The Chokepoint Paradox

There is a building in the southern Netherlands, in a town most Europeans cannot place on a map, where the modern world is quietly manufactured. Veldhoven. Twenty minutes from the Belgian border, ringed by the kind of low brick and managed grass that announces nothing. Inside, in rooms cleaner than an operating theatre and stiller than a vault, a company called ASML builds the single most complicated machine the human species has ever made — and the only one of its kind on Earth.

The machine is an extreme-ultraviolet lithography scanner. It is the size of a city bus, costs more than a wide-body jet, and there is a waiting list. To make the chips that run everything you will touch today — the phone, the car, the data centre answering your search, the insulin pump — you must, at the leading edge, print circuitry finer than a virus onto silicon. To do that you need light at thirteen-and-a-half nanometres, which no lamp emits, so ASML’s machine makes it on purpose: it fires a laser at a falling droplet of molten tin fifty thousand times a second, twice per droplet, flashing it into a plasma that glows in a colour no human eye has seen. That light is then steered by mirrors so flat that, scaled to the size of Germany, their tallest imperfection would stand under a millimetre. This is not a flourish borrowed for effect. It is the specification.

ASML makes 100% of the EUV scanners on Earth. Not the lead. Not the plurality. All of them. Around 94% of every lithography system sold at any node traces back to this one Dutch firm, and something close to 85% of advanced chip manufacturing depends on its tools to exist at all. In 2025 it booked €32.7 billion in revenue doing it, at a gross margin north of fifty per cent. If you wanted to name the one place where Europe is not a follower, not a worthy fast-second, but the irreplaceable, sole, planet-wide monopolist of the thing the entire digital economy is built on — it is this town you cannot find on a map.

Hold that for a moment, because the rest of this essay will be unkind, and you should know it begins from genuine wonder. Europe built this. European physicists, European engineers, a supply chain stitched across the continent solved a problem the United States looked at, with all its money, and decided was too hard. The machine in Veldhoven is one of the high-water marks of the civilisation. Look at it. Drag the slider below and watch the leverage radiate out from the node — every ray a thing the world cannot do without the Dutch.

Now drag it the other way.

Because here is the part nobody in Brussels likes to say aloud. The most important machine on Earth does not, in any sense that matters, belong to the people who built it. The light source at its heart — the laser-and-droplet engine that makes the impossible colour — comes from a company called Cymer, in San Diego, California, and it is 100% American. The optics come from Zeiss, a single German supplier with no redundancy. Pull either one and the machine is a sculpture. And sitting above all of it is a quieter dependency that appears on no bill of materials: American law. Because enough of the machine’s content originates in the United States, Washington asserts the right — under what trade lawyers call the Foreign Direct Product Rule — to decide who ASML may sell to, and what it may service, anywhere on the planet.

That reach is not theoretical, and in the last eighteen months it has hardened into architecture. The Netherlands began licensing ASML’s exports to China in January 2025 and widened the net in June 2026 to cover the servicing and spare parts that keep installed machines alive — controls written in The Hague but unmistakably authored in Washington. Around them the United States assembled a multilateral export-control bloc the diplomats took to calling “Pax Silica”, launched in December 2025 and joined by the European Union itself in June 2026. Europe’s crown jewel now runs under a coordinated foreign licence — and it negotiated its way into the cartel that holds the key.

You can watch the cost in a single line on ASML’s own income statement. For most of the decade China was its largest market — about 49% of sales at the 2024 peak. Then the controls tightened, and the number fell off a cliff: 49% in 2024, about 36% by late 2025, then 19% by the first quarter of 2026 — an entire market not lost to a competitor or a better product, but deleted by a foreign-policy decision Europe did not make and could not veto. The record-revenue year and the structural amputation are the same year. They are the same chart. Raise the machine’s American content past the rule’s threshold below, and watch the leash from San Diego snap taut.

This is the paradox the essay is named for, and it is worth stating precisely because everything follows from it: the strength and the exposure are the same node. The thing that makes Europe indispensable is the thing that makes it controllable. There is no version of the ASML story where Europe holds enormous leverage and sits safely outside someone else’s grip — the leverage is the grip, seen from the other end. Europe holds the key. Washington owns the lock. And — this is the part that should keep a European minister awake — Europe appears to have decided that this is fine.

I have spent thirty years in and around this industry, on both sides of the Atlantic, watching European boards reassure themselves with the word world-class. Let me be careful, because the lazy version of this argument is a declinist sneer and I have no patience for it. Europe is not stupid and it is not poor. It is the richest pool of savings on the face of the Earth. It has the engineers, disproportionately so. It writes the papers. It built the machine in Veldhoven. The thesis here is not that Europe cannot. It is the far more uncomfortable claim that Europe does not — that with every structural advantage a continent could ask for, it has arranged its affairs so the value created here is systematically captured somewhere else, and that this is not bad luck or American cunning but a series of specific, nameable, repeated choices.

The pattern is not abstract, and the proof is a roll-call of things you use every day. The World Wide Web was written at CERN, in Geneva, and given to the world for nothing; the MP3 was compressed into being at Germany’s Fraunhofer institute; Skype was built by Estonian engineers for Scandinavian founders; the most important AI laboratory of the 2010s, DeepMind, was founded in London. Every one of them became someone else’s franchise — an American platform, an American acquisition, an American market capitalisation. Europe has invented an astonishing share of the modern world and captured the returns on almost none of it, and the regularity with which this happens is the tell. Once is misfortune. A century of it is a machine.

The roll-call is longer than the famous cases, and quieter. Python, the programming language that now runs a vast share of the world’s data science and AI, was created by a Dutchman, Guido van Rossum, at a research institute in Amsterdam; Linux, the operating system underneath almost every server and cloud on Earth, was begun by a Finn, Linus Torvalds. The capacitive touchscreen that makes the smartphone possible was prototyped by a Danish engineer at CERN in the early 1970s, decades before Apple scaled it into the iPhone. Europe did not merely invent a few products it failed to keep; it invented several of the foundations the American technology economy is built on, let them go, and now rents the industry they made possible. The divergence that followed is brutal in the aggregate: European output per person has slipped from roughly three-quarters of the American level around 2008 toward something nearer half today — the compounding result of one economy that scaled its inventions and one that exported them.

And the drift is not only of listings but of companies bodily. A steady trickle of European champions simply moves: the French data-science firm Dataiku, the search company Algolia, the treasury-software maker Kyriba, the telephony startup Aircall — all founded in France, all relocating their centre of gravity to the United States, most citing the same plain reason: it is simply easier to be an ambitious company there. The founders stay European, the first risk was European, and the headquarters, the cap table and the eventual windfall end up American. When the most ambitious version of a European company turns out to be an American one, the continent is not failing to start companies. It is failing to keep the ones it starts.

Zoom all the way out and the scoreboard is stark. Of the fifty most valuable companies on Earth, something like forty are American and fewer than ten European; there is no European Google, no European Amazon, no European Apple, and the continent’s most valuable company is a maker of the machines that other people use to build the future. Europe spends about 2.2% of its output on research and development against America’s 3.45%, and has for years — a gap that looks small in any single year and compounds into a chasm over a generation. None of this is destiny. It is the cumulative read-out of a thousand smaller choices, each individually defensible, which is exactly why the pattern is so hard to see and so hard to break: there is no single villain to point at, only a machine, running smoothly, in the wrong direction.

And lest this seem like a story only about the AI age, recall that Europe has run this exact play before, at the top of the last technology wave, and lost. In May 2000 Nokia was worth some three hundred billion euros and Europe owned the mobile phone; Sweden’s Ericsson held perhaps forty per cent of the world’s handset market; the continent had written the global GSM standard. Within a decade it was gone. Nokia lost 98% of its value and was sold for parts to Microsoft; Ericsson retreated to network gear; Siemens dumped its phone division on a Taiwanese firm that promptly went bankrupt. Europe did not lose mobile because its engineers were beaten on the hardware. It lost because, when the value migrated from the device to the software platform — the app store, the operating system, the ecosystem — the Americans built the platform and the Europeans kept polishing the handset. The chokepoint paradox is not a new affliction. It is a recurrence, and the AI age is merely its largest instance yet.

The people running Europe’s biggest companies see it clearly, and say so. Pascal Soriot, who runs AstraZeneca, one of the few European firms that can stand unembarrassed on a global stage, described the continent bluntly as “losing ground” — too focused, he said, on managing costs and distributing benefits, behaving “like one company in decline.” It is a striking thing for a European chief executive to say out loud, and a useful corrective to any suspicion that this essay is an outsider’s sneer. The diagnosis here is not foreign and it is not new; it is the quiet consensus of the people closest to the machine, who have mostly concluded that saying it changes nothing, and gone back to managing the decline with the competence for which Europe remains, genuinely, world-class.

And the abstraction has a price that lands in ordinary lives. The average American household now has a real income something like a third higher than the average European one, a gap that has widened steadily as the productivity lines diverged — which is to say that the chokepoint paradox is not only a story about companies and cap tables but about the slow erosion of what a European salary can buy relative to an American one. For now it is masked by Europe’s genuine advantages: the healthcare, the cities, the security, the holidays, the sense that life is for more than work. Those are real, and worth defending. But a continent cannot indefinitely fund a superior quality of life on the proceeds of an economy it is steadily ceding, and the bill for choosing comfort over dynamism, decade after decade, does not disappear. It is merely deferred — and handed, with the rest of the ledger, to the next generation.

The drift shows up most starkly where the money is finally counted: the stock exchange. Across the fifteen years to 2025, London hemorrhaged hundreds of listings toward New York and Nasdaq; when Sweden’s Klarna finally went public in 2025 it chose the New York Stock Exchange, not Stockholm, and Wise, the British payments champion, began shifting its own primary listing across the Atlantic. The founders are European, the engineering is European, the first risk was European — and the equity, the liquidity and the compounding all come to rest on an American exchange. Europe builds the company; America banks it. That is the whole essay compressed into a single transaction, repeated until it is a structure.

Those choices form a machine of their own, and this essay takes it apart stroke by stroke. It runs in four. Europe saves more than anyone — and routes the savings abroad. It funds the companies that will compete with it — through other people’s venture funds, because its own will not. It trains the talent — and exports it to be paid three to five times as much elsewhere. And then it buys back, at retail, wearing an American logo, the very capabilities it invented and sold at wholesale. Save, fund, train, buy back: four strokes to a turn, and the wheel turns a little wider each cycle, because every rotation hands a rival the capital and the people to win the next one. By the time we reach the engine in full you will be able to put a hand on the flywheel and feel how smoothly it runs. It is, in its way, as well-engineered as the machine in Veldhoven. It is simply pointed the wrong way.

If a continent owns the one irreplaceable machine, banks the deepest savings, and trains the best minds — and still ends each decade poorer, relative to its rivals, than it began — then the problem is not capability. It is nerve.

Hold that as a question, not yet an accusation. It is not money, and it is not even, mostly, regulation, though the regulation is coming. It is the willingness to bank leverage instead of spending it, to fund the future at home instead of renting it from abroad, to insulate a forty-year strategy from a four-year electoral cycle. None of that requires a miracle. All of it requires a decision Europe has, so far, declined to make.

We start small, and close to home, because the whole pattern shows up in miniature in one country before it shows up across the continent — and it happens to be the country that builds the machine. The Netherlands is the canary. It is Europe in high resolution: the same wealth, the same brilliance, the same quiet, polite, catastrophic refusal to back itself. If you want to know how a rich, clever, well-meaning continent talks itself into tenancy, watch a rich, clever, well-meaning country do it first. Veldhoven is where it starts. It is not where it has to end.

The most damning document about the Netherlands was written by the Netherlands. It is a competitiveness strategy, hundreds of pages long, commissioned by people who love the country and know it intimately, and it reads less like a policy deck than a confession. The phrase it keeps circling is “research colony” — a place that generates world-class ideas and watches them get commercialised somewhere else. That is the diagnosis, in the country’s own hand. The rest of this chapter is mostly evidence, and the evidence is unusually clean, because the Dutch did the unkind work themselves.

Start with why the Netherlands is the right place to begin. It is not a poor country having a hard decade. It is one of the richest, best-educated, most open economies on Earth, sitting on the single most important machine in the global economy and a wall of savings most nations would envy. If this country — with every advantage loaded in its favour — systematically ships its value abroad, then the continent-wide version of the story cannot be blamed on poverty or bad luck. The Netherlands is Europe with the contrast turned up. It is the canary precisely because it should be the last one to faint.

And yet the deck’s own headline is a slow suffocation. If the current trajectory holds, it puts the cost of stagnation at €1,700 per citizen, per year in forgone purchasing power — the compounding gap between the growth the Netherlands is managing and the growth it used to manage. Dutch labour-productivity growth averaged roughly 0.4% a year across 2014–2024, against a historic norm closer to 1.8%. (2025 brought a cyclical rebound, which is welcome and does not undo a lost decade.) That sounds abstract until you turn it into money and let it run. Scrub the years below and watch the abstraction become a salary.

A productivity gap is not a number on a dashboard; it is the raise that never arrived, the public service that quietly thinned, the pension that will be a little lighter. Compounded over a decade and divided by a population, it is €1,700 a head a year — and the uncomfortable thing about the figure is that nobody chose it. No minister announced a policy of national stagnation. It accreted, decision by avoided decision, in exactly the way the rest of this essay describes.

And the country is not trying especially hard to reverse it, which is the part that turns sympathy into impatience. A nation that wants to out-grow its productivity problem invests in the future; the Netherlands spends only 2.23% of GDP on research and developmentagainst a 3% target it set itself — a shortfall the deck puts at roughly €6.7 billion every year, and which, closed properly across a decade, would run to something like €86–107 billion in innovation funding alone. Under- investment is half of it. Under-adoption is the other half. Dutch manufacturing runs at a robot density of 209 machines per ten thousand workers; South Korea runs at 1,012. The deck draws the line in plain words: that automation lag correlates directly with the productivity stall. A rich country can fall behind not because it lacks the technology, but because it declines to install it — which, for the nation that builds the machines the technology runs on, is a particular kind of irony.

So where does the value leak out? Not at the lab bench. Dutch science is genuinely excellent — Wageningen is the world’s number-one agritech university, the photonics and quantum work is world-class, the citation counts are the envy of far larger nations. The leak is downstream, at the precise point where an idea is supposed to become a company. The deck calls it the missing middle, and the data behind it is brutal: zero AI unicorns founded in the Netherlands. Not few. Zero. Sweden, with roughly half the population, has minted dozens across its modern history — the figure the Dutch benchmark cites is 41 — and produces them at about four times the Dutch per-capita rate, off the back of the Spotify and Klarna founder-mafias that recycle capital and nerve into the next cohort. The Netherlands produced Adyen and Mollie, world-class payment companies both — and the deck’s own verdict is that they succeeded despitethe ecosystem, not because of it, often forced into early American exits to find the capital to grow.

The kill happens earlier and more completely than most people imagine. New biotech company formation in the Netherlands collapsed from 48 firms in 2018 to 12 in 2024 — a three-quarters fall in the pipeline feeding the Leiden–Amsterdam life-sciences cluster. Of the research that does spin out, only a sliver ever reaches revenue. The picture is not a tidy bar chart of attrition; it is a haemorrhage, and it bleeds at one specific wound — the jump from prototype to pilot, where the cheque that should exist does not. Hover the gaps below and read the size of the cheque that was missing at each stage.

The Valley of Death is not a Dutch invention — every innovation economy has one — but the Netherlands has dug it unusually deep, and underneath the missing cheque sits a missing instinct. Dutch policy, the deck argues, is corporatist by reflex: it backs the incumbents it already knows — Shell, Unilever, Philips — over the challengers it does not, so generic support flows to exactly the firms least likely to build the next industry. The capital markets compound it. Where Israel and the United States run on venture equity that is paid to take risk, the Netherlands runs on banking dominance over venture capital — debt for the safe, subsidies for the rest — and subsidy-dependent start-ups do not become champions. Even the universities work against the grain: restrictive technology-transfer rules turn every spin-out into a negotiation, and the excellence at Delft and Wageningen stays siloed, never consolidating into the industrial clusters that turn a lab into a sector. Each of these is a decision about who gets the benefit of the doubt. The Netherlands keeps giving it to the past.

And the most telling fact in the whole deck is the one place none of this happens. Dutch agritech is not a research colony — it is an empire. The Netherlands is the world’s second-largest agri-food exporter, achieved on roughly 0.4% of global production volume: a small, water-logged country that out-exports whole continents by turning Wageningen’s science into greenhouses, seeds and systems it actually owns and scales at home. It is proof, in the country’s own figures, that the Dutch canconvert world-class research into a durable, home-grown, globe-leading industry — when they decide to. Which makes the failure to do the same in AI, biotech and deep tech not a ceiling but a choice, and the choice is most visible in where the money goes.

Which brings us to the part that should be impossible to write with a straight face. The Netherlands has €1.6 trillion in pension assets — one of the deepest pools of patient, long-horizon capital on the planet, money explicitly saved for the forty-year future — and it invests almost none of it at home. Dutch pension funds hold only around 6% of their assets in European equities; the domestic allocation is a rounding error, parked instead in foreign sovereign bonds and, above all, US equities. Read the sequence slowly. The country starves its own scale-ups of late-stage capital — and then posts the savings of its workers to Wall Street, where a good deal of it funds the American firms that will acquire those same starved Dutch scale-ups. The capital that could have built the bridge across the Valley of Death is, instead, financing the far bank.

The venture layer tells the same story in miniature. In the large, late-stage rounds that actually decide which companies scale — the €50-million-plus raises — the share coming from domestic investors has collapsed from 61% to 15%. The Dutch built a world-class research base and a world-class savings base and then routed the connection between them through San Francisco. A promising Dutch company that wants to grow does not find Dutch money; it finds American money, and American money, reasonably, wants the company near the rest of its portfolio. The relocation is not theft. It is the predictable physics of who showed up with the cheque — and the reason no Dutch trustee is ever punished for owning US Treasuries and never rewarded for backing a Delft spin-out.

A nation can own the deepest savings on the continent and the best science in its weight class, and still be a research colony — if it insists on lending the savings to its rivals and giving the science away at the seed round.

Why does the money not stay where it is needed? Part of the answer is structural cowardice dressed as prudence. But part of it is something the deck names with startling candour: the Netherlands is governed on the wrong clock. Strategic industries — chips, grids, biotech, energy — run on forty-year horizons. Dutch politics runs on four. And every time the political dial resets, the long-term strategy is knocked back toward zero. The clearest casualty is the Nationaal Groeifonds, the €20-billion National Growth Fund built precisely to make forty-year bets — whose future rounds, some €6.8 billion, were cancelled in 2024, barely a cycle into its life. Spin the political dial below and watch the forty-year asset refuse to move while the strategy resets around it.

