This is an opinion piece. I’m not telling you what to think. I’m asking you to think.
This is the longer version of this article (C1 for B2 readers). It has more research, more data and more voices from the classroom. If you prefer shorter sentences and simpler vocabulary, there is a B1-B2 version of this article here with the same argument, the same classroom quotes and the same question at the end.
I taught the same lesson all week while the country cooked.
It was one of those heatwaves that stops being weather and starts being a headline. Trains slowed down so the rails wouldn’t buckle. Half the news blamed emissions, and half of that half blamed the other half. And I sat in air-conditioned meeting rooms across the south of Germany, sweating anyway, teaching a lesson about a man in a leather jacket who makes computer chips.
The man is Jensen Huang. The company is Nvidia. If you’ve heard of neither, here’s the one line that matters: almost every serious artificial-intelligence system on the planet runs on chips his company designs, and there is no real second supplier. Call it what it is: a near-monopoly.
So every group I taught kept landing on the same question. Is that a strength or a danger — that the whole AI revolution, the thing that’s supposedly about to reshape your job and mine, depends on one company’s chips, designed in America, built in Taiwan? The heatwave and the chip question sat side by side in my week like two strangers on a train who slowly realise they’re going to the same place.
They arrived on Friday. I had a one-to-one session with a client who works in the semiconductor trade. He buys and sells chips for a living, so what he knows is just his Tuesday — first-hand, unglamorous, and usually right. Somewhere in that ninety minutes, the two questions fused into one, and I’ve been unable to unfuse them since.
This is where I’ve landed. I’ll show you the research first, then the classroom, then what I actually think. You’re under no obligation to agree with the last part. In fact, I’d rather you argued with it.
Key Vocabulary — The Man, the Machine, and One Dutch Company
fabless | designing chips but owning no factory to make them | fabless, ohne eigene Fertigung
foundry | a factory that manufactures chips for other companies | Chipfabrik, Auftragsfertiger
lithography | the process of printing microscopic circuits onto silicon | Lithografie
leading-edge | the most advanced generation of a technology | Spitzentechnologie, neueste Generation
a chokepoint | a narrow stage where everything must pass through one place | Engpass, Nadelöhr
a single point of failure | one part whose collapse brings down the whole system | einzelne Schwachstelle, kritischer Ausfallpunkt
to outsource | to pay an outside company to do work you could do yourself | auslagern, fremdvergeben
The Man, the Machine, and One Dutch Company
Start with the surprise. Nvidia doesn’t make Nvidia’s chips.
The company is fabless — it employs some of the best chip designers alive and owns not a single factory to build what they design. The building is outsourced almost entirely to one company in Taiwan, TSMC, the world’s most important foundry. In 2025, Nvidia quietly overtook Apple to become TSMC’s biggest customer, which tells you how much silicon the AI boom is swallowing. Roughly 80 to 90 per cent of the world’s AI chips, by revenue, are Nvidia’s. For the chips that train the biggest models, it’s over 90. Whatever you think a monopoly looks like, it looks like this.
Now go one layer deeper, because this is the part the popular story usually skips. TSMC can only build leading-edge chips because of a machine made by a Dutch company called ASML. The machine does something close to magic: it prints circuits a few nanometres wide using lithography so precise it needs light most of us have never heard of. ASML has a 100 per cent share of that technology. Not a dominant share. All of it. There is no plan B, no rival, no garage start-up about to disrupt them.
Stack it up and you get a dependency chain with a very thin waist: the world’s AI runs on Nvidia, Nvidia runs on TSMC, and TSMC runs on ASML. Each link is a chokepoint. The Dutch machine at the bottom is the real single point of failure — the one nobody outside the industry talks about, and the one that would take the longest to replace. If ASML sneezes, the whole AI economy reaches for a tissue.
Hold that shape in your head, a tall tower balanced on a very narrow base, because the truly interesting question is the one hiding underneath it: why did one particular part of the world decide to stand at the bottom and watch everyone else build?
