is built on American chips... and part-funded by Gulf money.

Alex: Read it again and it still lands the same way.

Sam: So is that just hypocrisy? Are they being fake about the whole sovereignty thing?

Alex: No — and this is the important bit. It's not hypocrisy, it's the actual shape of the constraint. Nobody makes frontier AI chips outside that American-designed, Taiwan-built supply chain. Nobody. So "sovereignty" in European AI can't mean self-sufficiency, because that's not on the menu for anyone. What it can mean is controlling the hosting layer — where the model runs, whose laws govern the data, and crucially, who's able to pull the plug.

Sam: So you can rent the engine, as long as you own the garage and hold the keys.

Alex: You own the garage and you hold the keys. That's a genuinely meaningful thing to own. It's just a far more modest sovereignty than the word suggests when a president says it from a podium.

Sam: Okay, this is where I want to push, though. You keep saying Europe's given up on the frontier. But there must be somewhere it's genuinely, seriously trying. It can't be nowhere.

Alex: There is. And it's exactly where relying on America stops being merely uncompetitive and becomes actually dangerous. The clearest example is defence.

Sam: Ah. Because you're not going to run your military's AI on a foreign company's cloud.

Alex: You are absolutely not. And investors know it. There's a Munich defence-AI company called Helsing — it raised one-point-eight billion dollars in 2026, at an eighteen-billion-dollar valuation. That makes it one of the most valuable private defence-tech firms on the planet.

Sam: Eighteen billion. That's nearly Mistral money, just for battlefield AI.

Alex: Because a continent that has very recently rediscovered the cost of depending on others for its security is not about to outsource its weapons' brains. And that's the same logic — once a capability becomes strategic, leaning on a foreign supplier for it turns into a vulnerability a government will pay serious money to close. We did a whole episode on that exact nerve — America showing it could throttle even its allies' access to frontier AI. Quick aside if that grabs you: it's called "Sovereign AI: America Built an AI Kill Switch," episode seventeen, from earlier this year. It's the backstory to everything Europe is doing here.

Sam: So defence is one pillar. What else is actually working?

Alex: A cluster of genuine winners — and here's the tell — almost all of them in applied AI, not foundation models. The voice company ElevenLabs hit an eleven-billion-dollar valuation on about three hundred and thirty million in annual revenue. Germany's Black Forest Labs — their open image models power a huge chunk of the ecosystem — raised at three-and-a-quarter billion. Add the automation firm n8n, the video company Synthesia. That cohort pulled in over five billion euros across 2025.

Sam: That's not nothing. That's a real roster.

Alex: It's a real roster. But notice what none of them is. Not one is a frontier lab. Every single one of them — Mistral included — ultimately sits downstream of the base intelligence being pushed forward in American labs. Europe is winning the floors of the building it doesn't own the foundation of. And on that foundation layer — the open models a country could host on its own soil — the whole strategic argument for open-weight AI is its own tangle, which, quick aside, we pulled apart in "Open-Weight AI: The Prisoner's Dilemma You Can't Escape," episode thirty-five.

Sam: So the strategy, if you squint, is actually kind of coherent. Own the defence AI you can't outsource. Fund a sovereign model layer for your regulated industries. Win the apps and tools close to your customers. Just... don't try to build the base brain.

Alex: That's it. That's the whole real strategy, said out loud. Narrow, deliberate, defensible. It's just nothing like the "first AI Continent" branding on the poster.

Sam: Alright. You promised me a moment where Europe basically admits all this out loud. Where is it?

Alex: Watch what Europe is doing to its own proudest achievement. The EU AI Act. This was meant to be the "Brussels effect" in action — the idea that Europe sets the world's rules because it can't set the world's pace. Can't build the fastest AI? Fine. Regulate it, and make everyone else follow your rulebook.

Sam: Right, I remember this. Europe as the world's referee.

Alex: The world's referee. And then competitiveness anxiety hit — that Draghi report really changed the mood — and Europe blinked. Late 2025, the Commission proposes a "Digital Omnibus" to simplify its own rulebook. By mid-2026, an AI Omnibus is in force, pushing back the Act's toughest obligations — the high-risk stuff — from 2026 to the end of 2027, and some of it to 2028.

Sam: So they wrote the landmark law, told the whole world it was the future of AI governance, and then... quietly hit pause on it.

Alex: Quietly hit pause on the very law they'd been evangelising. Macron starts talking about "more sovereignty and more autonomy." The Commission reframes its entire posture — from regulator to would-be builder. Ninety-six new initiatives, an "Apply AI" strategy, the whole thrust of which is getting European industry to use AI rather than make it.

Sam: Honestly, though? That's the most self-aware thing in the whole story. It's a bit sad, but it's clear-eyed. They looked at the scoreboard and went, being the referee's worth nothing if we've got no team on the field.

Alex: I think that's the most honest thing Europe has done in this whole saga. The open question is whether stapling deregulation onto a still-fragmented capital market and a still-slow grid actually grows you a team. But the direction of travel is unmistakable. And it's an admission.

Sam: Okay, you teased a twist earlier. A reframe about where Europe's real contribution goes. I've been waiting. Hit me.

Alex: So the sharpest, most deflating version of this whole question is: is Europe basically just a talent farm and a shopping mall for American AI? And the honest answer is — largely, yes. But how it happens is the fascinating part.