The contrast the deck draws is with Denmark, which anchors thirty-year sectoral agreements its governments are bound to honour across electoral turns — the reason Vestas and Ørsted grew into global champions rather than promising pilots. The Dutch instrument for that kind of patience, the Growth Fund, was killed the moment it became politically convenient. You cannot build a forty-year capability on a four-year attention span, and the volatility premium — the compounding cost of every reset — is paid by exactly the deep-tech founders who most need a stable horizon and least control the political weather.

Then there is the talent, which is the part that should make a Dutch reader wince, because it is the most self-inflicted. The Netherlands trains superb engineers and scientists and then prices them to leave. A Dutch AI lecturer earns on the order of €60,000; the same person commands €120,000 to €300,000 and more in the United States or Switzerland. So they go — the deck’s estimate is that 90% of Dutch AI PhDs leave academia, many poached by Google, DeepMind and their peers before they have even finished. And the few foreign experts the country might import to replace them hit a wall that has nothing to do with science: a housing shortage of some 900,000 homes, which makes relocating to the Randstad a punishing proposition. The Netherlands has built a talent machine that runs in reverse — an exporter of the one input it cannot afford to lose.

The cracks reach further down than the salary scale. The foundational pipeline is thinning too: Dutch fifteen-year-olds have shed roughly twenty points across the PISA assessments in maths, reading and science — a quiet erosion of the very numeracy a high-tech economy runs on. And the firms that do manage to scale meet a second wall that has nothing to do with talent or capital: a power grid flashing “code red” for congestion across nearly every province, with industrial electricity priced some 65% above its 2021 level — reindustrialisation blocked at the socket. (That grid earns a chapter of its own, later; note here only that the foundations are cracking on more than one axis at once.)

Underneath all of it sits a culture the deck is brave enough to name. Doe maar gewoon, dan doe je al gek genoeg — just act normal, that’s already crazy enough. It is a genuinely lovely social ethic and a catastrophic industrial one: a quiet levelling instinct in which conspicuous ambition is faintly embarrassing and conspicuous success faintly suspect. It is the soft tissue around all the hard numbers — the reason a brilliant Delft postdoc starts a consultancy instead of a company, and the reason the trustee, the minister and the dean all find it easier to do the normal thing.

It is worth dwelling on what the winners do instead, because the deck’s own benchmark is a catalogue of roads not taken. Israel funds high-risk deep tech through an Innovation Authority on repay-on-success terms — the state carries the downside, takes royalties only if the venture works, and treats a failure like SpaceIL’s crashed lunar lander as national pride rather than national embarrassment, on the way to a 6.3% R&D intensity, the highest in the world. Sweden routes 74% of its venture capitalinto impact and sustainability and lets its Spotify and Klarna alumni recycle money and mentorship into the next cohort. Switzerland leans on pharma-scale corporate R&D — Roche, Novartis — to top the global innovation index year after year. Denmark binds its governments to those thirty-year agreements. None of these is a secret, and none is beyond a country with the Netherlands’ resources. They are simply decisions the Dutch have watched others make.

So this is not a mystery, and it is not a tragedy in the Greek sense — there is no flaw of the gods here, only a sequence of survivable decisions the country keeps declining to revisit. The science is there. The savings are there. The machine, literally, is there, an hour’s drive from the pension funds that will not invest in the companies it could seed. What is missing is the willingness to point them at each other. The Dutch wrote that conclusion themselves, in a deck designed to be read by the people who could change it — which is either the most hopeful thing in this chapter or the most damning, depending on whether anyone acts on it.

It would be dishonest to pretend the Netherlands never holds on to a winner, and the honest exceptions sharpen the rule rather than soften it. Adyen, the Amsterdam payments processor, listed on Euronext in 2018, kept its headquarters at home, and now moves more than a trillion euros a year — Europe’s most valuable fintech, still European. Booking.com, Amsterdam-built, has kept its head office on the canal-ring for two decades even while listing in New York. These prove the outcome is not fated. But notice how the exceptions cluster in payments and travel — not in the frontier industries, AI and chips and foundational software, where the next two decades of value will be decided. Europe can keep the champions of the last era. It keeps losing the ones that will own the next.

And the deeper trap is measurable, not anecdotal. The Netherlands ranks among the world’s top ten for innovation inputs — R&D spending reached 2.44% of GDP, some €12.5 billion, in 2023 — yet slips toward the high teens on innovationoutputs: the patents, the scale-ups, the commercial franchises that turn research into rent. The Brainport cluster around Eindhoven is a genuine marvel of roughly eleven hundred companies and a hundred thousand jobs — but some seventy per cent of them sit in manufacturing and hardware and barely 8% in software, against a third or more in Silicon Valley. It is a hardware powerhouse in a software century: world-class at building the tools, strangely unable to own the platforms those tools enable. The inputs are Dutch. The compounding is somewhere else.

One number captures the whole Dutch paradox. The Netherlands has the highest density of AI professionals in Europe — nearly eleven for every ten thousand people — and almost no AI champions to show for it. A country that ranks sixth in the world for overall competitiveness and third for infrastructure also sits far down the table for the digital-skills depth and scale-up capital that turn dense talent into companies. The brilliance is present and measurable; the machine to compound it into ownership is the thing that is missing — and its absence is a choice, not a shortage of clever people.

When a Dutch champion does try to conquer America head-on, the result is often a cautionary tale. Just Eat Takeaway, born of the Dutch food-delivery pioneer Takeaway.com, bought the American firm Grubhub in 2021 for around seven billion dollars at the pandemic peak — and offloaded it a few years later for a fraction of that, a multi-billion-dollar write-down that stands among the costliest European attempts to scale into the US market on American terms. The lesson Dutch boardrooms drew was not “try harder” but “don’t try” — which is its own kind of defeat.

Even the Dutch crown jewels are, on inspection, the children of a retreat. ASML, NXP and BE Semiconductor are all spin-offs of Philips, once the towering electronics giant of Europe — brilliant companies set free as their parent shrank from a global champion into a health-tech mid-cap. The cluster is real and world-beating. But it is a constellation thrown off by a contracting star, and nothing in the Dutch system has since assembled a new giant to replace the one that broke apart. Europe is unusually good at producing excellent fragments and unusually bad at keeping, or rebuilding, the whole.

The exception that proves the rule is made of glass. In the Westland, a strip of South Holland carpeted with greenhouses, Dutch growers achieve tomato yields of around ninety-five tonnes a hectare against a world average near eighteen, and the country exports more food by value than almost anyone on Earth from a territory the size of a large city. It is one of the genuine wonders of applied science. And notice how it is organised: around Wageningen, the world’s leading agricultural university, knowledge moves by licensing and operating expertise, by family firms and cooperatives refined over generations — not by venture capital and equity scaling. The Dutch are world-beaters precisely where the model is patient, rooted and incremental, and stranded precisely where it demands a founder, a fast cheque and a willingness to bet the company. The competence is total. It is simply sector-locked into the shape the chokepoint machine cannot capture — which is also, unfortunately, the shape that does not build the next trillion-dollar platform.

Two final Dutch details complete the portrait. When a genuine Dutch deep-tech winner does emerge — DataSnipper, an Amsterdam document-automation company that found real product-market fit — its path to scale ran through acquisition by the American automation giant UiPath, its intellectual property consolidated into a New-York-listed group. And when the government finally answered the competitiveness alarm with a headline €2.8-billion technology-and-talent fund, the first-year allocation arrived at around €340 million — an eighty-eight per cent gap between the announcement and the cheque, the four-year political clock starving a forty-year bet before it began. The ambition is real and recurring. So is the under-execution. That, in one country, is the whole machine.

The Dutch state knows exactly how exposed it is, because in 2024 it had to pay to keep its crown jewel from leaving. When ASML signalled it might expand abroad rather than at home, the government scrambled together a two-and-a-half-billion-euro package — nicknamed “Operation Beethoven” — to unblock the housing, grid and talent bottlenecks around Eindhoven and persuade the company to stay. A country does not improvise billions to retain a single firm unless it understands that the firm is the economy. And the pull the other way is relentless: foreign acquisitions of Dutch scale-ups roughly doubled between 2020 and 2025, from sixty-six a year to a hundred and twenty-nine. The canary is not only singing. It is being carried, cage and all, out of the mine.

Hold the Netherlands in your mind now as a high-resolution scan of the patient, because every symptom you have just seen — the savings that flee, the scale-ups that emigrate, the talent priced to leave, the strategy that resets every four years, the grid that cannot carry the load — recurs at continental scale, with more zeros on the end. The canary has fainted. The next chapter walks into the mine shaft to see why the whole of Europe breathes the same air: a standing order, signed by no one in particular, that wires a quarter-trillion euros a year out of the continent and has gone uncancelled for a decade.

In September 2024, Europe was handed the most lucid diagnosis of its own decline it has ever commissioned, and it has spent the eighteen months since proving the diagnosis right by ignoring it. The document was Mario Draghi’s report on European competitiveness — four hundred pages, written by a former central banker with no election to win and no incentive to flatter, at the request of the European Commission itself. It is not a pamphlet. It is the establishment’s own verdict on the establishment, and its central finding is the one this essay keeps arriving at from every direction: Europe is falling behind, the gap is widening, and the cause is not bad luck but a failure to act.

The prescription was specific and enormous. To stop the slide, Draghi argued, Europe needs an additional €750–800 billion of investment every year — on the order of 4.4 to 4.7% of EU GDP, the largest sustained investment surge since the post-war reconstruction, roughly twice the scale of the Marshall Plan relative to the economy. He broke it down to the euro: about €450 billion a year for decarbonisation, €150 billion for digitalisation, €100–150 billion for breakthrough innovation, €50 billion for defence. And he was explicit that perhaps 80% of it must come from private capital— which, given that Europe is the richest savings bloc on the planet, ought to have been the easy part. The map out was not vague. It came with a budget, a timeline, and a list of 383 specific recommendations.

Before the autopsy of what was done with it, look at the diagnosis itself, because the gap Draghi measured is not closing while you read about it — it is opening, on every axis that compounds. Click through them below.

Each of those is a different instrument reading the same disease. Europe deploys a fraction of the venture capital the United States does — the American market runs four to five times larger, and in the late-stage rounds that build giants the asymmetry is starker still: Europe accounts for around 9% of global late-stage funding against America’s 68%. On the capital expenditure that is currently deciding the AI era, the gap is not a multiple but a chasm: America’s big platforms are spending something like $700 billion in 2026 on data centres and AI, against a sovereign European figure nearer €10.6 billion — a ratio in the region of fifty-five to sixty to one. The continent that invented the technologies is now a rounding error in the build-out that will monetise them. Draghi’s report did not predict this so much as see it arriving and name the cause.

So what did Europe do with four hundred pages of correct, expensive, establishment-sanctioned advice? It noted them. Eighteen months on, the most generous accounting of the 383 recommendations finds that only about 15% are fully binding law — and that figure is the improvement, up from roughly 11% the previous autumn. The crawl is the story: four points of progress across a year on a plan whose own logic was that the window was already closing. Whole domains — energy, defence, pharmaceuticals, the automotive sector — register essentially zero structural progress. Scrub the timeline below and watch the map sit on the table, drawn in full, almost entirely unwalked.

It would be easier to forgive if the problem were disagreement. It is not. Almost nobody of consequence in Brussels disputes the Draghi diagnosis; ministers cite it approvingly in the speeches where they announce that it will be considered. The blockage is not intellectual. It is the thing the report itself named, in a phrase that should be carved over the door of the Berlaymont: inertia as a rule of law. Europe has built a governance machine whose steady state is deliberation, in which the production of a strategy is treated as interchangeable with the execution of one, and in which any individual government can slow the whole convoy to the speed of its most reluctant member.

The clearest case is the one most central to Draghi’s fix: the single market for capital that would let Europe’s €33 trillion of private savings fund Europe’s own future. A Capital Markets Union has been official EU policy since 2015. It has been relaunched, rebranded, and re-committed to roughly once per Commission ever since — most recently as a “Savings and Investments Union,” with the bloc’s six largest economies issuing a fresh call to accelerate in May 2026 and a legislative package now aimed, with a straight face, at “summer 2026”. Eleven years of summers. The compass below is the most honest possible diagram of European industrial strategy: a needle that spins with great energy, points authoritatively at each new deadline, and never once settles on a bearing.

And here is the part that turns farce into something colder. As the action has deflated, the rhetoric has inflated to fill the vacuum. The word that now saturates every European tech communiqué is “sovereignty” — digital sovereignty, technological sovereignty, strategic autonomy. It is a fine word, and Europe has discovered it at precisely the moment it has the least of the thing. The continent talks sovereignty while running 70 to 85% of its cloud on American hyperscalers, while its savers keep 34% of their wealth in bank deposits earning nothing rather than funding European firms, while €300 billion a year of that wealth flows out to capitalise the rivals. The vocabulary of independence is doing the work that independence will not.

Letta’s parallel report on the single market, published five months before Draghi’s, had already supplied the most damning number of all: of Europe’s vast savings, some €300 billion leaves the continent every year, much of it to be intermediated by Wall Street and lent back to Europe at a markup, or invested in the American companies that will buy Europe’s best startups. A continent that cannot fund itself is not sovereign in any sense the word has ever carried, however many times a press release uses it. Sovereignty is not a slogan you adopt. It is a set of cheques you are willing to write.

A diagnosis filed under “noted” is not neutrality. It is a more articulate way of losing — the difference between a patient who never saw the doctor and one who read the chart, agreed with it, and changed nothing.

None of this means the door is shut. Draghi’s map is still on the table, and the striking thing about it is how much of the required capital already exists inside Europe — it is a problem of plumbing and nerve, not of poverty. The Savings and Investments Union, if it were ever actually built, would route the €300 billion home. The 15% implemented could become 50%. The summer that never comes could, in principle, arrive. But hope is not a plan, and the first eighteen months are not encouraging, because they revealed the true obstacle: not that Europe does not know what to do, but that knowing has become, for an entire governing class, a comfortable substitute for doing.

The continent has not failed to notice, and the record of its attempts to answer is its own kind of evidence. Gaia-X, launched in 2019 as Europe’s grand sovereign-cloud project, curdled into what one analysis called a “paper monster” — years of working groups and standards documents, little running infrastructure, and a fatal decision to welcome the very American hyperscalers it was meant to counter into the consortium. Meanwhile the European providers’ combined share of their own cloud market has slid from about 29% in 2017 to roughly 15%, and stuck there — treading water in a market growing some 24% a year, which means losing ground in everything but the press release. OVHcloud, the French flag-carrier, only crossed a billion euros of revenue in 2025; the three American hyperscalers spend more than that on data centres in a good fortnight.

The newest attempts are more serious and worth crediting. Deutsche Telekom’s T Cloud went live in 2025, and with Nvidia it stood up an industrial-AI cloud in Munich on more than ten thousand Blackwell GPUs — a billion-euro build that lifted German AI compute by about half, and runs, of course, on American chips; SAP launched a sovereign cloud in France in 2026 through the Orange-and-Capgemini venture “Bleu,” chasing the French state’s SecNumCloud qualification. These are real and they matter. But read the architecture closely and the dependency keeps reappearing one layer down — the chips are Nvidia’s, the reference designs American — and “sovereign” increasingly means a European-run building full of someone else’s technology, governed by a contract that promises Washington cannot reach inside. It is better than nothing. It is not yet ownership.

And the dependence deepens precisely as it appears to localise. The American hyperscalers have learned to answer Europe’s sovereignty worries by building on European soil — AWS, Microsoft and Google have committed tens of billions of euros to data centres and “EU sovereign cloud” regions across Germany, France and beyond. It sounds like reshoring. It is closer to the opposite: a European building, full of American servers, running American software, billed on a subscription that can be re-priced or, in the limit, switched off from abroad. The concrete is European; the control is not. And the gap in who actually spends tells the real story — American AI and data-centre capital expenditure runs into the hundreds of billions a year against a European sovereign-cloud figure still measured in low tens.

And the legal foundation under all of it has never actually been made sound. Ever since the Schrems II ruling struck down the previous EU–US data deal, European data sitting on American clouds has lived in a state of permanent legal provisionality — patched by successive frameworks that the same Austrian litigant who toppled the last two is already challenging. The result is a quiet institutional doublethink: European bodies know the platforms they run on are governed, in the last instance, by American law, and use them anyway, because there is no European substitute at the scale they need. Industry analysts now write about hyperscaler “permanence” in Europe as a settled fact rather than an open question — the dependence hardened from a procurement choice into the assumed shape of the world.

There are, to be fair, signs of a European cloud worth rooting for. OVHcloud has begun winning real public-sector and regulated-industry business on the explicit promise of immunity from foreign law — and in April 2026 the European Commission itself awarded a roughly €180 million sovereign-cloud contract to a slate of European providers (OVHcloud, STACKIT, Scaleway, Proximus) — the first procurement to apply explicit sovereignty criteria, though one winning group leans on the Thales–Google joint venture S3NS, which critics were quick to call “sovereignty-washing”; a coalition of European providers calling itself Euclidia emerged to argue, pointedly, that “sovereign” should mean European-owned and European-controlled, not merely American technology in a local building. These are the right arguments, made by the right people. But they are made from a market position of around fifteen per cent and slipping, against rivals who spend more on data centres in a quarter than the European challengers are worth. The will is finally there in places. The scale is not — and in cloud, as in chips, scale is the whole game.

The most decisive counter-move so far came from Paris. In 2026 France’s interministerial digital directorate, DINUM, issued a directive pushing every ministry and affiliated public body toward sovereign and open-source tools rather than the American defaults — an order touching the software of roughly two and a half million civil servants, and a direct challenge to the assumption that the dependence is permanent. It is the most concrete state-level attempt yet to cancel the standing order. Whether it survives the friction of actually migrating — the retraining, the integration, the thousand small reasons it is always easier not to — is the test that matters, and the one the rest of Europe will be watching.

The price of lock-in became impossible to ignore the moment a vendor decided to test it. When Broadcom acquired VMware, the virtualisation software a great share of European data centres run on, it raised prices on some European cloud providers by figures variously reported between eight hundred and a thousand per cent — an increase you can only impose on customers who have nowhere else to go. The shock did, at least, concentrate minds: France moved its national Health Data Hub off Microsoft Azure to the homegrown provider Scaleway, and the German retail group Schwarz poured billions into its own sovereign cloud. The lesson European institutions are learning, expensively, is the oldest one in commerce: a supplier you cannot leave is not a supplier. It is a landlord.