Key Vocabulary — The Alternative We Won’t Build
sovereignty | a country’s power to control its own affairs | Souveränität, Selbstbestimmung
a subsidy | government money given to support an industry | Subvention, Fördergeld
capacity | how much a factory or region is able to produce | Kapazität, Produktionsvermögen
trailing-edge | older, simpler technology, a few generations behind | ältere Technologie, nicht Spitzenklasse
rare earths | scarce metals essential for magnets, motors and electronics | Seltene Erden
permitting | the official process of getting legal permission to build or mine | Genehmigungsverfahren
to hollow out | to remove the strength from inside something while it still looks whole | aushöhlen, entkernen
The Alternative We Won’t Build
Europe talks about technological sovereignty the way some people talk about going to the gym. Sincerely, frequently, and mostly next week.
The numbers are brutal. In 2000, Europe held about a quarter of the world’s chip-making capacity. Today it’s somewhere around 8 to 10 per cent, and even that figure flatters us, because most of what Europe makes is trailing-edge — the older, sturdier chips that go into cars and washing machines, not the leading-edge ones that run AI. Roughly 80 per cent of the suppliers serving European industry sit outside the EU. We don’t just import chips. We import almost the entire ability to make the good ones.
Brussels noticed. The European Chips Act set a headline target: double Europe’s share of global output by value to 20 per cent by 2030, and mobilise around 43 billion euros in subsidies and private money to get there. Ambitious. Then reality filed its report. The European Court of Auditors (the EU’s own auditors, not some outside critic) concluded the 20 per cent target was “very unlikely,” with internal models pointing to a real figure closer to 11.7 per cent. And in July 2025, Intel indefinitely postponed its planned 30-billion-euro factory in Magdeburg, the flagship of the whole strategy, blaming high energy costs and weak demand. The centrepiece of Europe’s plan to build chips was cancelled by the company that was supposed to build them.
It goes further down than chips, into the ground itself. AI hardware needs rare earths — scarce metals for magnets and motors. Here’s the part that should sting: Europe has them. Sweden sits on some of the continent’s largest known deposits. But permitting a mine there takes six to eight years, tangled in environmental review and genuine, legitimate conflict with the Sámi communities whose reindeer cross that land. So the deposits sit there, geologically rich and politically frozen, while China processes 85 to 91 per cent of the world’s rare earths and we buy them back, refined, at a markup.
My Friday client put it in one flat sentence I keep hearing in my head. We have the materials here, he said, but we don’t have the process to extract them. The materials exist. The will and the capability have quietly left the building.
And that’s the word for it. Hollowed out. On paper, European technology is impressive. Push on it and you find the load-bearing bits have drifted somewhere else: the fabs, the processing, the engineers. Which brings me back to the room, because the sharpest version of this didn’t come from a report. It came from a man describing his own industry.
Key Vocabulary — What the Room Actually Said
to circle (a question) | to keep returning to the same issue from different angles | um eine Frage kreisen
hindsight | understanding an event only after it has happened | Rückblick, Nachhinein
the boiling frog | the idea that slow danger goes unnoticed until it’s too late | der Frosch im Topf, schleichende Gefahr
to hedge | to protect yourself by not depending on one option | sich absichern, streuen
to push back | to challenge or resist an idea politely | Widerspruch einlegen, kontern
day one | the very first day of a crisis, before anyone can react | Tag eins, der erste Krisentag
brain drain | the loss of skilled people to another country | Abwanderung von Fachkräften
What the Room Actually Said
I run these lessons with a warm-up: everyone researches one small piece of the story before we read anything, then brings it back to the group. It means the best facts of the week rarely come from me. They come from a sales lead, an engineer, a lawyer, each holding one tile of a mosaic none of us can see whole.
So when the groups circled the strength-or-danger question, they came at it from every side.
The management thinkers went straight to Jensen Huang’s famous line that his company is always “thirty days from failure.” Hartmut, in my most advanced group, defended it as the only honest posture a company can hold, and reached for the image of the boiling frog: the water warms so slowly that comfort itself becomes the danger, and by the time you feel the heat, you’re lunch. “For me,” he said, “there’s no alternative to this kind of mindset.” Gudrun wasn’t buying it. Tell her every month that the company might die in thirty days, she said, and after the third time she’d simply stop listening. Fair. Fear has a shelf life.
Then Tobias said the thing I’ve quoted to three people since. Turn the question around, he suggested. We only ask whether Huang had vision because he won. “If you fail, it’s a negative connotation. If you win, you get a positive one.” That’s hindsight doing its quiet magic — dressing up luck as genius after the final whistle. The same bet, made by a founder who went bankrupt, would be a business-school warning. We judge the result, then reverse-engineer a flattering story about the decision that produced it.