Sam: Start with the talent farm bit. Why do the people leave?

Alex: Follow one engineer. A senior machine-learning engineer at a top European lab earns maybe a hundred and twenty to a hundred and eighty thousand euros. That same person at an American frontier lab? Three hundred and fifty to seven hundred thousand dollars.

Sam: So three, four times the pay. For the same brain. That's not a nudge to leave, that's a tractor beam.

Alex: It's a tractor beam. But here's the twist you were waiting for — the talent doesn't even have to emigrate anymore.

Sam: What do you mean? How do you hire someone in another continent without them moving?

Alex: You bring the lab to them. Anthropic opened offices in six European cities — London, Dublin, Zurich, Paris, Munich, Milan — in under a year. It's tripling its international headcount, and its European revenue has grown something like ninefold year on year. So the brilliant researcher stays in Paris, keeps her flat, keeps her life — and works for a company whose equity, whose compute, and whose profits all live in America.

Sam: Oh, that's so much sneakier than a brain drain. The brain doesn't even drain. It just... gets quietly re-pointed while sitting at the same desk.

Alex: And it's structural, it's not a fluke. DeepMind — arguably the single most important AI lab of the last decade — was founded in London. It's wholly owned by Alphabet, and much of its frontier work happens in the US. Of all the startups spun out by DeepMind's own alumni, the United States absorbed around seventy of them. Britain kept twenty-eight. The rest of Europe? A handful.

Sam: So Europe is running the world's best AI talent academy, and then shipping the graduates — or their paychecks — straight across the Atlantic.

Alex: The talent is European. The equity, the compute, and the profit are not. That's the indirect participation, and it is real, and it is enormous. But — and this is the part the gloomy story misses — there's that deliberate carve-out we talked about. The slice of the stack Europe is choosing not to rent. It's small. But a defence champion the state will protect, a sovereign model layer regulated industries will pay for, an applied-AI sector sitting right next to European customers — that's not nothing. It's a real, if humble, answer to the only question that matters now.

Sam: Okay. Land it for me. If someone gets off the train right now and asks you, one sentence — where does Europe actually stand in AI?

Alex: Europe will not be a frontier player this decade. Full stop. The compute gap is fifteen to one. The capital gap is structural. The energy and grid disadvantages are physical, not political — you can't legislate a grid connection into existing faster. And its best people are already being paid three to four times more to work, often from a European desk, for American companies.

Sam: That's the bleak version. Give me the useful version — because I don't think your take is actually "Europe's doomed."

Alex: It isn't. The useful version is that Europe has finally stopped playing the game it can't win, and started playing one it might. Its realistic prize was never a champion lab. It's strategic autonomy in the few places where dependence is genuinely intolerable — defence, a sovereign model-and-hosting layer, critical industrial adoption. Own the things you truly cannot afford to rent.

Sam: And whether it actually pulls that off — what's it hinge on?

Alex: Three things, very specifically. One: can it finally pool capital across its borders at frontier scale, instead of twenty-seven countries each doing their own small thing. Two: can it build and power gigawatts of data centre faster than a decade at a time. And three — the big one — can it keep enough of its own talent to actually staff the effort, rather than exporting it.

Sam: And if those three don't move?

Alex: Then the gloomy instinct is just right. Europe's role in the AI revolution will be to supply the brilliant Europeans who go and build it for somebody else. The talent that lights up the whole thing — flowing out of the continent that raised it.

Sam: That's a genuinely haunting place to land. And it completely rewired how I hear that "first AI Continent" line — it's not a lie, exactly, it's aspiration wearing the costume of an achievement.

Alex: Aspiration dressed as an achievement. That's the episode in five words.

Sam: And that's where we'll leave it. Quick recap of what to actually walk away with. One: the whole thing is decided by compute and capital, and on both Europe is structurally outgunned — five percent of the chips, four cents on the dollar. Two: its one real hope, Mistral, is a sovereignty hedge, not a frontier rival — and even it runs on American chips and Gulf money. And three: the real, credible European effort is narrow and smart — defence and applied AI — and Europe deregulating its own signature law is the tell that it knows exactly where it stands.

Alex: And thank you, genuinely, for spending this time with us. I hope you come away seeing a bit more clearly where all of this is heading — because it's a fast-moving, high-stakes picture with a brutally short shelf life on what any of us thinks we know, and that's exactly what makes it worth following closely.

Sam: One honest note on how the show is made. It's AI-generated. Dan builds a custom stack of AI tools to research, analyse, verify and illustrate the questions worth understanding — mostly to learn them himself, and then publishes it for anyone who'd like to follow along. AI-assisted, fact-checked, and always worth a second look.

Alex: And before we go, one genuinely useful thing you can do. Follow the show. Whatever app you're listening in right now, there's a follow or a plus button — one tap, it's free, and it does two things at once. You'll get every new episode the moment it lands, and honestly, for a small independent show like this one, a follow is the single biggest lever there is for helping it reach other people trying to make sense of all this. So if this was worth your time — go ahead and hit follow.

Sam: And one last quick thing. This show gets better when you push back on it. So if there was something in here you'd argue with, or a thread you want us to pull much harder on next time — a company, a country, a number that didn't sit right — tell us. The address is podcast at connectiveshift dot com. We read every single message, and it genuinely shapes what we go dig into next.

Alex: It really does. Until next time — keep paying attention. This one's worth it.