The race is now being run in concrete, and the scoreboard is instructive. On the European side, beyond Deutsche Telekom’s Munich build the retail group Schwarz is pouring some eleven billion euros into STACKIT, complete with a planned cluster of a hundred thousand AI chips — serious, sovereign, German-owned bets. On the other side, Amazon alone committed nearly ten billion euros to a single expanded region around Frankfurt and launched a separate “European Sovereign Cloud” staffed by Europeans — a clever move that answers the sovereignty objection in marketing while keeping the technology, and the ultimate legal control, American. Both sides are building furiously. Only one of them can also, when it chooses, set the prices and write the terms for the other.

And when Europe did try to write real sovereignty into its rules, it discovered who objects. The EU’s cloud-security certification scheme, EUCS, originally proposed to require that the most sensitive European data sit only with providers immune from foreign law — which is to say, not the American hyperscalers. After sustained lobbying, those sovereignty requirements were quietly stripped out; and in early 2026 the US Secretary of State reportedly instructed American diplomats to press European governments to drop “data sovereignty” and data-localisation demands altogether. The dependence, in other words, is not passively accepted by Washington as a happy accident of market share. It is actively defended — because the tribute documented in this essay is, from the other side of the Atlantic, simply revenue worth protecting. Europe is not only failing to build the alternative. It is being lobbied, firmly, out of even requiring one.

The market’s response to all this has been a sudden proliferation of clouds wearing the word “sovereign” like a label of origin. Deutsche Telekom is building an industrial-AI cloud in Munich packed with Nvidia chips; Oracle and IBM have rushed out their own EU “sovereign” offerings; even AWS and Microsoft now sell European-staffed sovereign regions — the same firms that, in mid-2026, faced preliminary findings that their cloud businesses should be designated gatekeepers under the Digital Markets Act. The word has become a marketing category precisely because the underlying anxiety is real and unmet. But a label is not a guarantee, and most of these “sovereign” clouds still run on American silicon, American reference designs, or American corporate parents. Europe has discovered that it can buy the adjective off the shelf. The noun — actual, unconditional control — is the one thing that is not for sale.

And the diagnosis sharpens to a single startling point. When economists decompose the widening gap in returns between European and American business, almost the entire difference — by one analysis around 90% — turns out to come from one place: the technology sector. Europe is not broadly less productive than America across the board; in many traditional industries it holds its own. It is being out-run almost entirely in the digital, software and AI economy that now drives growth everywhere — the gap showing up as labour-productivity growth of well under one per cent a year in the euro area against several times that in the United States over the same span. The continent did not fall behind at everything. It fell behind at the one thingthat compounds into everything else, and then mistook a tech problem for a general malaise it could manage rather than a specific failure it had to fix.

And the fragility of the arrangement was demonstrated, accidentally and globally, on a single morning in July 2024, when a faulty update from the American security firm CrowdStrikecrashed some eight and a half million Windows computers at once. Across Europe, hospitals turned away non-emergency patients, airports ground to a halt, banks and broadcasters went dark — not because of an attack, but because of a botched software patch pushed from abroad to machines the continent does not control. The episode put a number on the dependence: something on the order of ten billion dollars in global losses from one bad file. A continent whose hospitals, airports and banks can be felled by an American vendor’s routine Tuesday update has outsourced not just its convenience but its resilience — and resilience, unlike convenience, is the thing you only discover you have surrendered on the morning it fails.

Draghi himself put the indictment in a single devastating sentence to the European Parliament: there is, he noted, no EU company with a market value above €100 billion that has been founded from scratch in the last fifty years — while the United States has minted six of them past a trillion. He has a phrase for the predicament, too: the “middle-technology trap,” a Europe stuck specialising in the mature industries of the last century while the high-growth sectors of this one are colonised by others. And a year on from his report, his verdict on Europe’s response was four words long: “every challenge has worsened.”When the most respected economist the European establishment could find to diagnose itself returns, twelve months later, to report that the patient has declined further while the prescription went unfilled, the polite fiction that reform is quietly under way becomes impossible to sustain.

Which forces the question the rest of this essay exists to answer. If the savings are here, and the science is here, and the machine is here, and even the plan is here, correct and costed and signed by the establishment’s own oracle — then where, precisely, does the money go instead? Not metaphorically. Mechanically. The next chapter opens the engine that carries Europe’s wealth out of Europe, four strokes to a turn, and shows you the part no report quite says out loud: that the outflow is not a leak the continent is failing to plug. It is a standing order it has chosen, for ten years running, not to cancel.

A standing order is the quietest instrument in finance. You authorise it once, and it pays out forever — no decision, no signature, no meeting — until someone actively cancels it. Europe has one. It wires a few hundred billion euros a year to the United States, it has been running for at least a decade, and no one in Brussels has cancelled it, partly because no one quite admits it is there. Every symptom in this essay — the scale-ups that emigrate, the champions that list in New York, the pension funds that won’t back their own economy — is a withdrawal from the same account. This chapter is about the engine beneath the symptoms: the machine that carries European wealth out of Europe, and runs a little wider every year.

The mechanism turns in four strokes, and once you see them you cannot unsee them. Europe saves more than almost anyone — and routes the savings abroad. It funds the early life of its best companies — then hands the profitable adulthood to foreign capital. It trains the talent — and exports it to be paid three to five times more elsewhere. And then it buys back, at retail and wearing an American logo, the cloud, the chips and the models it could have owned. Save, fund, train, buy back: four strokes to a turn, and each rotation hands a rival the capital and the people to win the next one. Step through the engine below; there is a toggle to run it backwards, which is the whole argument of the essay in one switch.

Begin with the first stroke, because it is the one that should be impossible. Europe is not capital-poor; it is the most capital-rich bloc on the planet. Households and institutions across the European Union hold something like €33 trillion in private financial wealth, of which roughly €10 trillion sits idle in bank deposits, earning next to nothing — about 70% of household savings parked in cash, against closer to 30% in the United States. This is a reservoir of patient money the size of the entire EU economy and then some. And every year, instead of irrigating the firms next door, around €300 billion of it flows over the dam to be invested in foreign — overwhelmingly American — markets. The continent that saves the most has built its plumbing so the water runs uphill, away from its own fields. Move the valve below and watch how little it would take to turn the flow around.

The second stroke is subtler and crueller, because here Europe does show up — just never when it counts. European capital is present at the birth of its companies and absent at their adolescence. It leads roughly 78% of early-stage rounds, the seed cheques that get an idea off the ground. But by the late-stage rounds — the large raises that actually build a global company — the European share of lead investment collapses to about 18%. Put the other way: some 82% of European scale-up rounds are led by foreign capital, mostly American, and that capital comes with a quiet, reasonable condition — move closer to where the rest of the portfolio lives, to the deeper market, to the New York listing. Europe pays for the childhood and signs the adoption papers at adolescence. Scrub the funding stages below and watch the domestic money fall off the cliff exactly where the company needs it most.

The continent is dimly aware of this one. Its answer, announced in 2025, is the Scaleup Europe Fund — some €5 billion, managed by EQT, with a first close pencilled in for autumn 2026: a single fund to patch a structural hole through which tens of billions drain every year. It is a thimble bailing a reservoir, and the fact that it counts as bold tells you how low the bar has sunk.

The third and fourth strokes show up on the cap table, and this is where the abstraction becomes a row of names. Europe’s champions are conceived in Europe and captured in America, and the captures are recent, dated, and verifiable. Klarna, the Swedish fintech, listed not in Stockholm or Amsterdam but on the New York Stock Exchange in September 2025. Wise, the British money-transfer company, moved its primary listing to Nasdaq in May 2026, keeping London only as a secondary. Arm, the crown jewel of British chip design, is SoftBank-owned and Nasdaq-listed. DeepMind, the most important AI lab Europe ever produced, is a division of Google. Adyen and Mollie took American capital to grow. The blades of the scissors are value created on one side and value captured on the other, and they open a little wider with every champion. Click through them below.

And then there is the detail that turns the whole chapter from negligence into something stranger — the sight of Europe funding the very firms that buy its children, with its own retirement money. Norway’s sovereign wealth fund, the €1.7-trillion pension pot built on Norwegian oil, is a multi-billion-euro shareholder in exactly the American giants at the centre of this story: on the order of $49 billion of Apple and $42 billion of Microsoft, around 1.3% of each. (It is not, contrary to the easy line, a top-five holder of any of them — that tier belongs to Vanguard, BlackRock and State Street — but it is a major one.) A European nation took the windfall under the North Sea and used it to become a landlord of Silicon Valley. There is nothing wrong with the investment; Apple and Microsoft are fine assets. There is something deeply wrong with a continent that can find tens of billions for American incumbents and almost nothing for its own challengers.

Which is the purest expression of the whole machine: the European pension system, sitting on the deepest retirement savings in the world, allocates a vanishing fraction of it to the venture capital that builds the future. The European figure is on the order of a hundredth of a per cent of assets, against an American system that commits roughly a hundred timesmore. The savers’ own money, managed in the savers’ own name, declines to back the economy the savers will retire into — and flows instead to the firms that will sell that economy its software at a markup. The standing order does not need a villain. It needs only a thousand prudent people each doing the normal, defensible, individually-blameless thing.

No one is stealing Europe’s wealth. Europe is wiring it out, on a standing order it set up itself, renews by default, and has decided — for ten years running — not to cancel.

That is also the good news, such as it is. A standing order is cancellable. Nothing in this chapter is a law of physics; every stroke of the engine is a policy, a default, a habit of mind that could be changed by people who are still alive and still in office. The €33 trillion is real and it is here. The plumbing that sends it west is plumbing, not destiny. What it would take to reverse the flow is not genius or money — Europe has both in surplus — but the nerve to point its own savings at its own future, and to keep pointing them there across the four-year cycles that keep resetting the aim.

Follow the cheques and the pattern is unmistakable. Klarna raised round after round from Sequoia and other American funds before listing in New York in 2025; Revolut, valued around thirty-three billion dollars, leaned on SoftBank’s Vision Fund and General Catalyst and stayed private into 2026 on American late-stage money; Graphcore, once Britain’s great hope to build an alternative to Nvidia’s AI chips, ran short of European growth capital and was sold to Japan’s SoftBank in 2024. The shortfall is structural, not anecdotal: the median European growth-stage round runs roughly €18–24 million against €45–65 million in the United States, and the missing middle — the Series B-to-C cheque that turns a promising company into a global one — is exactly where American capital steps in and takes the cap table.

The deepest irony sits in the pension statements. European pension funds allocate a vanishingly small slice — barely a hundredth of a per cent of their assets — to domestic venture capital, a fraction of even the modest American rate, leaving a quarter-trillion euros a year of European retirement savings not backing European companies. Where does it go? Norway’s sovereign wealth fund, the largest single pool of European capital on Earth, holds its biggest single positions in Apple and Microsoft — on the order of forty to fifty billion dollars each — and ranks among the larger holders of Nvidia besides. Which means a Dutch machine’s biggest customers are funded, in part, by the pensions of Norwegian oil workers, whose returns then ride on American technology giants that depend on a Dutch monopoly the fund holds no special stake in. The savings are European. The equity is American. The circle closes on everyone except Europe.

The household end of the same pipe is just as telling. Only about 17% of European household financial assets sit in securities, against roughly 43% in the United States — nearer nine per cent in Germany, five in France — so the European who does invest typically reaches for a global index fund, and the global index is itself a wealth-transfer machine. The most popular such product on European retail platforms tracks an all-world benchmark that is around two-thirds American and tilted hardest toward US technology, which means a German saver buying “the whole world” through a Frankfurt app is, in practice, wiring most of the money to Cupertino and Redmond. The Commission has diagnosed the disease — its Savings and Investments Union, launched in 2025, names an investment gap of €750–800 billion a year and aims to keep European savings at home — but the prescription is years away from changing the plumbing.

The most candid admission of the whole dynamic was written into a trade deal. As part of the 2025 EU–US understanding struck under tariff pressure, the European side put a figure of some $600 billion of European investment into the United States by 2029 on the table — framed, precisely, as the “interest” of European companies rather than a binding pledge the Commission could compel. The hedge matters and cuts both ways: Brussels can no more order private capital across the Atlantic than order it to stay home. But that a transatlantic deal could even be framed around hundreds of billions of European money flowing west — offered as a concession rather than resisted as a loss — tells you how completely the direction of the current is now simply assumed. The leak is no longer hidden. It has become a bargaining chip.

The cap-table capture has a flagship case, and it is the most valuable chip-design company outside the foundries. ARM, founded in Cambridge and the brains inside nearly every smartphone on Earth, was bought by Japan’s SoftBank in 2016 and re-listed not in London but on Nasdaq in 2023 — every euro of value it has compounded since accruing on an American exchange under a Japanese owner. It is not the exception but the median: of European late-stage rounds large enough to build a global company, the clear majority are led or co-led by American investors, who set the terms, the domicile, and ultimately the destination of the exit.

The failure has a precise location on the funding ladder, and European founders have a name for it: the Series B cliff. A European startup can raise a seed round and an A round at home well enough; the angels and early funds exist. It is the next cheque — the twenty-five to a hundred million euros that turns a promising company into a category leader — that goes missing, because Europe has too few funds large enough to write it. So at exactly the moment a company is proving it can scale, the only investors in the room with the chequebook to fund that scaling are American. They invest, and reasonably ask for what capital asks for: a board seat, a US holding company, a path toward a New York listing. The founder says yes, because the alternative is to stall — and another European company quietly becomes, in all the ways that matter, an American one. Not through failure. Through the precise absence of a single missing tier of domestic capital.

Klarna is worth one last look, because its arc rehearses the whole machine in a single company. A Swedish founder built it on European soil; it scaled on American venture capital; in 2024 its chief executive boasted that AI had let it do the work of some seven hundredcustomer-service staff, a story that helped power the narrative toward a New York listing; and by 2026, with service quality slipping, it had quietly begun rehiring humans into a hybrid model. Read it slowly and every stroke is there: build in Europe, fund from America, optimise for the exit narrative rather than the durable business, and list in New York so the compounding accrues to US public markets. The founders bank the wealth; the value-creation machine itself ends up on the American side of the ledger.

The aggregate scale of the shortfall is its own kind of indictment. Atomico’s annual survey of European technology reckons the continent underfunded its tech sector by some $375 billion over the past decade relative to the United States, and would need a trillion dollars more in the next one merely to stop the gap widening. And the exits keep voting with their feet: Klarna finally went public on the New York Stock Exchange in September 2025 at about fifteen billion dollars — down two-thirds from its 2021 private peak, yet still chosen over any European venue — while Wise moved its primary listing to Nasdaq in 2026, keeping London only as a secondary afterthought. Each is a small, rational decision by a single company; together they are a continent quietly conceding that the place where ambition gets financed and rewarded is somewhere else.

The single starkest number in the whole capital story is the pension allocation, and it deserves to be stated baldly. European pension funds put on the order of 0.018% of their assetsinto venture capital; American ones put closer to 1.9% — a hundredfold difference, rooted in a 1979 American rule change that freed pensions to take the risk, which Europe never matched. The Netherlands’ ABP and APG between them steward over a trillion euros, much of it working in American markets. The good news is that the diagnosis has finally started producing instruments: France’s Tibi programme has corralled tens of billions toward deep tech, and a European Tech Champions Initiative is assembling a fund-of-funds to write the big late-stage cheques Europe has always lacked. Whether these stay pilot-scale or become the new default is the hundred-billion-euro question on which the next decade turns.

The founders themselves say it without euphemism. Pascal Gauthier, who runs the French crypto-security firm Ledger, put the capital problem in a single sentence: “Money is in New York today — it’s nowhere else in the world, certainly not in Europe.”He was talking about his own sector, but the diagnosis generalises across the whole essay. The people who run European companies are not confused about where the capital is; they live the shortage daily, and a striking number quietly conclude that the rational response is not to fix Europe but to go where the money already sits. The capital drain is not, at bottom, an accident of policy. It is the aggregate of thousands of founders making the same clear-eyed, individually correct decision to follow the money west.

London tells the same story in miniature, and faster. In the first half of 2025, companies raised a mere £160 million in initial public offerings on the London market — the weakest in more than three decades — while a parade of its biggest names voted with their feet: Wise shifted its primary listing to Nasdaq, the gambling group Flutter moved to New York, and Arm, the Cambridge chip-design jewel, had already chosen New York over its home exchange. A half-per-cent stamp duty on UK share trades that New York does not levy, and fintech valuations that run a third to two-thirds higher on Nasdaq, make the arithmetic brutally simple: for a growth company, listing in London is, in the cold words of more than one adviser, “not rational.” The City that once financed the world now watches the world’s ambition list somewhere else.

Underneath the venture shortfall is a deeper, more cultural gap: Europeans simply do not own equity the way Americans do. Around 62% of American adults hold stocks; in Germany the figure is closer to 15%, in Britain a third. European companies raise some 85% of their financing from banks, against barely half in the United States — which means European capital is structurally biased toward the safe, collateralised loan and away from the risky equity stake that funds a frontier company. And where Europeans do invest in markets, they increasingly do it through American hands: US asset managers — BlackRock, Vanguard, State Street — now run close to half of all European assets under management. So the continent not only sends its savings to American markets; it increasingly pays American firms to manage the journey. The drain is not a single leak. It is the plumbing.

Some of the drain is written directly into Europe’s own rulebook. Under Solvency II, the regime governing the continent’s vast insurance industry, holding equity — the patient, risk-bearing capital that builds companies — carries a punitive capital charge, while government bonds are treated as nearly free. So the very institutions that in America are among the largest backers of long-term innovation are, in Europe, regulated into parking their trillions in safe debt instead. It is the chokepoint paradox turning on itself: a continent that complains it lacks risk capital has built a prudential system that taxes risk and rewards the safe option, then acts surprised when its insurers and pension funds behave exactly as the rules instruct. The savings are not missing. They are being regulated into timidity.

The next chapter is where the cost of not doing so stops being a financial abstraction and becomes a physical one. Because the same continent that exports its capital also owns the single most strategic asset in the modern economy — the machine in Veldhoven — and has spent that chokepoint as carelessly as it spends everything else. We turn to the semiconductor, the crown jewel Europe holds and somehow still manages to lose.

Everything so far has been about what Europe gives away. This chapter is about the one thing it truly owns — the single most strategic asset in the modern economy — and how it has managed to spend even that. The asset is the chokepoint we opened this essay with: ASML’s extreme-ultraviolet lithography machine, the sole tool on Earth that can print the most advanced chips, built in Veldhoven and nowhere else. By every law of strategy a monopoly that absolute should be the most valuable bargaining chip a continent has held since the war. Watch, instead, how Europe has turned the crown jewel into a wealth pump for everyone but itself.