The technical people went somewhere else entirely: down the supply chain, into the risk. Anton, in a Cologne group, gave the most vivid scenario of the week. If China moves on Taiwan, he said, you won’t be buying a new laptop for a while. The chips stop, the machines to make more chips take the better part of a year each to build, and so, on day one, whatever electronics you already own might be what you’re using for the next ten years. There’s no real way to hedge against a shock like that, either — you can’t exactly stockpile ten years of laptops. Suddenly the thing on your desk stops being a gadget and turns into a ration.
And in a small, quiet session (I was ill that week, and it showed), Carina and a colleague pulled the whole thing one layer up, into software. Their own company’s product had recently switched its AI engine to Claude, an American model. Weeks later, that exact kind of dependency would detonate in public. What struck me was Carina’s honesty about her own reluctance to change. “I like control, not change,” she said — which is, if we’re honest, the quiet motto of an entire continent. Nobody in the room pushed back on that one. We just sort of nodded.
The Friday client carried the sharpest blade, though, precisely because he wasn’t theorising. He sells this stuff. He confirmed the fabless reality from the inside, named ASML as the underrated chokepoint the newspapers miss, and then told me a small story I can’t shake. A little engineering firm near Freiburg, bought by a big American company about a year ago. On paper, he said, the company still exists. But the people don’t work there. The engineers took the better money and went west. A whole European capability, brain-drained out through the back door while the sign stayed up over the front.
That’s the moment my week stopped being about chips.
Key Vocabulary — We Export Our Emissions
to offshore | to move work or production to another country | ins Ausland verlagern
carbon leakage | when climate rules in one country push emissions into another | Carbon Leakage, Verlagerung von Emissionen
to relocate | to move something from one place to another | verlagern, verlegen
performative | done for show and approval rather than real effect | zur Schau gestellt, symbolisch
self-inflicted | a wound or problem you cause to yourself | selbst verschuldet, selbst zugefügt
a ledger | an account book recording what is owed and owned | Kontobuch, Bilanz
prudent | sensibly cautious; wise about risk | umsichtig, klug vorsichtig
self-sabotage | harming your own interests, often without meaning to | Selbstsabotage
We Export Our Emissions
Here’s what I actually think, and you can take it or leave it.
Everyone worries about the dependence. I worry about the thing sitting underneath it: Europe has quietly made itself unable to build the alternative, and then learned to call that inability a virtue. The dependence is only what that inability costs us.
We keep asking the wrong question. “Is it dangerous to depend on Nvidia?” is a fine question for a newspaper, but it points the camera at Taiwan and America and lets us off the hook. The more honest question points at us. Why can’t we build our own? And the uncomfortable answer is that we could. We have the deposits, we had the fabs, we still train brilliant engineers. We’ve simply made a long series of choices that add up to won’t — and then told ourselves a noble story about why.
A large part of that story is environmental. We don’t want to dig the mine, because mining is dirty. We don’t want the fab, because fabs are thirsty and hungry and loud. We don’t want the smelter, the refinery, the coal-black end of the supply chain, because we can see it, and it offends us, and so we pay someone far away to do it instead. And we call that being green.
I want to be very careful here, because this is the exact point where a reader who sweated through that heatwave starts to hear an enemy. So let me say it plainly: if the heat worried you, you should be even angrier than I am about how we actually handle emissions. Because we don’t cut them. We relocate them. We offshore the dirty work to a country with a coal-heavier grid and looser rules, and then we award ourselves a medal for the clean air we’re now importing.
That reflex has a technical name: carbon leakage. When climate rules make production expensive at home, some of it moves abroad, and the emissions move with it. At the macro level, the fingerprints are all over the data. In 2023, the EU’s consumption-based emissions (everything we actually consume, wherever it was made) ran about 21 per cent higher than the emissions produced inside our own borders. Germany alone carried a consumption footprint of over 900 million tonnes of CO₂. The gap between those two numbers is, quite literally, the stuff we buy but don’t make. The pollution is still ours. It’s just filed under someone else’s address.
Now the honest caveat, because I won’t sell you a cleaner story than the evidence supports. This holds at the level of the whole economy, and it gets messier the closer you look. Careful studies of French and Danish firms find that companies rarely offshore in order to pollute more cheaply. They move for wages, for trade access, for proximity to customers, and the global emissions effect is often roughly neutral, sometimes even positive as production modernises. So I’m not claiming that everything made abroad is automatically dirtier. That would be lazy, and wrong.