Begin with the leash, because it has only tightened. The Prologue showed you the machine and the American hand on its triggers; here is that hand closing, click by click, over seven years. In 2019 the United States led the move that blocked EUV exports to China. In 2023it reached down the product stack to restrict advanced DUV tools. In January 2025 the Netherlands wrote its own tiered controls; in December it joined the US-led “Pax Silica” export-control bloc, which the EU itself signed up to in June 2026. That same month the Dutch extended the controls to cover servicing and spare parts — the umbilical that keeps installed machines alive — and in Washington a proposed MATCH Act would ban all DUV exports and servicing to China outright. You can read the cost in one line of ASML’s accounts: China fell from about 49% of sales in 2024to 36% by the end of 2025 to roughly 20% in 2026. Scrub the timeline and watch the dial turn — it has only ever turned one way.

Note what this is and is not. It is not that Europe lacks leverage; it is that the leverage belongs, jurisdictionally, to someone else. ASML is a Dutch company, Euronext-listed, and the most valuable technology firm Europe has — its market capitalisation crossed $740 billion in mid-2026, up by a third in a single half-year on the back of the AI build-out. And yet a foreign government decides which of its customers it may serve, on penalty of losing access to the American technology embedded in the machine. Europe holds the asset and rents the authority over it. That is the chokepoint paradox in its purest form: the more indispensable the jewel, the more completely it is controlled from abroad.

You might expect a continent sitting on this asset to have built a fortress around the rest of the chip supply chain. It announced one. The EU Chips Act set a clean, quotable target: Europe would reach 20% of global semiconductor manufacturing by 2030, double its share. It is not going to happen. Europe’s share sits near 10% today and is projected to reach perhaps 11.7% by 2030 — and the European Court of Auditors, in a December 2025 report, called even hitting the target “very unlikely”. The €43-billion Act did catalyse some €80 billion of investment, but the rivals it was racing simply invested more, and faster. The flagship project meant to anchor the whole strategy — Intel’s giant fab in Magdeburg, some €30 billion backed by nearly €10 billion of German state aid — was cancelled in July 2025. The site is reverting to a generic business park. Flip the toggle below between the press release and the field in Saxony-Anhalt.

To be fair to the ambition, there is real building underway — TSMC’s ESMC joint venture in Dresden topped out in January 2026 and will make mature 28-nanometre and 16-nanometre chips, useful and strategic ones, from late 2027. But notice the shape of even the success: the most important new fab on European soil is majority-owned by a Taiwanese company making trailing-edge nodes, years from output, on a continent that builds the machine the cutting edge depends on. Europe manufactures the tool that prints the future and imports the company that uses it.

The deeper reason the 20% was always a mirage is the same disease as the capital chapter, in a different organ: fragmentation. The Chips Act’s roughly €86 billion is not a war chest; it is a scatter of national subsidies — a German cheque here, a French one there — with only a sliver, the Court of Auditors estimates on the order of 5%, actually managed centrally by the Commission. Set that against rivals who move as one: cumulative semiconductor capital spending by the US, Taiwan, Korea, China and Japan ran to something like $580 billion across 2020–2023, behind coordinated public programmes — the US CHIPS Act’s $52.7 billion in federal money, China’s $40-billion-plus state funds. Twenty-seven national chip strategies do not add up to one industrial policy, and the gap shows.

And the part of the chip stack where Europe was genuinely world-class — design — has quietly left the building too. Arm, the British architecture at the heart of nearly every phone on Earth, is owned by Japan’s SoftBank and listed on Nasdaq. Graphcore, Britain’s most promising AI-chip challenger, was absorbed by SoftBank in 2024. The pattern is by now familiar enough to be monotonous: Europe holds the irreplaceable physical tool, loses the authority over it to Washington, loses the leading-edge factories to Asia, loses the design IP to SoftBank and Nasdaq, and keeps — what, exactly? The wages of the engineers in Veldhoven, until the next acquisition. The compounding goes elsewhere.

A monopoly this absolute should have been a fortress. Europe turned it into a tollbooth it does not own the road to — collecting the wages while someone else keeps the toll.

This is what makes the semiconductor chapter the hinge of the whole argument. Everywhere else, you can at least tell a story about Europe being out-competed — out-funded, out-built, out-paid. Here it is not out-competed at all. It won. It holds the single un-substitutable node in the most important supply chain on the planet, and it has still contrived to capture a minority of the value, surrender the strategic control, miss its own target by half, and watch its flagship factory turn back into a field. If a continent can lose with the crown jewel, the problem was never the cards. It was the player.

Look closely at what Europe is actually building with its chip money and the ambition shrinks further. The one major new fab to break ground is ESMC in Dresden — a ten-billion-euro joint venture led by Taiwan’s TSMC with Bosch, Infineon and NXP — and it is slated to produce 28- and 22-nanometre chips when it opens around 2027: competent, useful, and roughly a decade behind the leading edge ASML’s own machines make possible elsewhere. Europe builds the tool that prints the most advanced chips on Earth, then builds itself a factory for the chips of 2015.

And the money behind the ambition is a study in fragmentation. The Chips Act’s headline is €86 billion, but the European Commission directly controls only about €4.5 billion of it — barely five per cent — with the rest scattered across member states under no obligation even to report progress on what they fund. A continental strategy run as twenty-seven national cheque-books is how you arrive at a 20%-of-the-world target that the Commission’s own forecasts now expect to reach about 11.7% by 2030.

The cruelest irony of the crown jewel is that its very importance is what made it a hostage. Because ASML’s machines are indispensable, controlling who may buy them became the single most powerful lever in the technology cold war — and that lever is pulled in Washington, not The Hague. The proof is written on ASML’s own income statement: China, for years its largest market at close to half of all sales, was throttled by export controls to around a fifth in barely two years — an entire market amputated, not by competition or a better product, but by a foreign-policy decision the company could neither shape nor veto. Europe owns the most valuable chokepoint in the world economy and discovered, the moment it mattered, that owning the chokepoint and controlling it are two different things.

Step back and the strategic picture is starker still: for all the Chips Act money and ambition, there is, as of 2026, no leading-edge logic fab in Europe at all — nothing making chips at the three-nanometre class that ASML’s own machines enable in Taiwan, Korea and Arizona. Europe builds the irreplaceable tool and then buys back the finished chips it cannot make at home, at prices set by the firms that bought the tool. It is the four-stroke engine in silicon: invent the critical capability, sell it wholesale, and re-import the value-added at retail. The most advanced manufacturing technology on Earth is European; the most advanced manufacturing is somewhere else.

The leverage runs in more than one direction, and 2026 made that vivid. ASML began shipping its next-generation High-NA EUV machines — at around $350 million each, the most expensive commercial tools ever built — deepening the dependence of every advanced chipmaker on one Dutch firm. But China answered the export controls with leverage of its own, throttling exports of gallium and germanium, raw materials the chip and defence industries cannot do without, until European prices spiked several-fold. The semiconductor war is a war of chokepoints, and Europe sits astride the single most important one — which is precisely why it has so little freedom to act. To hold the decisive node is also to become the decisive target.

There is, in fairness, one corner of semiconductors where Europe is genuinely strong, and it is worth naming because it shows what holding on actually looks like. In power electronics— the unglamorous chips that manage electricity in cars, trains, motors and grids — Germany’s Infineon and the Franco-Italian STMicroelectronics are world leaders, and Infineon is building a five-billion-euro fab in Dresden to press the advantage. This is the model that works: a real product, real customers, patient capital, kept at home. But notice where it sits — in the mature, industrial, slower-moving end of the business, not the bleeding-edge logic that runs AI — and notice that even here the share is slipping as Chinese rivals scale. Europe can hold a chokepoint when it builds in its own industrial heartland. The puzzle of the whole essay is how rarely it tries.

The demand for the Dutch machines, meanwhile, has never been more naked. In 2026 the Korean memory-maker SK hynix placed what was reported as the largest EUV order in history — on the order of eight billion dollars for some thirty machines — a single foreign customer committing more to ASML’s tools than the European Commission directly controls in its entire chip strategy. Brussels, for its part, answered the Chips Act’s shortfall by drafting a “Chips Act 2.0” with another twenty-billion-euro ambition. The contrast is the chapter in miniature: the world bids billions for the European tool, and Europe answers its own strategic shortfall with another plan, another target, another round number, run through the same fragmented machinery that produced the shortfall in the first place.

You do not have to take a critic’s word for any of this; take ASML’s. Peter Wennink, who ran the company until 2024, called the EU’s flagship goal of 20% of global chip production by 2030 “totally unrealistic,” noting that Europe holds maybe eight per cent at best and that hitting the target would mean roughly quadrupling output while everyone else expanded too. When the chief executive of the single most important company in the entire supply chain publicly calls his own continent’s strategy unattainable, the polite fiction that the plans merely need a little more time collapses. The people closest to the machine know exactly how far behind Europe is. It is the political class, drafting its targets in round numbers, that keeps insisting otherwise.

And the pain is not confined to the cancelled mega-fabs; it is hitting the firms that stayed. STMicroelectronics, the Franco-Italian chipmaker, announced some five thousand job cuts over three years as it restructured against Asian competition; when Intel walked away from Magdeburg, it took with it not just three thousand promised jobs but an estimated eighteen thousand more in the supplier base that had begun to assemble around it. A semiconductor ecosystem is a delicate, decades-long thing to grow and a quick thing to lose, and Europe keeps discovering that announcing one is far easier than sustaining it. The strategy keeps mistaking the press release for the factory — and the factory, unlike the press release, demands patient capital, cheap power and political constancy that the continent has struggled to supply all at once.

And in 2025 the chip dependence cut from the other direction, on European soil. Nexperia, a Dutch chipmaker spun out of Philips and now Chinese-owned, became the centre of a control fight: when export restrictions tangled its operations, the disruption rippled straight into European car factories that rely on its humble but essential components, briefly threatening to halt production lines across the continent. It was a small, sharp lesson in how many directions the vulnerability runs — a Dutch-made, Chinese-owned, export-controlled chip with the power to idle German assembly plants — and a reminder that Europe’s semiconductor exposure is not only the glamorous matter of the leading edge but the unglamorous one of the ten-cent parts no car can be built without, and which it long ago stopped reliably making for itself.

The contrast with Asia is the most instructive part of the whole chapter. Taiwan built TSMC out of a government research institute in fourteen patient years; South Korea has sustained a state-and-conglomerate semiconductor partnership for half a century; both treated chips as a multi-decade national project and never wavered. Europe passed a Chips Act in 2023 and watched its flagship, the Intel fab in Magdeburg, be abandoned within two years — the German and Polish subsidies, some twelve billion euros, left undrawn. The difference is not money or talent; it is time-horizon and constancy. Asia’s chip champions were built by states that decided what they wanted and held the course across electoral cycles and downturns alike. Europe keeps deciding, then re-deciding, then quietly letting the decision lapse — which is precisely how a continent ends up owning the world’s most important machine and almost none of the industry it makes possible.

It is worth crediting the genuine progress, because it shows what realistic ambition looks like. In July 2026 Infineon opened a five-billion-euro fab in Dresden — on time, even early — making the power-management chips Europe actually leads in; the silicon-carbide market it serves is forecast to grow nearly tenfold this decade, and European firms hold a real share of it. The same smarter bet shows up in the segments alongside the logic die: STMicro began high-volume production of a silicon-photonics interconnect platform at Crolles in 2026, aimed squarely at the AI data centre, and Italy’s Silicon Box won “Open EU Foundry” status for the advanced chiplet-packaging that stitches finished dies together. This is the strategy that can work: build in the segments where Europe is strong — power, photonics, packaging, the profitable rim around the leading-edge die — in its own industrial heartland, with patient capital. But hold it against the scoreboard. A five-billion-euro fab for mature power chips is a fine and sensible thing; it is also roughly what a single American hyperscaler spends on AI infrastructure in a week. Europe can win the races it enters seriously. The trouble is that the race which will decide the next economy — leading-edge logic for AI — is the one it has, so far, declined to seriously enter.

There are, to be fair, the beginnings of something. In June 2026 GlobalFoundries in Dresden reported the first fully European, end-to-end “sovereign” chip flow — designed, manufactured and delivered without leaving the continent — for a security-critical navigation chip. It is a real and meaningful milestone, and it is also, tellingly, at a mature node, for a niche application. Meanwhile the leash on the leading edge tightens: ASML now remotely monitors every EUV system it has ever shipped, a customer cannot so much as relocate one without the company’s involvement, and the Dutch export rules now require a licence even for the spare parts and software updates that keep an installed machine alive. The most European thing in the entire industry — the irreplaceable machine — is also the most tightly leashed, and the hand on the leash, as the prologue warned, is not entirely Europe’s own.

Which is the right moment to be fair to the other side of the ledger. Critics of an argument like this one always have a ready reply: but look at the bright spots — ASML, Arm, Mistral, Novo Nordisk, the defence-tech upstarts. Europe is not a museum; it makes world-beaters. That is true, and the next chapter takes the steelman seriously, jewel by jewel. It just turns out that every one of them, examined closely, has a foreign outlet, a foreign owner, or a foreign off-switch — that the bright spots are not counter-evidence to the thesis but its most polished proof.

At this point a fair reader is getting restless, and rightly. Every argument like this one has an obvious rebuttal: but look at what Europe makes. ASML. Arm. DeepMind. Mistral. Novo Nordisk and its miracle drugs. Spotify, Adyen, a whole defence-tech generation. Europe is not a museum; it is a workshop that still turns out world-beaters. That is true, and this chapter takes it seriously — jewel by jewel, at face value, with the numbers. The trouble is not that the bright spots are dim. It is that, examined closely, almost every one of them has a foreign outlet, a foreign owner, or a foreign off-switch. They are not the exception to the thesis. They are its most polished proof. Europe grows the crop; someone else owns the silo.

Start with the cap table, because that is where ownership actually lives. The pattern is so consistent it has been measured: by the time a European startup breaks through to its big growth rounds, around 73% of the lead investors are American. The roll-call writes itself. Arm, the British chip architecture in nearly every phone alive, is ~87% owned by Japan’s SoftBank and listed on Nasdaq. DeepMind, the best AI lab Europe ever produced, is a wholly-owned arm of Google. Wayve, Britain’s self-driving hope, raised into an $8.6-billion valuation on the backs of SoftBank, Microsoft and Nvidia. Helsing, Germany’s defence-AI champion, took its $18-billion round from a US-led syndicate. Aleph Alpha, once Germany’s great sovereign-AI hope, was absorbed into Canada’s Cohere in 2026. And the late-stage money that decides all of this isn’t close: US growth funding runs near $141 billion against Europe’s $12 billion — roughly nine to one. Click through the jewels below and read each one’s foreign landlord.

It would matter less if the bench behind these names were deep. It is not. Strip out the handful of European giants — ASML at $743 billion, SAP, Novo Nordisk, LVMH — and there is startlingly little behind them. Europe’s market leadership is two or three players deep where America’s is seven: the US “Magnificent Seven” alone are worth more than $20 trillion combined, a figure that rivals the entire listed value of large parts of Europe. A shallow bench is a fragile one — when a single star goes down, there is no depth to absorb the blow. Bench Novo Nordisk below and watch how much of Europe’s top-tier value leaves with one injury.

And Novo is not a hypothetical injury; it is the most spectacular European corporate collapse of the decade, and it happened in eighteen months. In June 2024, riding the obesity-drug boom, Novo Nordisk was Europe’s most valuable company, worth around $604 billion — a Danish firm briefly larger than the Danish economy. By mid-2026 it was worth about $197 billion: a fall of roughly two-thirds, half of it in 2025 alone. In the same window its American rival Eli Lilly climbed past it to become the first healthcare company ever worth $1 trillion, in November 2025 — now roughly five timesNovo’s size. The proximate trigger was a single disappointing clinical trial (Novo’s CagriSema delivered 22.7% weight loss against a ~25% target). But the deeper lesson is the thin bench again: when Europe’s one giant stumbled, there was no second giant to carry the flag. Scrub the eighteen months below.

Then there is the frontier that matters most for the next decade, where Europe’s flag is carried by a single company: Mistral, valued around €20 billion after its 2026 round. It is a genuinely good lab. It is also a fraction — a small fraction — of the American frontier labs it is meant to rival, which are valued well into the hundreds of billions of dollars; Anthropic alone is on the order of forty times its size, and Mistral trains on rented Nvidia silicon, part-owned by ASML. Europe’s entire answer to OpenAI, Anthropic and Google DeepMind is one company worth less than those labs raise in a single round. You can admire Mistral enormously and still see that one plucky challenger against a phalanx of giants is not a sovereign AI capability; it is a brave tenant on a very large American farm.

The bright spots are real. That is precisely the problem: Europe is fully capable of growing world-class crops, and has built an entire economy in which the silos, the granaries, and the land registry all belong to someone else.

Notice what this chapter is not saying. It is not saying these companies are failures — they are triumphs of European science and nerve. It is not even saying foreign capital is malign; SoftBank’s money built Wayve, and Wayve is better for it. It is saying that a continent which produces this many jewels and retains ownership of so few has a capture problem, not a creation problem — and that “but look at the bright spots” is, on inspection, the strongest evidence for the prosecution, not the defence. The crop is magnificent. The lease is the issue.

Run the roll-call and the pattern is total. Spotify, Sweden’s streaming champion, is incorporated in Luxembourg, listed in New York, and counts China’s Tencent among its shareholders. Supercell, the Finnish studio behind Clash of Clans, was bought by Tencent for some ten billion dollars. Mojang, the Swedish maker of Minecraft, went to Microsoft for $2.5 billion; Skype, the Estonian miracle that taught the world to make free calls, was bought by Microsoft for $8.5 billion and then, in December 2025, simply switched off and folded into Teams — a European household name discontinued by an American product manager, with no European in a position to object. And DeepMind, London’s crown jewel of artificial intelligence, has been a division of Google for over a decade; its Nobel-winning breakthroughs are Alphabet’s property. Every one of these is a genuine European triumph. Not one of them is European-owned.

And ownership aside, even the champions Europe still nominally holds are tethered. ARM, the Cambridge chip-design house inside nearly every phone, is SoftBank-owned and Nasdaq-listed; Wise and Revolut, Britain’s fintech stars, run on American venture capital and answer increasingly to American exchanges and IPO timetables; and Mistral, the great hope of European AI, is valued at roughly a fortieth of its American rival Anthropic, rents all its compute from Nvidia, and was rescued into relevance by a stake from ASML — one European chokepoint propping up another. The bright spots are real. They are also, almost without exception, working someone else’s land on a lease they did not write.

The newest tenant farm is the most telling, because Europe built it on purpose as a champion — and, as we have already seen, even Mistral is a chain of dependence: a European model trained on American chips, part-owned by a European monopoly that itself operates under American export law. There is no disgrace in any single link — each is the rational move on the board as it stands. But see the whole clearly: Europe’s sovereign-AI champion is sovereign in branding and tributary in fact.

The most expensive lesson of all is being taught right now in pharmaceuticals, and it is worth watching because it is the thesis at full speed. Novo Nordisk, the Danish maker of Ozempic and Wegovy, more or less invented the modern obesity-drug category — a genuine scientific triumph that briefly made it Europe’s most valuable company. Then the American challenger, Eli Lilly, arrived with deeper pockets, a louder marketing machine and the full weight of the US health system behind it, and out-scaled Novo in its own invention. Through 2025 Novo’s market value roughly halved; in 2026 Lilly’s pushed toward a trillion dollars. Europe discovered the molecule. America is keeping the market.

The same drama plays out one tier down, in the biotech that ought to be Europe’s natural stronghold. There is real consolidation — Denmark’s Genmab agreed an eight-billion-euro purchase of the Dutch antibody firm Merus in 2026, and Germany’s BioNTech absorbed its compatriot CureVac — proof that Europe can, when it chooses, build scale inside its own borders. But the capital market underneath still drains the other way: of the European-domiciled biotechs that went public over the past six years, all but one listed outside Europe, and when Belgium’s Agomab reached the public markets in 2026 it did so on Nasdaq, in dollars, like nearly everyone before it. Europe can invent the science and even, increasingly, own the mergers — and still watch the moment of capitalisation, the part where the wealth is actually minted, happen on an American exchange.

And the pattern is already pre-installed in the next generation. Wayve, the British self-driving-AI company widely held to be Europe’s best, raised its mega-round not from European funds but from SoftBank, Microsoft and Nvidia — which means that if it succeeds, the upside compounds to Tokyo and Redmond and Santa Clara, and if it ever needs an exit, there is no European acquirer remotely large enough to be the buyer. The bright spots are not counter-evidence to the argument. Examined honestly, one after another, they are the argument: world-class invention, foreign capture, on repeat, and faster each cycle.

Take the single most celebrated case and look at it without flinching. DeepMind was founded in London in 2010, the most important artificial-intelligence laboratory Europe has ever produced; in 2014 it was sold to Google for a few hundred million dollars and has been a division of an American hyperscaler ever since. Its crowning achievement, AlphaFold — which cracked the fifty-year problem of predicting protein structures and won its leaders a share of the 2024 Nobel Prize in Chemistry — is, as intellectual property, Alphabet’s. A British lab made a discovery that will reshape biology for a century, and the value of it compounds in Mountain View. There is no prouder European achievement in modern AI, and no cleaner illustration of the thesis: the genius is European, the ownership is not.

Two more cases sharpen the point from opposite ends. Elastic, the Dutch-founded search company, listed on the New York Stock Exchange in 2018 partly to escape the gravity of the American cloud giants — whereupon Amazon simply forked its open-source code into a rival product, “OpenSearch,” and captured much of the market the Dutch company had created. And Adyen, the rare European champion that stayed home — Euronext-listed, Amsterdam-headquartered, with the fattest margins in global payments — still processes only around three per cent of global payment volume, hemmed permanently beneath Visa and Mastercard. List in New York and the platforms eat you; stay in Amsterdam and you hit a structural ceiling. The genius is real on both paths. The ownership of the future is on neither.

Consider the Italian software consolidator Bending Spoons, which built itself from a forty-thousand-dollar seed into the owner of Evernote, Vimeo, AOL and Eventbrite, partly by rewriting its products with AI — by early 2026 more than ninety per cent of its code was machine-authored. For years it stayed defiantly private, its founder arguing that European firms could not command American valuations on public markets. Then, in June 2026, it filed to go public anyway — on Nasdaq, at some nineteen billion dollars, not in Milan or on any European exchange. Even the rare Italian champion that built something genuinely large concluded, when the moment came to cash the achievement in, that the only serious place to do it was New York. The gravity is not subtle, and it does not spare even the ones who tried hardest to resist it.

And the tether does not always run west. Some of Europe’s most prominent media and tech assets are bound, increasingly, to Chinese capital. Tencent holds around a fifth of Universal Music Group, the Amsterdam-listed company that is the largest record label on Earth; Spotify and Tencent swapped minority stakes years ago, lacing Europe’s streaming champion into a dependence on Beijing-aligned capital and on Chinese goodwill for its largest Asian market; Prosus, the Dutch investment giant that aspires to be Europe’s most valuable tech company, draws much of its worth from a roughly quarter-stake in Tencent. So even the European champions that escaped American gravity often did so by drifting into a different orbit. The point is not that China is the villain in America’s place. It is that a continent which does not own its own platforms ends up owned, in pieces, by whoever does.

There is one category where Europe genuinely held a giant, and its predicament is its own kind of warning. SAP, the German business-software titan, is the most valuable technology company in Europe, and Dassault Systèmes, Schneider’s software arm and Sweden’s Hexagon are real, durable, world-class firms in industrial and design software. But watch what is happening to them now: as the business shifts to AI-driven, cloud, subscription and “token-based” pricing, their comfortable legacy margins are compressing, and the frontier of that shift — the models, the clouds, the chips it all runs on — is owned by the Americans. Even Europe’s software champions are being pushed from owning the product toward renting the platform it must now live on. SAP’s own answer, a twenty-billion-euro sovereign cloud built to stay interoperable with the US hyperscalers, captures the bind exactly: the best European software company can build a sovereign alternative, but it must keep it plugged in to the very giants it is meant to provide an alternative to.

Pharmaceuticals deserve a closer look, because they are the field where Europe’s scientific pedigree runs deepest and the erosion is therefore most telling. Europe’s share of the world’s manufacturing of active pharmaceutical ingredients — the molecules that make a drug a drug — collapsed from around 63% in 1981 to about 6% today, almost all of it ceded to China and India, so that the continent that invented modern medicine now imports the substance of it. And the capital pattern is the familiar one: European biotech raised perhaps €25 billion in venture funding over a decade against more than €200 billion in the United States, and of sixty-seven European biotechs that went public over six years, sixty-six listed abroad. Europe still does world-class biology. It simply manufactures it elsewhere, funds it elsewhere and lists it elsewhere — keeping, as ever, the discovery and exporting the value.

Even the consolidation plays tell the story. Zalando, Europe’s biggest fashion-e-commerce company, spent over a billion euros buying its rival About You in 2025 to build the scale to resist Amazon — and still leads only the narrow niche of cross-border fashion while Amazon leads the broad market in all five of Europe’s largest economies. Prosus, the Dutch giant that owns much of the European delivery and classifieds landscape, is in truth a holding company whose worth rides on a quarter-stake in China’s Tencent. The pattern even at the top of the European table is consolidation around the American and Chinese platforms rather than displacement of them — European champions merging with one another to hold a defensible corner of a market whose centre belongs to someone else. It is a rational survival strategy. It is not the behaviour of an industry that expects to lead.

And when these champions do reach the public markets, their own financial statements quietly confess the ceiling. Spotify, for all its seven hundred million users, hands roughly 70% of its revenue to the American music labels that own the rights, capping its margins permanently beneath them. Klarna, having listed in New York, trades down more than half from its offer price, valued by the market not as a rocket but as a mature payments processor. Zalando is worth three-quarters less than at its 2021 peak. The market is not being irrational; it is pricing a structural truth these companies live daily — that a European champion sitting beneath an American platform, an American rights-holder or an American exchange captures only the slice of value the layer above it chooses to leave. They are real, profitable, admirable businesses. They are also, every one of them, tenants reading the terms of a lease they did not write.

Which points at the input underneath all the others, the one Europe is most lavishly endowed with and most carelessly exports: not capital, not companies, but people. The next chapter is about the talent — the brains Europe trains at its own expense and then rents out to the firms that will use them to buy the rest.

Strip the savings, the companies and the machine away and you reach the input underneath all of them: people. This is the resource Europe is most lavishly endowed with — it educates, freely or nearly so, some of the finest scientific and engineering minds on the planet — and it is the one it exports most carelessly of all. Europe runs, in effect, the world’s finishing school: it takes in the raw talent, polishes it to a global shine at public expense, and then watches it graduate to be paid, patented and capitalised somewhere else. You cannot lose a capital war and a talent war and expect to win anything in between.

The mechanism is almost insultingly simple: money. A senior AI researcher in a European hub earns total compensation somewhere around €120,000 to €180,000 — good money, a fine living. The same person at an American frontier lab earns $600,000 to over $1,000,000. Not a premium; a multiple — three to five times, and far more at the very top. There is no mystery about why the best leave; the only mystery is why anyone expects them to stay. Move the slider below and watch the paycheck multiply as the researcher crosses the Atlantic — same person, same skills, a different zero.

So they go, and the tide is measurable. Europe’s net inflow of tech talent — the balance of who arrives versus who leaves — halved in two years, from about +52,000 in 2022 to +26,000 in 2024, and on the AI frontier Europe remains a net exporter to the United States. The density gap underneath is even starker than the flow: the top American talent hubs run something like five to ten times the concentration of AI talent of the best European ones — San Francisco at roughly 23.9 specialists per thousand workers against Ireland’s 4.2. Europe’s strength is breadth, a thin even layer across a continent; America’s is depth, a few places so dense they reach escape velocity. Scrub the tide below.

And the loss is not only the people; it is what the people would have made. Here the create-versus-capture gap that haunts every chapter of this essay shows up in its purest form. Europe produces something like 17 to 20% of the world’s most highly-cited research — it writes the papers the future is built on. It then captures roughly 5% of the world’s AI patents. The ideas are European; the intellectual property is not. The same asymmetry runs through the founders: of the European company-builders who pulled up stakes and relocated, around 57% moved to the United States. Europe teaches the class and America hands out the diplomas that pay. Watch the cohort flow through the finishing school below — in as ideas, out as someone else’s product.

Europe is the only education system in the world generous enough to train the global elite for free and humble enough to let someone else collect the tuition — in patents, in paychecks, in the companies its graduates build on the other side of an ocean.

There is, for once, a flicker of better news in this chapter, and honesty requires reporting it. The drain is slowing — partly because America has spent 2025 and 2026 making itself harder to move to, with visa friction and rising costs, and partly because Europe has finally started to fight. The EU’s “Choose Europe for Science” initiative, launched at around €500 million in 2025, has grown to roughly €900 million across some 101 national and regional schemes by early 2026, and it is drawing interest — applications from senior non-EU researchers have surged. For the first time in this essay, a door is visibly being propped open rather than left to swing shut. Whether it is wide enough, and whether it stays open past the next budget cycle, is exactly the test of nerve the whole essay is about.

The fight is real at the institutional level too, and worth naming. France’s Inriahas begun offering early-career AI chairs worth around €1 million over four years; the Max Planck Society stood up a dedicated AI network with guaranteed compute and tenure-track posts; the European Innovation Council put some €1.4 billion into its 2026 programme with a new talent scheme attached. These are serious, well-designed instruments. But they compete on the same soil as the poachers: in 2026 Anthropic alone opened six European offices, including its first in London, and the American labs no longer need to lure Europe’s researchers across an ocean — they can hire them down the street, on a European address, at an American package. The counter-offensive is genuine; it is also outspent and out-equitied by rivals who have simply moved into the building.

But a retention scheme, however welcome, treats a symptom. The disease is that Europe has built an economy where the rational move for its most talented person is to leave — where the pay is abroad, the capital is abroad, the scale is abroad, and the only thing reliably at home is the training that made them worth poaching. Fix the pay and the capital and the scale, and the talent stays without being begged. Beg the talent to stay while leaving the pay, capital and scale broken, and you are bailing a boat without finding the hole.

Put names to it and the abstraction turns concrete. Jan Leike, who had led OpenAI’s superalignment work, moved to Anthropic in 2024; Yann LeCun, the French Turing laureate who built Meta’s AI research lab and personally anchored its Paris outpost, left in late 2025 to start his own world-model company. The pay gap underneath these moves is not subtle: a senior AI engineer who commands €120,000–180,000 in Western Europe can earn two to four times that, often far more once equity is counted, at a US lab. And the equity is the point — the European salary is a wage, the American package is a claim on the upside, and a decade of upside is what compounds into a house, a fortune, or the seed of the next company.

The cruelest detail is geographic. The American labs no longer need Europeans to emigrate; they simply open an office down the road. Anthropic hired scores of staff in Ireland inside a single year and now runs research and operations out of London, Dublin, Zurich, Paris and Munich — European talent, European cities, American payroll and American equity, the continent reduced to a staging ground for someone else’s firm. Europe’s answer, the “Choose Europe for Science” programme launched in 2025, put about five hundred million euros on the table to lure researchers home — a real gesture, and roughly a fortieth of what the bloc proposes to spend on the AI gigafactories those researchers would work in. We will fund the machines. We are still pricing the people as an afterthought.

The trend line is the alarming part, and it has already been measured: that net inflow, as we saw, has roughly halved in just two years, as the United States, Canada and Australia pulled harder while Europe’s own friction — visa drag, equity gaps, regulatory uncertainty — pushed. A continent that educates its workforce at public expense and then watches the net flow run the wrong way is not suffering a talent shortage. It is quietly running a talentexport business, and booking the loss as someone else’s gain.

Density is its own kind of destiny, and it is worth dwelling on why the concentration gap measured earlier — the top American hubs running five to ten times denser than the best European ones — matters so much more than the bare ratio suggests. Density is not vanity; it is the mechanism. It is how a researcher hears the unpublished result, how the second start-up gets founded by the first one’s alumni, how capital and talent and ambition collide often enough to ignite. Europe’s talent is real but diffuse, spread thin across twenty-seven national systems and as many languages; America’s is concentrated to the point of criticality. You can hold the same number of brilliant people and get a fraction of the output, simply because they are too far apart to catch fire.

The pipeline has a postcode. For two decades the route ran from Europe’s great university towns — Cambridge, Oxford, Zürich, Paris, Amsterdam — to a handful of square miles around San Francisco Bay, and the AI era has only widened it. The pattern is so consistent that scholars have a name for the arrangement: the research colony, a territory that supplies the raw intellectual material — the doctorates, the papers, the foundational ideas — to an imperial core that turns them into products, profits and equity. Europe educates the mind at public expense; America rents it, captures its output, and books the returns. It is a colonial relationship inverted into the knowledge economy, and the most uncomfortable part is how willingly, and how cheaply, the colony exports its most valuable crop.

It would be too easy to make this only about money, and the deeper truth is worse for Europe. Ask the researchers who leave why they went, and the pay gap is rarely the first answer. They went because that is where the work is — the frontier models, the near-unlimited compute, the colleagues operating at the absolute edge, the sense that the most important problems in the field are being solved in that building and not this one. A continent can sometimes close a salary gap with a grant. It cannot close an ambition gap with one, because ambition follows the frontier, and the frontier followed the capital and the compute across the Atlantic years ago. Europe is not only out-paying its talent; it is, more damagingly, out-inspiring it.

The machinery of retention is broken at both ends. Europe trains an enormous share of the world’s science and engineering talent in superb public universities — and then makes it administratively harder to keep a brilliant non-European graduate than the United States makes it to recruit one, while offering a fraction of the pay and almost none of the equity upside. A doctorate earned in Delft or Zürich or Munich is a globally portable asset, and its holder faces a calculation Europe loses at every term: more money, more compute, more ambitious colleagues, a faster path to a green card, all on the far side of the Atlantic. The continent subsidises the most expensive part of the pipeline — the education — and then declines to compete for the cheap part, the keeping. It is the most expensive imaginable way to run a talent policy: pay to grow it, pay nothing to hold it, and then call the loss a brain drain as though it were the weather.

The poaching is not abstract; it has names and dates. In June 2025 Meta hired away three of the researchers who had built OpenAI’s Zürich office — Lucas Beyer, Alexander Kolesnikov and Xiaohua Zhai, themselves recruited out of Google DeepMind’s Zurich lab months earlier. Anthropic, opening its own Zurich hub, hired the DeepMind veteran Neil Houlsby to run it. The American labs now wage their talent wars inside Europe, trading European researchers among themselves on European soil, with European institutions merely the venue. And the tax code holds the door open: a stock-option grant that a US engineer eventually pays capital-gains rates on, around fifteen to twenty per cent, a German engineer pays ordinary income tax on at exercise, north of forty — so even the equity meant to bind talent to a European startup is taxed as though it were a salary. Europe penalises the one instrument that might let it compete.

The pattern even shapes Europe’s rare successes. Arthur Mensch, the founder of Mistral — the closest thing the continent has to a frontier AI champion — did not build it straight out of a European lab. He spent years at Google DeepMind in Paris first, learning the craft inside an American hyperscaler, before leaving in 2023 to start something French. So even the bright spot runs on talent Europe first had to lose to the Americans and then win back; the training that mattered most happened on a US payroll, on European soil, in a lab Europe does not own. The continent can, occasionally, reclaim its people — but it does so one founder at a time, against a tide running the other way by the tens of thousands, and then calls the rare reversal a triumph rather than the exception that measures the scale of the loss.

And then, in 2025, something genuinely new happened: for the first time in a generation the tide briefly slackened — not because Europe grew more attractive, but because America grew less so. As the new US administration cut research budgets and leaned on universities, applications from US-based scientists to European programmes surged: France’s hastily-launched “Safe Place for Science” drew hundreds of American applicants in its first round, the European Research Council reported a sharp jump in interest from across the Atlantic, and Europe found itself, improbably, a refuge. It is the essay’s most hopeful accident: proof that talent flows toward stability and freedom and funding, and that these are things Europe, when it chooses, can still offer better than almost anyone. The window is real. Whether Europe funds it at the scale required, or lets it close again as the American panic subsides, is exactly the test of nerve this essay keeps returning to.

The composition of the workforce tells its own story. Europe’s tech hubs increasingly run on imported talent — by one measure roughly a quarter of Ireland’s AI workforcedid its undergraduate degree in India — even as the continent’s own graduates leave for America; Europe is, in effect, a transit lounge, drawing skilled people in from the developing world and losing them onward to the United States. And the pipeline is narrow where it should be widest: women hold only about a quarter of AI roles worldwide, and Europe does no better, leaving half its potential talent under-tapped at the exact moment it can least afford to. A continent serious about a talent shortage would be widening every intake it has. Europe treats its human capital the way it treats its financial capital: abundant, world-class, and quietly allowed to flow out faster than it is built up.