What I’m claiming is narrower and, I think, harder to wriggle out of: offshoring production is not the clean win we pretend it is. The atmosphere doesn’t care whose ledger the carbon sits on. Move the smokestack out of your eyeline and your conscience cools beautifully; the planet stays exactly as warm as it was.
And that’s the performative version of environmentalism: the kind that counts only the emissions on your own books and conveniently ignores the ones you’ve posted overseas. Real environmentalism counts the carbon wherever it happens. The performative kind counts it wherever it’s convenient. The most expensive kind of virtue is the kind that just moves the problem somewhere you can’t see it — and then sends you the bill in a currency you didn’t expect. That currency is dependence.
Because here’s the twist that makes it worse. When you offshore the mine and the fab and the refinery, you don’t only export the emissions. You export the jobs, the engineers, the know-how — the whole industrial muscle that would have let you say no to someone later. That firm near Freiburg is the story in miniature. We told ourselves we were being clean and sensible, and we woke up unable to build the thing everyone now agrees is essential. A self-inflicted wound we’ve been trained to read as good manners.
Then the whole pattern repeats one floor up, in software — and this time with a date on it.
In June 2026, the American AI company Anthropic released two very powerful models, Claude Fable 5 and Mythos 5. Days later, the US Commerce Department decided they were too dangerous in the wrong hands (jailbreakable towards things like bioweapon design) and ordered access cut for all non-US nationals. Anthropic had no practical way to fence off foreigners, so it pulled both models for everyone, worldwide. A Canadian engineering team at an American start-up was locked out overnight and sued. The block was partly eased two weeks later, for a short list of vetted American organisations. Carina’s company, and thousands like it across Europe, had built on an engine that a foreign government could switch off with a memo — and briefly did.
Same shell game, one layer up. Chips are exhibit A. Frontier AI models are exhibit B. In both, Europe consumes what it has decided it can’t, or won’t, produce, and calls the arrangement responsible right up until the supplier stops picking up the phone. (In fairness, there’s a hedge here worth naming: open-weight models like Mistral and LLaMA can be run on European soil, and they blunt this particular risk. The escape hatch exists. We just have to be bothered to use it, which, see above, is rather the theme.)
So where does that leave the question everyone actually wants answered?
I’ve told you where I land. Europe’s caution looks less like wisdom and more like a trap we walked into with our eyes open, congratulating ourselves the whole way. But I promised you a genuine argument, not a verdict, so let me hand you the other side with both hands. Maybe this caution is prudent. Maybe a continent that chooses cleaner air, stronger environmental rules and slower, more humane permitting is making a mature, moral trade, and dependence is simply the price of it. Maybe “we won’t wreck our own back yard to win a chip war” is exactly the spine the rest of the world lacks.
I don’t think so. I think we’ve dressed self-sabotage up as virtue and I think the bill is already arriving, in chips today and in AI models tomorrow. But that’s the argument, and it’s a real one, with real weight on both sides.
So I’ll leave it where I left the room. Is European caution prudent self-protection — or self-sabotage dressed up as virtue? I’ve shown you my hand. Now show me yours.
Your Move
Forget the vocabulary drill this week. The task is a conversation.
Find someone whose judgement you trust, and argue the side you don’t believe — in English. If you think the dependence is madness, defend the caution. If you think the caution is wise, defend the risk-takers. The point is to hold a difficult position steadily in your second language, with the evidence on both sides, without reaching for the nearest slogan. That’s where real fluency lives: in the messy middle of an argument you can’t fully close.
You’ll fumble a few sentences. Good. That’s acquisition doing its job.
I’ll go first with the vulnerability, as usual. I can lecture a room on industrial strategy across three chip-making countries, and I still can’t reliably tell a German colleague whether I mean mir ist heiß or the far more revealing ich bin heiß after nearly twenty years here. We’re all operating machinery we don’t fully control — mine just happens to be German grammar rather than a lithography line in Taiwan. Same struggle, different dictionary.
This article grew out of a full week of teaching the same question across several groups. The companion three-level lesson (B1, B2 and C1, with the reading, the vocabulary and the full discussion task) is published alongside it. If you want to run this argument with your own team, that’s where to start.