The gap shows up before the first job, in what graduates even imagine doing. Only about 15% of European PhDs go on to found or join a startup, against something like 45% in the United States; the European doctorate still points, by default, toward the university, the public lab or the large incumbent, not the risky new company. And when Europeans do try to build, roughly seven in ten founders name the regulatory environment as a brake on expansion. So the loss compounds at every stage: a system that trains brilliantly, channels its brilliance away from enterprise, taxes the equity that would reward it, and regulates the scaling that would retain it — and then wonders where its founders went. They went to the place that does the opposite of all four things, and felt, on arrival, as though they had been let out of a smaller room.

And the loss is not evenly spread; it hollows out whole regions. Dozens of European areas — much of Bulgaria and Romania, swathes of southern Italy and Portugal — are now caught in what demographers call a “talent development trap”: their young and educated leave for the richer cities or for America, the population ages and shrinks, and the conditions that might have kept the next cohort decay further with each departure. Across the EU, youth unemployment sits around 15%, more than double the adult rate, even as firms complain they cannot find skilled workers — the paradox of a continent that cannot match its people to its opportunities because too few of the opportunities are being created at home. A brain drain, seen up close, is not an abstraction in a productivity table. It is a clever young person in Cluj or Catania doing the arithmetic of their own life and concluding, reasonably, that their future is somewhere else.

The next chapter is about the most physical hole of all — the one you can measure at the electricity meter. Because a continent can train the engineers and bank the savings and still fail to build anything, if it has priced the electrons that everything runs on out of reach.

You can train the engineers, bank the savings, own the machine, and still build nothing — if the electricity that everything runs on costs two or three times what your rivals pay. This is the most physical chapter in the essay, the one you can read off a meter, and it is in some ways the most damning, because energy is not a matter of culture or nerve or cap tables. It is arithmetic. A factory either makes money at the local power price or it does not, and across much of European heavy industry, increasingly, it does not. The same failure shows up twice — as the old economy switching off, and as the new one quietly choosing to build somewhere else.

Take the old economy first, because the damage is already booked. European industrial electricity runs at roughly 2.5 times the US price — about €0.199 per kilowatt-hour against €0.075 — and around 2.4 times China’s. Gas is worse: the European wholesale benchmark has traded near five times America’s. For an industry like chemicals, where energy is the feedstock, that is not a headwind; it is a verdict. Europe has lost something like 37 million tonnes of chemical capacity since 2022 — close to 9% of the total — with output down 11 to 15% from pre-crisis levels and on the order of 109,000 jobs gone or at risk. Nearly half of the announced closures cite energy costs as the reason. The new economy’s ledger is the mirror image: the AI build-out needs vast, cheap power, and a continent whose electricity costs double forecloses on the data centres before they are drawn — which is why Europe’s AI capital spending runs a fraction of America’s. Toggle the two ledgers below; they are charged to the same account.

The single most eloquent fact in the chapter is corporate, not statistical. BASF, the largest chemical company in the world and for a century the beating heart of German industry, inaugurated an €8.7-billion integrated Verbund site in Zhanjiang, China, in March 2026 — even as it closed plants at its historic home in Ludwigshafen, which has been posting billion-euro operating losses. A company does not move its century-old centre of gravity to the other side of the planet on a whim. It does so because the arithmetic at home stopped working, and the most German company there is voted with ten billion euros of capital. When BASF leaves, it is not a data point. It is a tolling bell.

You can stack the disadvantage up and read it as a single number: the tax a producer pays simply for being located in Europe rather than on the US Gulf Coast. Start at an American baseline and add the electricity premium, then the gas premium, then the carbon price, then the regulatory load, and the cost of European location climbs well above parity — not because European firms are inefficient, but because the ground they stand on is dearer. Build the waterfall yourself below. It is the clearest answer to the question Europe keeps not asking out loud: why would anyone build the next plant here?

And then, as if to remove any remaining doubt, there is the grid — because even the company that wants to build in Europe, and can afford the power, increasingly cannot get plugged in. In the Netherlands, the densest, richest corner of the continent, grid congestion has become an absolute brake: something like 47 gigawatts of connection requests sit on waiting lists — over fourteen thousand of them — with waits stretching to ten years. The national grid operator has warned, flatly, that the network will fail to meet electricity demand by 2030; the Randstad around Amsterdam and Schiphol is effectively frozen to the mid-2030s. A factory, a data centre, a newly-electrified production line, a housing estate — all of them now join the same queue, and the queue is measured in years. Add demand below and watch it freeze.

Europe priced its electrons out of reach, and then ran out of wires to carry the ones it has left. You cannot reindustrialise a continent that can neither afford the power nor connect the plug.

It is worth being precise about what is and is not Europe’s fault here. The 2022 gas shock was Russia’s doing, not Brussels’. But three years on, the persistence of the gap is a policy outcome, not an act of God: it reflects choices about nuclear, about LNG infrastructure, about grid investment, about whether to shield industry from the full force of carbon and energy pricing or let the market clear by closing the factories. America had its own gas; it also chose, through the Inflation Reduction Act, to pour money at the problem with brute coordinated force. Europe has the engineers to build the grid and the capital to fund it — the same €33 trillion from the capital chapter — and has so far chosen to let the connection queue grow to ten years instead.

And the Dutch queue is only the sharpest edge of a continental problem. Across the EU something on the order of 1.7 terawatts of would-be generation — most of it wind and solar, more than the entire capacity installed on the system today — sits in connection queues waiting for wires that do not yet exist. In 2025, for the first time, wind and solar together out-generated fossil fuels across the EU’s grid — a genuine milestone. And yet the cruelty of the bottleneck is exactly that: the clean, cheap power Europe has actually managed to build increasingly cannot reach the factories and data centres that need it, because the grid to carry it was never built alongside. It is the capital paradox again, written in copper — a continent rich enough to fund anything that has somehow under-invested in the one piece of infrastructure, the wires, on which every other ambition (electrification, reindustrialisation, AI) now physically depends.

BASF is the loudest case, not the only one. ArcelorMittal moved to idle blast furnaces at Florange in France and Liège in Belgium, putting thousands of steel jobs at risk and citing power costs running past twenty euro-cents a kilowatt-hour. Yara, the Norwegian fertiliser giant, mothballed around two million tonnes of nitrogen capacity at Herøya and Kårstø after a four-fold electricity-cost spike and shifted production to the US Gulf and Trinidad, where power runs nearer six cents. The pattern is not weakness; it is arithmetic — and Europe stacks on top of it one cost its rivals do not pay at all: the Emissions Trading System carbon price, around ninety to a hundred euros a tonne, which adds a further eighteen to twenty-five euros per megawatt-hour to a European smelter’s power bill that a Texan or Chinese competitor simply keeps as margin.

And it is no longer only the energy-hungry chemicals and metals; the malaise has reached the machine-makers at the very core of the European model. In Germany — the continent’s industrial heart — Volkswagen is cutting on the order of 35,000 jobs by 2030 and has begun shifting Golf production to Mexico; Bosch, the world’s largest car-parts maker, has announced around 22,000; ZF and Continental tens of thousands more between them. The chemical sector is in its fourth consecutive year of crisis, and the single most alarming data point is a survey finding: more than half of German industrial firms with over five hundred employees now say they are weighing moving production out of the country. This is not a downturn that ends with the cycle. It is the slow relocation of the continent’s manufacturing base toward places where the power — and increasingly the future — is cheaper, happening to the one country that was supposed to be immune to it.

The new economy tells the same story from the opposite direction. The AI build-out is, at bottom, a bet on cheap power, and the numbers have become almost comically lopsided: American hyperscalers have announced something on the order of four hundred gigawatts of new AI-compute capacity by 2030, against a European total well under forty. The reason is the meter. A data centre paying European industrial power prices — two to three times American ones — is uneconomic before the first server is racked, which is why no European hyperscaler has announced a major campus outside North America or the handful of cold, cheap-power corners like Iceland and Norway. Europe priced itself out of the old economy’s furnaces and the new economy’s data centres with one and the same tariff.

Behind the frozen connection queues sits a number that explains why they will not thaw soon. Europe’s own estimates put the grid investment it needs this decade at something like €584 billion, with some analyses pushing past a trillion once the full electrification of heat, transport and industry is counted — and the actual build-out is running far behind. Germany has completed only a fraction of its planned high-voltage transmission lines; the Netherlands is rationing connections in its richest region. This is the capital chapter again, in a hard hat: the money to build the grid is the same idle savings from Chapter Three, and the decision not to mobilise it at speed is the same failure of nerve — only now it shows up as a substation that does not exist and a factory that therefore cannot.

The bitterest detail is where the crunch bites hardest. Brainport, the Eindhoven region that hosts ASML and the densest concentration of high-tech industry on the continent, is itself running up against the limits of the local network — the single most strategically important industrial cluster in Europe, the one that builds the machine the whole digital world depends on, rationing its own ability to expand because the wires are full. More than ten thousand businesses sit in the Dutch connection queue behind it. When the country that makes the world’s most advanced technology cannot reliably plug in its own most advanced factories, the problem has stopped being an energy-market quirk and become a sovereignty question in its own right.

The single most painful loss is the one that was supposed to be the answer. Northvolt, the Swedish battery-maker, was Europe’s great hope to build a homegrown champion in the one industry on which the entire electric transition depends — and in March 2025 it filed for bankruptcy under roughly €5.8 billion of debt, the largest industrial collapse in modern Swedish history, its half-built gigafactory in the Arctic north left for others to pick over. Around it the older economy keeps closing: Speira shuttered primary aluminium smelting at Neuss; Tata Steel moved to close coke and blast-furnace lines at IJmuiden; Yara earmarked an ammonia plant in Belgium for closure. And the new economy cannot move in to replace them, because AWS and its peers now face grid-connection waits of up to seven years for new European data centres. The factories that leave do not come back, and the ones that might replace them cannot get plugged in.

And a hard truth sits underneath the prices: much of the gap is self-inflicted policy, not fate. France, which kept and kept building nuclear power, enjoys some of the cheapest, cleanest electricity in Europe; Germany, which shut its last reactors in 2023 in the middle of an energy crisis, locked in a deeper dependence on gas and a higher price for a generation. The continent that frets about strategic autonomy switched off its most strategically autonomous source of power on principle, then watched its chemical industry decamp over the resulting bills. Energy is the chapter where the verdict is least about American strength and most about European choice — which is also what makes it, in theory, the most fixable, if the nerve to reverse the choices could be found.

Two 2025 decisions capture the bind from both directions. ArcelorMittal cancelled some two-and-a-half billion euros of green-hydrogen steel projects in Germany and walked away from more than a billion in German subsidies, judging that clean steel simply could not pay at European power prices — the green transition foreclosed by the very energy bill it was meant to drive. And the AI build, when it does reach Europe, goes where the power is cheap: OpenAI’s Stargate programme is siting a hundred-thousand-GPU data centre not in the industrial heartland but in the far north of Norway, drawn to its abundant, cheap hydropower. Europe’s electrons have become a sorting mechanism: heavy industry priced out, AI compute pushed to the cold edges, and the populous, productive core left paying the most for the least.

The cruel twist is that the answer is proven and Europe has simply forgotten how to deploy it at speed. Finland’s Olkiluoto 3 and France’s long-delayed Flamanvillereactor both finally came online this decade and now pour cheap, clean, firm power into their grids — proof that the technology works. But Britain’s Hinkley Point C has slipped toward 2030 at a cost ballooning past forty-five billion pounds, a monument to how thoroughly the West has lost the muscle memory of building big things on time. Europe’s energy problem, in the end, is its capital and talent problems wearing a third costume: not an absence of the answer, but an inability to commit to it at scale, on schedule, against the short-term objections — the failure of nerve, rendered in concrete and cooling towers.

Nowhere is the industrial squeeze more visible than in the country that was supposed to be immune to it. Germany, the manufacturing heart of Europe, ground through a third straight year of stagnation in 2026 with the grim furniture of decline piling up: Volkswagen shut a German factory for the first time in eighty-eight years; Bosch announced twenty-two thousand job cuts; Thyssenkrupp, the storied steelmaker, agreed to shed some eleven thousand jobs — forty per cent of its steel workforce — and lined up a sale to an Indian group. German unemployment crossed three million for the first time in over a decade, and corporate insolvencies hit their highest level in years. The proximate causes are the ones this chapter has catalogued — energy at well over twice American prices — compounded by a second shock from the opposite direction: Chinese electric cars, led by BYD, eating the home market of the very industry that defined modern Germany. The country that was the rebuttal to every declinist argument about Europe is now the argument’s leading exhibit, and it got there by the same machine: priced out of energy, out-built on capital, out-scaled on the technologies that will define the next economy.

Watch how the Americans solved the same power problem, because the contrast is total. Unable to wait for the grid, the US hyperscalers simply bought their own nuclear plants: Microsoft signed a deal to restart a reactor at Three Mile Island to feed its data centres; Amazon struck a multi-billion-dollar agreement for the output of the Susquehanna nuclear station. When a company that size needs a gigawatt of firm power, it writes a cheque and gets one. A European firm in the same position joins a connection queue that now runs seven to thirteen years in Dublin, Frankfurt, London, Amsterdam and Paris alike. The difference is not technology or even, ultimately, money — Europe has both. It is the capacity to act: to build, to buy, to clear the path, at the speed the moment demands. The Americans are powering the AI age off the grid by sheer force of will. Europe is waiting in line for permission.

And the men who run German industry have stopped using diplomatic language. Peter Leibinger, head of the powerful federation of German industries, called the moment Germany’s “deepest crisis since the founding of the Federal Republic”; the billionaire industrialist Reinhold Würth warned that the country was caught in a “downward spiral of deindustrialisation.” These are not activists or professional declinists; they are the owners and chiefs of the very firms the energy crisis is hollowing out, and they are saying, in public, that the post-war German economic model is breaking. When the people with the most to lose and the most reason for optimism begin talking like this, the polite official insistence that the slump is merely cyclical becomes very hard to credit.

The closures have not slowed; if anything they have become routine enough to stop making headlines. BASF is shutting eleven plants at Ludwigshafen alone, including ammonia and TDI lines, as part of more than a billion euros of annual cost cuts; INEOS is closing chemical units in Germany; the cumulative European chemical capacity lost since 2022 has passed 37 million tonnes, nearly a tenth of the total. And the energy dependence that drove it has, if anything, deepened: having replaced Russian pipeline gas, Europe now imports the majority of its liquefied natural gas from the United States — some 63% of it, heading toward eighty — trading one external dependence for another, at a price that keeps its industry uncompetitive either way. The continent swapped a hostile supplier for a friendly one and called it security; what it actually bought was the same vulnerability with a better-tempered landlord.

The one country making a serious, sustained bet is, again, France, which committed some €73 billion to build six new EPR2 reactors — a multi-decade wager on cheap sovereign power that is exactly the kind of patient constancy the rest of the continent struggles to muster. Set against it is the counter-signal: Fluidstack, an AI-cloud company, decamped from Europe to list in the United States at an eighteen-billion-dollar valuation, taking its compute build with it. The two facts frame the whole chapter. Cheap, firm, sovereign power is buildable, and France is building it; but absent that power, and the speed to connect it, the AI economy that runs on electrons simply forms up somewhere else. Europe’s energy choices are not only about industry’s past. They are about whether the future plugs in at home or abroad.

The energy chapter, then, is the capital chapter and the talent chapter in physical form: a continent that has the resources to solve a solvable problem and keeps declining to spend the nerve. The next chapter turns to the place where Europe is now spending money fastest of all — defence — and finds, depressingly, the same machine running underneath: a historic surge of European money that ends up, by a different route, buying American.

For the first time in a generation, Europe is spending on defence like it means it. Spooked by Russia and by an American administration that has made the old guarantees feel conditional, NATO allies pledged at the 2025 Hague summit to reach 5% of GDP on defence by 2035, and the EU wrapped a roughly €800-billion “ReArm Europe / Readiness 2030” ambition around the surge. After thirty years of under-spending, this is the right instinct, arguably the most decisive collective move the continent has made in the whole essay. And it is, by the machine’s now-familiar logic, about to become the largest transatlantic wealth transfer of the decade. Europe is re-arming. It is mostly buying American.

The numbers are not subtle. The United States supplied 64% of European NATO members’ arms imports over 2020–2024 — up sharply from 52% in the prior five years. At the height of the post-invasion scramble, the Draghi report found that European governments spent some €75 billion on defence equipment with about 78% going to non-EU suppliers and roughly 63% of that to the United States. So the historic surge of European money, the one meant to buy European autonomy, drains across the Atlantic almost as fast as it is appropriated — into American jets, American missiles, American systems. The EU knows it, and has set targets to claw the share back (no more than 45% bought outside the EU by 2030, 40% jointly procured by 2027). Whether the targets survive contact with the urgency is the open question. Trace the leak below, and pull the “buy European” lever.

The flagship purchase is also the sharpest illustration of why the dependence matters beyond the money. European NATO states have ordered something like 630 to 670 F-35 fighters — the finest combat aircraft in the Western inventory, and also one whose continued operation runs through American hands. There is, the Pentagon insisted in March 2025, no literal “kill switch” — no secret line of code that bricks the jet. But there does not need to be. The spare parts, the mission-data files, the ODIN software-update pipeline, the maintenance — all of it flows from the United States, and any of it can be slowed. As defence analysts put it through the 2025 controversy, logistical control is operationally equivalent to a switch; you do not need to disable the aircraft if you can simply decline to keep it flying. Two NATO partners, Portugal and Canada, publicly reconsidered their F-35 commitments on exactly this fear. Turn the dependency dial below and watch the fleet go dark.

Run a finger down the spine of the European defence stack and the same vertebra keeps coming up American. The aircraft are largely US. The munitions, at the surge, were largely US. And the software brain — the AI-enabled battle-management layer that increasingly runs modern war — arrived in the form of Palantir’s Maven Smart System, adopted by NATO in March 2025 and rolled out across its commands. A continent that worried about depending on American chips and American clouds is now wiring its command-and-control through an American defence-software company. Autonomy, bought on a foreign platform, is not autonomy. It is a subscription.

And yet — for the second time in two chapters, and it is worth marking — this is the place where Europe has actually started to push back. Germany rejected Palantir in April 2026 over precisely these sovereignty concerns; France, the Netherlands and Denmark are building and testing European alternatives. The defence chapter, alone among the chapters of loss, contains a genuine countercurrent: a continent that, confronted with dependence at its most existential — the ability to defend itself — has finally found the nerve to flinch. Click down the spine below; the gold marks where the resistance has begun.

You cannot buy strategic autonomy from a single foreign supplier who keeps the spare parts, the software keys, and the right to say no. That is not sovereignty. It is a very expensive lease on someone else’s permission.

The lesson of the defence chapter is double-edged, and both edges matter. The first edge is the familiar one: even when Europe finally spends — decisively, urgently, at historic scale — the machine routes the money and the control westward, and re-armament becomes one more tribute. The second edge is the hopeful one, and the whole essay turns on whether it can be generalised: that fear finally did what a decade of competitiveness reports could not. The threat of being switched off concentrated minds that the threat of being out-competed never did. Germany did not reject Palantir because of a Draghi recommendation; it rejected it because it suddenly, viscerally understood what a foreign off-switch on its own military meant.

The defence surge is, at least, beginning to grow a European industry to spend into. The EU’s SAFE instrument — a hundred-and-fifty-billion-euro loan pot agreed in 2025 — carries a rule with real teeth: at least 65% of a funded weapon’s value must come from European or allied suppliers, an explicit attempt to keep the re-armament money on the continent. Rheinmetall, the German munitions maker, expects to roughly quintuple sales toward fifty billion euros by 2030; Helsing, a Munich defence-AI company backed by Spotify’s Daniel Ek, is raising at around eighteen billion. These are real green shoots in the one field where fear has concentrated minds. But the counter-example is just as instructive: the Franco-German FCAS next-generation fighter, nine years and billions of euros in, produced no flying prototype and effectively stalled in 2026 — the old European disease of a flagship programme dissolving into national in-fighting, even as the urgency screams.

The scale of the aircraft dependence alone is worth spelling out, because it is a continent-wide commitment, not a one-off. The United Kingdom plans some 138 F-35s; Finland 64; Italy ninety-odd; the Netherlands, Norway, Belgium, Denmark, Poland and the Czech Republic filling in behind — on the order of six hundred of the jets across European NATO, each one flying on American spare parts, American software updates and American mission data. Poland took its first airframes in 2026; deliveries run well into the 2030s. This is not a procurement a country unwinds in a budget cycle. It is a thirty-year structural dependence, signed willingly, by almost every air force on the continent at once.

And the spending is real and historic: NATO’s European members pushed collective defence investment past a hundred and thirty billion euros in 2025, a record, with the pledge to reach 5% of GDP by 2035 still ahead of them. The question the chapter keeps pressing is simplywhere it lands. Some, encouragingly, lands at home — Rheinmetall’s order book, Helsing’s valuation, the SAFE content rules. But the marquee capabilities — the fifth-generation jet, the battle-management software, the precision munitions in the quantities a real war consumes — still run substantially through American suppliers, which means the largest European rearmament since the Cold War is also, in its opening years, one of the largest single transfers of European money into American defence revenue on record.

And the soft kill-switch stopped being hypothetical in March 2025, in the most literal way imaginable. For a few days, amid a diplomatic rupture, the United States suspended intelligence-sharing with Ukraine — degrading, at a stroke, the targeting and early-warning picture Ukrainian forces depended on. The feed was restored within about a week, but the demonstration was total: the most important input to a modern military can be switched off from Washington, on a political whim, with no notice. Every European capital watching understood that the same hand rests on the same kind of switch over its own American-supplied systems.

The legal architecture makes it concrete. American components inside European weapons fall under ITAR, the US arms-export regime, which hands Washington a veto over the re-export — and sometimes the very use — of any system containing US-origin technology, a constraint that has repeatedly complicated European deliveries of missiles and aircraft to third countries. So even the weapons Europe builds for itself often carry an American licence buried in the bill of materials. The continent is spending its historic defence surge to acquire capability and, baked invisibly into much of it, a foreign permission slip it cannot revoke.

And there is one corner of the defence build-out where the gap is not large but near-total. Modern war is becoming software — drones that target autonomously, battle-management systems that fuse a thousand feeds, the AI layer that increasingly decides faster than any human staff can. American firms are pouring hundreds of billions into the underlying AI; Europe’s defence-AI spending, the figure that may matter most by 2035, rounds to a small fraction of that. The continent that worried about depending on American chips and clouds is now, by inattention, on track to depend on American military intelligence software too — which is exactly why Germany’s rejection of Palantir and the rise of Helsing matter far more than their euro values suggest. They are the first refusal to outsource the brain of European defence.

And here, more than anywhere in the essay, a genuine European industry is being born in real time. A wave of defence-technology start-ups has erupted out of Germany in particular: Helsingraising at around eighteen billion dollars, Quantum Systems and the drone-maker Stark and the autonomous-vehicle firm ARX all scaling fast, many of them explicitly engineering their supply chains to be free of American components so they qualify under the “buy European” rules. The procurement decisions are starting to follow: Denmark chose the Franco-Italian SAMP/T air-defence system over the American Patriot in 2025, and the Franco-German land-systems giant KNDS lined up a twenty-billion-euro listing. None of this undoes the F-35 dependence or the software question. But it is the most convincing evidence in the whole essay that, when the fear is sharp enough and the money real enough, Europe can still grow champions of its own — which only sharpens the question of why it does so almost nowhere else.

The procurement map is starting, in places, to redraw itself. The Franco-German land-systems champion KNDS — maker of the Leopard tank and the Caesar howitzer — lined up a listing reportedly valuing it around twenty billion euros, and Germany ordered its first domestically-built tanks and artillery in a generation. France’s Rafale fighter carries an order backlog of more than two hundred aircraft, much of it export, as buyers seek an alternative to the American jet and its strings; and even American primes are adapting, building ITAR-free European production bases specifically so their kit can qualify for “buy European” contracts. The continent is discovering, under the pressure of a real war on its border, that an industrial base is not a thing you can summon in a crisis. It is a thing you either kept, or did not.

The most encouraging movement is in the unglamorous business of actually making things. Stung by the discovery that it could not produce artillery shells fast enough to sustain Ukraine, Rheinmetall is racing toward 1.5 million 155-millimetre shells a year by 2027; Norway’s Nammo is restarting mothballed lines; two dozen nations have signed up to a European air-defence initiative. And Ukraine has become, grimly, Europe’s defence-tech proving ground: a swarm of new drone makers — Germany’s Quantum Systems and Stark, Croatia’s Orqa, a dozen others — are iterating weapons in a live war at a speed no peacetime procurement system could match, and scaling toward millions of units a year. This is what an industrial base looks like when it is finally allowed, or forced, to grow. The tragedy threaded through the whole essay is that it took a war on the continent’s edge to permit it.

And beneath the spending lies a set of capability gaps so specific they are almost embarrassing. For all its wealth, Europe can deploy only a handful of its own heavy strategic-airlift aircraft and air-to-air tankers, leaning on a literal three shared C-17s and a small pooled fleet for the basic job of moving an army; its civilian Copernicus satellites see the ground at a resolution too coarse to identify a military vehicle, leaving it dependent on American imagery to know what is happening on its own frontier; and no European air force maintains a credible capacity to suppress enemy air defences — the dangerous first task of any modern air campaign — without the United States. These are not luxuries. They are the load-bearing capabilities of an autonomous military, and a continent of four hundred and fifty million people, spending hundreds of billions, still cannot field them alone. That dependence is not a line in a budget. It is the difference between an alliance and a protectorate.

Space is the starkest gap of all, because it is where the next war will be won or lost. Europe has no equivalent of SpaceX — no cheap, reusable, high-cadence launcher of its own — and so depends on American rockets even to put its own satellites up; its sovereign satellite-communications network, IRIS², is not due in service until 2029, years behind the constellations already overhead. When Ukraine’s battlefield connectivity hung, at moments, on the goodwill of a single American billionaire’s satellite network, every European capital saw the future with terrible clarity: that the most decisive military infrastructure of the age is privately owned, American, and switchable — and that Europe, for all its engineers and its money, had not built itself an alternative. The continent that launched the European space age from Kourou is now, overhead, a tenant like everywhere else.

The most basic test of a war economy is whether it can make enough shells, and Europe has spent three years discovering, humiliatingly, that it could not. At the start of the Ukraine war the entire European Union struggled to produce in a year what Russia turned out in months; only now, after a frantic build-out, is NATO’s monthly artillery-ammunition target — around 267,000 shells — drawing level with Russia’s output, and only because firms like Germany’s Diehl are doubling missile lines and Rheinmetall is building new plants. Meanwhile France, the EU’s only nuclear power, moved in 2026 to harden its own deterrent and stopped disclosing the size of its arsenal — a quiet acknowledgement that the American nuclear umbrella over Europe could no longer simply be assumed. A continent that has to re-learn how to make ammunition and re-examine who guarantees its ultimate security is a continent discovering, late and at speed, just how much of its own defence it had quietly outsourced.

And the nature of war itself is changing under Europe’s feet in a way that should terrify and encourage it in equal measure. At a NATO exercise in Estonia in 2025, a handful of Ukrainian drone operators — ten of them — reportedly defeated two NATO battalions in half a day using cheap first-person-view drones: a vivid demonstration that the expensive, exquisite platforms Europe buys from America can be checkmated by hardware costing a few hundred euros. This is, for once, a contest where incumbency counts for little and Europe’s engineering culture could genuinely excel — the drone war rewards iteration, software and cheap manufacturing, not forty-year procurement cycles. Ukraine has become both the proof and the school. The question is whether Europe learns the lesson in time to build the new kind of arsenal itself — or whether it spends its historic defence budget re-buying, at vast expense, the last war’s weapons from the last war’s supplier.

The money has at last begun to move: in January 2026 the EU’s SAFE instrument approved its first €38 billion in disbursements to eight member states, and a new five-nation consortium, LEAP, formed to mass-produce the cheap drones the Ukraine war has shown to be decisive. But the gap between spending and capability is still wide and slow to close: the flagship European Sky Shield air-defence initiative now counts some two dozen nations and, years in, still has no integrated system actually deployed. This is the recurring shape of European ambition — the announcement arrives long before the capability, the cheque clears years before the shield is built — and in defence, more than anywhere, the lag between deciding and fielding is measured in the one currency that ultimately matters: time the continent may not have.

Which raises the question the back half of this essay keeps circling: if existential fear is what it takes to make Europe act, does the economic version of the same threat — the slow, bloodless capture documented in every other chapter — register as fear at all? Or does it take a fighter jet going dark to make a continent feel what a closed factory and an emigrated founder never could? The next chapter looks at the one weapon Europe has reliably reached for instead of spending or building — the rulebook — and asks whether the drawbridge it keeps raising is keeping anyone out, or only sealing itself in.

When Europe cannot spend and will not build, it does the one thing it is still unmatched at: it writes a rule. For a decade this was sold as a superpower — the “Brussels Effect,” the idea that the EU’s vast single market lets it set the standards the whole world must follow, from privacy to AI. There is something to it. But the effect has curdled, and this chapter is about the curdling: a regulatory moat dug so deep, and policed by so many hands, that it no longer keeps the rivals out so much as it seals Europe’s own challengers in — a fortress with the drawbridge stuck down, the garrison taxed for the privilege of defending it.

The clearest evidence is the one regulation Europe is proudest of. The GDPR was a genuine moral achievement and a measurable economic own-goal: a 2025 study from the National Bureau of Economic Research found that, after it took effect, the number of EU venture deals led by US investors fell by about 21%, and EU technology venture investment dropped roughly 26% relative to the United States. The rule meant to protect Europeans from Big Tech also, quietly, made it harder for the next European challenger to Big Tech to get funded. And the pattern generalises through the compliance cost itself, because compliance is a regressive tax: the roughly €200,000 to €600,000 a high-risk AI-Act provider must spend — on top of a quality-management system and per-model assessments — is a rounding error for a hyperscaler and an extinction event for a seed-stage startup. The total bill runs to an estimated €3.3 billion a year across the EU. Scrub firm size below and watch the burden invert: heaviest where Europe can least afford it.

The burden is multiplied by the thing that should have been Europe’s greatest asset and is instead its greatest tax: the market is not single. The Draghi report counted more than 13,000 EU legal acts passed between 2019 and 2024 — against roughly 5,500 in the United States over a comparable span — and over 270 digital regulators spread across the member states (and the true volume of implementing acts is almost certainly higher). A US startup writes its software once and sells it into one market of 340 million people. A European startup writes its compliance 27 times, threads 27 national interpretations, and answers to a constellation of overlapping authorities. The single market is, for a small company with no compliance department, twenty-seven mazes wearing one name. Run a founder through it below.

And here is the confession, the moment the whole strategy looked in the mirror and flinched. Europe passed the most ambitious AI law in the world, the AI Act, with its toughest obligations for high-risk systems due to bite on 2 August 2026. Then, with the deadline in sight and some 78% of organisations reporting they had taken no compliance steps at all, Brussels reached for a lever it had never publicly admitted owning: it stopped its own clock. Through the “Digital Omnibus” agreed in May 2026, the high-risk deadline was pushed back to 2 December 2027 — a sixteen-month reprieve from a rulebook the EU had written itself. It is hard to think of a more eloquent admission that the regime had outrun the economy it governs. Advance the clock below and watch Europe drag its own deadline backwards.

None of this means the targets of the rules are sympathetic, or that the rules do nothing. When the EU fined Apple €500 million in 2025 for blocking developers from steering users to cheaper deals, and forced its App Store commissions down from the old 15–30% toward 7–10%, that was a real win for a real abuse — the Brussels Effect doing exactly what it says. But notice the asymmetry even in victory: Apple appealed, kept collecting on the vast majority of the App Store economy, grew its services revenue anyway, and absorbed the fine as a cost of doing business. The giant pays the toll and walks on. It is the challenger, the one the rules were nominally meant to make room for, who finds the drawbridge in its face.

The cost of the rulebook is most visible in the products Europeans simply do not get, or get late. Apple Intelligence was withheld from the EU at its 2024 launch over Digital Markets Act uncertainty; Meta delayed its AI assistant in the bloc for the same reason; Google geofenced its Gemini model out of several EU countries after a French-regulator challenge; and when the Chinese lab DeepSeek arrived, European data authorities moved to block it from training on EU data. Set aside whether each call was right — the cumulative signal to any company deciding where to launch is unmistakable: Europe is the market where the newest things show up last, wrapped in the most legal doubt. A continent that cannot reliably even receive the frontier, let alone build it, has quietly priced itself out of being a serious customer for the future.

The machinery behind all this is genuinely vast. Under the Digital Markets Act the Commission formally designated six gatekeepers — Apple, Alphabet, Amazon, Meta, Microsoft and ByteDance — controlling some two dozen “core platform services”, each triggering its own compliance regime, audits and interoperability duties. It is, in the abstract, a serious and in places admirable attempt to discipline genuine market power. But step back and the shape is unmistakable: five of the six firms being regulated are American, one is Chinese, and none is European — because Europe has no platform large enough to need disciplining. The continent has built the world’s most sophisticated apparatus for governing digital giants and has not produced a single one of its own to govern.

The privacy regime has its own landmark, and it cuts the same way. In 2023 the Irish regulator fined Meta €1.2 billion — the largest GDPR penalty ever — for shipping European users’ data to American servers, the climax of a decade-long saga driven by a single Austrian activist, Max Schrems, who twice toppled the legal frameworks governing EU–US data flows. It was, again, real enforcement of a real principle. And again the deeper picture is the trap: the data was crossing to American servers in the first place because the services Europeans use are American, hosted on American clouds — and the most a European regulator can ultimately do is fine the foreign company for the terms of a dependence it has no power to end. You cannot regulate your way out of not owning the infrastructure.

And there is a final twist that complicates the whole picture, because the retreat from the rulebook is being authored partly by the very giants it was meant to constrain. The Digital Omnibus simplification of 2026 — the package that stopped the AI Act’s clock and trimmed reporting duties — arrived after sustained lobbying from American big tech and its allies, who found Brussels suddenly receptive to the argument that its own rules were throttling competitiveness. So Europe spent a decade building the most ambitious digital rulebook in the world, then began dismantling parts of it under pressure from the firms it was written to discipline. Whether that is wise course-correction or regulatory capture depends on where you stand — but either way it is not the posture of a power that sets the terms. It is the posture of one that keeps revising them under pressure from somebody else.

The enforcement, when it comes, lands on American and Chinese platforms because those are the only ones big enough to enforce against. In late 2025 the Commission fined X some €120 million under the Digital Services Act over its “blue check” design and ad-transparency failures; in early 2026 it moved against TikTok for addictive design, the first action of its kind. These are, once more, real attempts to govern genuine harms. But notice the shape one final time: the regulator is European, the regulated are not, and the very existence of a sophisticated enforcement apparatus with no domestic platforms to enforce against is the cleanest possible statement of the chokepoint paradox. Europe has made itself the world’s referee in a game it no longer fields a team in.

The newest rules sharpen the irony to a point. The AI Act’s toughest tier, its obligations for “systemic-risk” models, is triggered by a training-compute threshold so high that essentially only American models — GPT-class, Gemini, Claude — cross it; Europe is, in effect, writing the operating manual for machines it does not build. Meanwhile the compliance bill lands hardest on the small: cookie-consent and data rules can cost a European SME tens of thousands of euros a year, and enforcement keeps escalating — Google alone was fined another €325 million in 2025 over its ad and cookie practices. The giants pay it as a toll and continue; the startup pays it as a tax it can barely afford, on its way to losing anyway. The rulebook is, with grim consistency, regressive at every level.

Enrico Letta, asked to diagnose the same disease, put his finger on the deeper cost: “Fragmentation directly weakens Europe’s innovative capacity and strategic autonomy.” The regulatory machine and the fragmented market are the same problem seen twice — a continent that produces rules faster than it produces companies, then wonders why the rules have no domestic champions to protect. The cure Letta and Draghi both prescribe is not less ambition but more coordination: one market, one rulebook, one set of forms. It is telling that the two most senior establishment figures Europe could commission to study itself both came back with the same answer — integrate or decline — and that the answer has, so far, mostly been filed under noted.

By 2026 the friction had hardened into open refusal and visible deprivation. Meta simplydeclined to sign the AI Act’s code of practice, betting that the cost of fighting the rules was lower than the cost of following them; Apple blocked its flagship AI assistant, the upgraded Siri, from the European Union indefinitely rather than reshape it to the Digital Markets Act, so that European iPhone owners simply do not get features their American counterparts take for granted. Set against that, the cookie banner that greets every European on every website — some two-thirds of major sites show one, and by one study only a sixth are actually compliant — is almost a comic emblem: a continent that experiences the internet as a daily obstacle course of consent pop-ups, having successfully regulated the user experience while regulating none of the platforms beneath it. The rules are felt everywhere. The power they were meant to check sits, untouched, in California.

It would be unfair to end the chapter without conceding what the rulebook gets right, because the honest case is more damning than a one-sided one. When the EU mandated a single USB-C charger for all phones and laptops, it cut waste and saved consumers real money, and the rest of the world quietly followed; its right-to-repair rules and repairability labels are nudging an entire industry toward durability; the GDPR, for all its costs, genuinely raised the global floor on privacy. This is the Brussels Effect working as advertised — Europe setting standards the world adopts. The trouble is the asymmetry of what it has chosen to be good at. Europe has become the world’s indispensable regulator of products and a negligible producer of them; it writes the rules of the digital economy with one hand and rents the digital economy with the other. Being the referee is an honourable job. It is simply not the same as being a player — and a continent cannot regulate its way back onto the field.

For a single human-scale illustration, take Bird (formerly MessageBird), one of the Netherlands’ few homegrown tech unicorns. When it cut a fifth of its workforce, its chief executive did not blame the market or his own strategy first; he pointed at the compliance burden of operating across Europe’s thicket of rules, and made plain that the next jobs would be created somewhere friendlier. Multiply that one founder’s calculation across every scaling European company weighing where to put its next hundred engineers, and the regulatory reflex stops being an abstraction about sovereignty and becomes a steady, quiet export of exactly the high-growth employment Europe says it most wants to keep. The rulebook does not only fail to produce champions; it actively encourages the ones Europe already has to do their growing elsewhere.

And the burden falls most heavily where it is least affordable. The headline GDPR fines land on Google and Meta, but the quieter enforcement increasingly targets ordinary small and mid-sized firms — a German property company, Deutsche Wohnen, was fined some fourteen million euros over data-retention practices, the kind of penalty that is survivable for a hyperscaler and lethal for a mid-cap. For a small European company the compliance arithmetic is brutal: cookie-consent tooling, a data-protection officer, breach-notification processes and the standing risk of a fine measured as a percentage of global turnover can run to tens of thousands of euros a year before a single product ships. The giant treats all of it as a line item; the startup treats it as one more reason to incorporate in Delaware. Regulation meant to constrain the powerful keeps, with grim reliability, landing hardest on the small — which is to say, on precisely the firms Europe needs most.

Lay out the marquee enforcement of the last two years as a single list and the asymmetry becomes almost comic. Apple, fined €500 million; Meta, €200 million under the DMA and €390 million before that; TikTok, €530 million for shipping European data to China; X, €120 million; Temu, €200 million for unsafe goods; Clearview AI, €30 million for scraping faces — a roll-call of billions in penalties, and not one of the firms on it European. The point is not that the fines are wrong; most punish genuine abuses. The point is what the list is: an inventory of the digital economy’s most powerful companies, compiled by the one body on the continent still able to make them flinch, on which Europe appears nowhere as a defendant because it appears nowhere as a builder. The fines are the consolation prize of a power that gave up competing and settled for refereeing.

And the deprivation runs the newest way too. Meta has withheld its most capable multimodal Llama models from the European Union, citing regulatory unpredictability; Apple has kept its upgraded Siri off European iPhones; the Commission has opened formal proceedings against X over the risks of its Grok chatbot. The cumulative effect is a continent that increasingly receives the frontier of AI late, in cut-down form, or not at all — not because its engineers cannot handle it, but because its rules make it the most legally fraught market in the rich world to ship into. Europe set out to govern the technology and has ended up, in a growing number of cases, simply not being offered it. That is the chokepoint paradox in its newest and most literal form: a market so busy regulating the future that the future has begun to route around it.

A moat only protects you if the drawbridge goes up for your enemies and down for your friends. Europe built the deepest moat in the world and then lowered the bridge for the giants and raised it against its own.

The deepest problem with the regulatory reflex is not any single rule; it is what the reflex reveals. Regulation is the tool of a power that has decided its job is to govern an economy someone else will build — to be the referee, the standard-setter, the conscience — rather than to build the economy itself. It is the posture of a landlord, not an owner; a regulator of other people’s platforms, not a builder of its own. And a continent that regulates what it does not own ends up, by a slow and dignified route, owning nothing but the rulebook. Which brings us, at last, to the bill — the one chapter that totals the whole machine, in euros, and asks where a quarter of a trillion of them go every single year.

Now the bill. Every chapter so far has traced one stream of the outflow — the savings, the companies, the talent, the electrons, the defence euros, the value the rulebook fails to keep. This chapter totals them, in the one currency that ends arguments: euros, on a ledger, paid annually. The headline number is the cloud, and it is the cleanest single measure of the whole machine. Europe pays the United States something like €264 billion a year for digital infrastructure and software — a quarter of a trillion euros, every year, for the right to run its economy on someone else’s computers. It is not a debt, which would at least end. It is a tribute, which renews by default.

Watch where the money actually goes, because the shape of it is the whole argument. Of Europe’s professional cloud spending, roughly 80 to 83% flows to US providers, with about 70% captured by the three American hyperscalers — Amazon, Microsoft, Google. What Europe keeps is the thin end: the reseller margins, the wages of the local sales teams, the concrete and steel of the data centres it builds on its own soil to host American software. The value that compounds — the equity, the intellectual property, the platform — books elsewhere. Europe is, quite literally, paying to construct the buildings in which its digital dependence is housed. Step down the waterfall below and watch the basis points leave.

The accumulated result of paying that tribute for a decade is a gap in raw corporate scale that has stopped being a gap and become a different order of being. The seven largest American technology companies are now worth, together, more than $20 trillion — on the order of the entire GDP of the European Union, and well over the combined value of every company listed on every European exchange. Drill into a single one and the point turns surreal: Nvidia alone, at about $4.8 trillion, is worth roughly as much as the entire German economy — one American chip company, on the scale of Europe’s industrial heartland. Pan across the landscape below; the American mass on one side and the European on the other are not two teams in the same league. They are two different sports.

A defender of Europe has, at this point, one number left to reach for — the trade statistics, which appear to show Europe running a healthy surplus with the world in services, anchored by Ireland’s spectacular export figures. It is the last bright number on the board, and it is a mirage. Ireland’s headline GDP, around €563 billion, sits about 75% above its real national income (GNI*, near €321 billion) — the gap is almost entirely the accounting shadow of US multinationals booking global profits through Dublin for tax reasons. Strip that distortion out, and the genuine transatlantic digital balance is not a surplus at all but a deficit on the order of −$350 billion. The number that looked like Europe winning was American profit, parked in a European postbox. Flip the toggle below from the mirage to the real ledger.

And the tribute buys more than dependence; it buys exposure, because the platform Europe rents comes with a foreign government attached. Around 80% of the EU public sector’s productivity software is Microsoft’s, and data held by American companies falls, under the US CLOUD Act, within reach of American law wherever in the world it physically sits. For years this was an abstraction lawyers worried about. In 2025 it stopped being abstract: after the US administration sanctioned the International Criminal Court’s chief prosecutor, Karim Khan, his Microsoft email went dark, and he decamped to a Swiss provider. Microsoft insists the Court chose to disconnect him rather than the company cutting him off — the attribution is genuinely disputed — but the lesson landed regardless: the email of the world’s leading war-crimes prosecutor became inaccessible the moment Washington was displeased, on a platform that runs Europe’s ministries too. Throw the switch below.

A quarter of a trillion euros a year, a market worth a continent, a surplus that is a mirage, and an off-switch in a foreign capital. This is the ledger, and it has been open, and unbalanced, and unremarked, for a decade.

It would be easy to read this chapter as the bleakest of the lot, and in euros it is. But notice the faint, repeated counter-melody, because the essay turns on it: Euro-Office, a sovereign European software suite, launched in June 2026; Denmark, Austria and France have begun migrating their public sectors off Microsoft; the Karim Khan affair did more to concentrate European minds on data sovereignty than a decade of position papers. The tribute is enormous and it is real. It is also, like every other line on this ledger, a standing order — and the one thing we know about standing orders, from the first chapter to this one, is that they can be cancelled by anyone with the nerve to sign the form.

The exposure is not a worst-case hypothetical; it is the default legal architecture. The US CLOUD Act of 2018 lets American authorities compel data held by American companies anywhere on Earth, with no requirement for the host country’s consent — and since something like 80% of EU public-sector data sits on American platforms, the Karim Khan episode was less an aberration than a live demonstration of a standing reality. The same concentration shows up on the tax ledger: Ireland now draws roughly 46% of its corporation tax from just three US firms — splendid until the day Washington changes its own rules and a third of a small country’s revenue turns out to hinge on decisions made in a foreign capital. Dependence, it turns out, is not only something you pay for in cloud invoices. It is something you quietly bank your public finances on.

The counter-melody, faint but real, is that some governments have finally started to walk out. After the Khan affair, Denmark began moving its public administration off Microsoft email and productivity software; Austria’s federal computing centre started shifting departments to open-source LibreOffice and Nextcloud; France pressed ahead with its sovereign-suite plans. These are early, partial and technically painful — a ministry does not leave Outlook in an afternoon — and they will not, by themselves, move the €264-billion figure much. But they matter as proof of concept: the tribute is not a law of physics. It is a contract, and contracts can be cancelled by a customer with the will to absorb the switching cost. The whole essay keeps asking whether Europe has that will at scale, or only in a brave few capitals.

And the dependence is deepest exactly where it matters most. European firms run a meaningful slice of ordinary computing on local providers — but for the strategic workloads, the AI training and inference that will define the next decade, the hyperscalers’ share runs toward nearly 100%, because only they have the scale of accelerated compute to do it. So the tribute is not a flat tax across the whole economy; it is concentrated, and rising, precisely on the frontier. Europe can keep renting yesterday’s computing from itself. Tomorrow’s it rents, almost entirely, from three American firms — which makes the quarter-trillion figure not a ceiling but a floor, set to climb as the workloads it cannot yet host at home become the whole game.

To feel the asymmetry, set the scoreboards side by side. The seven largest American technology companies are together worth more than every listed company in Europe combined, by something like half again — and there is no European company within an order of magnitude of the largest of them, the chip-maker whose single name now stands against the scale of a major European economy. This is not a gap that closes with a good quarter or a clever policy; it is the accumulated interest on twenty years of the standing order, compounded — and still compounding.

Even the “sovereign” answers keep routing back to the same firms. Germany’s flagship sovereign-cloud project, Delos Cloud, is built on Microsoft Azure; the much-trumpeted Microsoft EU Data Boundary, which promised European data would stay in Europe, turned out on inspection to carry documented carve-outs letting engineers outside the EU reach support data. And the dependence has a price as well as a politics: when Broadcom bought VMware, the virtualisation software a great deal of European enterprise runs on, and sharply raised its prices, a coalition of European cloud firms filed a formal antitrust complaint — because they had nothing else to switch to. “Sovereign,” in practice, keeps collapsing back into “American, with a European label and a reassuring clause” — and the clause is only as good as the day Washington decides otherwise.

If you want the dependence stated in a single sentence, a Microsoft executive supplied it under oath. In June 2025 the company’s French legal director, questioned by the French Senate, conceded that Microsoft could not guarantee that European customers’ data would be shielded from American authorities under the CLOUD Act — could not, in other words, promise the one thing “sovereign cloud” is sold to deliver. It was an honest answer, and a damning one. Around it the extraction continues in quieter forms: when European governments levied digital-services taxes on the platforms, the platforms simply passed the cost back to European advertisers as a surcharge. The tribute is not only large; it is structured so that the attempts to claw it back are themselves billed to Europe.

The money does flow back in trickles, where governments fight for it. France collects several hundred million euros a year from its digital-services tax; Italy widened its own by scrapping the revenue threshold so that any sales into the country are caught. But these are rear-guard skirmishes over a sliver of the outflow, and they invite retaliation — Washington has repeatedly threatened tariffs against countries that tax its champions. The more telling movement is on the procurement side, where the German military joined the civilian agencies in refusing Palantir over sovereignty concerns. It is the same lesson arriving from every direction at once: you cannot tax your way back to sovereignty, and you cannot litigate your way there either. You can only build or buy your way there — and Europe has spent two decades doing neither.

And the squeeze comes from two directions at once, which is the part that makes it feel inescapable. If the United States owns the top of the technology stack — the platforms, the frontier models, the cloud — China increasingly owns the bottom: it refines something like 92% of the world’s rare earths and produces the overwhelming majority of the gallium and germanium that chips, sensors and weapons require, and it has already shown it will throttle those exports as leverage. Europe sits in the middle, dependent on America for the software that runs its economy and on China for the materials — and increasingly the manufactured goods, the electric cars, the solar panels, the batteries — that fill it. To be a chokepoint empire is to be feared. To be caught between two of them, holding decisive leverage over neither, is to be the place where the squeeze is felt.

And in 2026 the bottom of that squeeze stopped being abstract. BYD’s electric-car registrations in Europe rose around 180% year on year in early 2026, vaulting it into the continent’s top three despite a combined EU duty above forty per cent; CATL, already the supplier of more than half of Europe’s batteries, is building gigafactories on European soil while holding a global battery share north of forty per cent. Roughly 95% of Europe’s solar panels come from China, and when Beijing trimmed an export rebate in 2026 the price of European solar simply rose. Where Washington reaches for export controls, Beijing reaches for the mirror image: when the EU put tariffs on Chinese electric cars, China answered with duties of up to 42.7% on European dairy, plus probes into its pork and brandy — aimed, with precision, at the farm vote — even as it began forcing Huawei out of European 5G cores. The continent’s 2025 goods deficit with China widened to some €360 billion.

The same two-front logic now reaches the frontier itself. China’s open-weight models — DeepSeek and its successors — deliver capability close to the American frontier at a fraction of the cost, and they slip past US export controls entirely, because anyone can simply download them. For a continent that cannot match American capital intensity, the cheaper Chinese tier is genuinely tempting — and reaching for it would only swap one dependence for another. That is the trap in its final form: Europe can rent the top of the stack from America or the bottom of it from China, but the one option fifteen years of strategy has not produced is the third — owning enough of either to set its own terms.

Widen the lens beyond cloud and the tribute only grows. Europe runs a services trade deficit with the United States of around €178 billion a year, much of it digital; global cross-border payments for the use of intellectual property — the licensing of patents, brands and software — passed a trillion dollars, with the United States the largest single collector. Even the regulator’s biggest swings barely dent it: the Commission fined Google three billion euros in 2025 over its ad-tech dominance, lifting its cumulative EU antitrust fines past eight billion — sums that are at once the largest in the history of competition law and a rounding error against the revenue the same firm pulls out of Europe every year. The tribute is paid in a hundred currencies — cloud fees, ad revenue, IP royalties, app-store commissions — and totalled, it is the quiet transfer of a great power’s economic surplus to another.

Even the most basic plumbing of the economy runs on foreign rails. Every time a European taps a card, the transaction almost certainly flows through Visa or Mastercard — two American networks that together handle well over half of the continent’s cashless payments and clear some twenty-odd trillion euros a year, taking a sliver of each and seeing the data trail of a continent’s spending. Europe has tried for two decades to build its own: the “Monnet Project” collapsed when national banks would not cooperate, and only now, with Wero — a pan-European wallet that moved billions in its first year and, in 2026, linked arms with Italy’s, Spain’s, Portugal’s and the Nordics’ national systems — is a genuine alternative finally taking shape. It is the whole essay in miniature: a sovereignty so basic most Europeans never think about it, surrendered for decades to American firms, and reclaimable only once the continent finally decides to stop competing as twenty-seven and start acting as one.

The extraction runs through the screen as well as the wallet. Meta alone pulls something like forty to forty-four billion euros a year in advertising revenue out of Europe; Google takes more; the app stores levy their tithe on every download and subscription. And one layer deeper, the financial system itself is denominated in someone else’s currency: the US dollar appears in roughly half of all international payment messages and the overwhelming majority of foreign-exchange trades, which lets Washington turn the plumbing of global finance into an instrument of foreign policy at will. Europe has its own currency and one of the world’s great trading blocs, and still its companies advertise, transact and settle on rails it does not own. The tribute, fully tallied, is not one bill but a stack of them — and almost every one is addressed across the Atlantic.

And the rawest resource of all, Europe gives away for free. Every search, post, purchase and location ping a European generates is data, and data is the feedstock of the AI economy — and the overwhelming majority of it is harvested, refined and monetised by American platforms, then sold back to Europe in the form of the very services that collected it. Europe’s answer, a Data Act meant to treat data as shared European infrastructure, took effect in 2025 — and pointedly excluded the largest gatekeepers from its core data-sharing duties, the loophole quietly swallowing the rule. A continent that frets about importing oil has spent two decades exporting something more valuable, in unlimited quantity, for nothing: the recorded behaviour of four hundred and fifty million people, the single richest training set on Earth, shipped abroad as fast as it is generated and bought back as a finished product.

The Karim Khan affair had a sequel that is the most hopeful thing in the chapter. In October 2025 the International Criminal Court — having watched its prosecutor’s Microsoft account go dark under US sanctions — formally migrated off Microsoft entirely, moving to openDesk, an open-source suite built by Germany’s public Centre for Digital Sovereignty. It is a small institution and a small migration, technically painful and far from complete. But it is the proof of concept the whole essay has been circling: that when the threat becomes concrete enough, the supposedly immovable dependence can be moved — that “sovereign,” open, European infrastructure is not a fantasy but a working alternative, waiting only for the will, and the fright, to adopt it. The tribute is enormous. It is also, as the ICC has just demonstrated, cancellable.

The dependence reaches even the information Europeans use to govern themselves. The continent has no search engine, no social network, no app store of any scale — so the news its citizens read, the debates they have and increasingly the facts they believe are mediated by American platforms whose incentives are not European. When Google’s AI-generated answers began keeping users from clicking through, European publishers watched their traffic fall by roughly a third and their viability, in the Commission’s own words, come “at risk in nearly all member states.” When the owner of X chose to throw his platform’s weight behind a far-right party before a German election, there was no European venue of comparable reach to counterbalance it. A continent can write all the content rules it likes; if it does not own the pipes the information flows through, it does not, in the end, control its own public square.

Tally the everyday digital life of a European and the absence is total. The streaming they watch is American — Netflix, Disney, YouTube and the rest take some 80% of online video minutes across the major European markets; the online shopping runs largely through Amazon, which leads all five of the continent’s biggest e-commerce markets; the search is Google, which holds close to 90% of the European market against low-single-digit shares for home-grown alternatives like Ecosia; the browsers are overwhelmingly Chrome; the phones run iOS or Android; the social feeds are American or Chinese; the cloud behind all of it is one of three US firms. There is no European Google, no European Amazon, no European Netflix, no European iPhone — not because Europeans would refuse to use them, but because Europe never built them, and, having not built them, now lives its entire digital existence as a paying guest in structures it does not own. A continent can be wealthy and cultured and free and still, in the medium that increasingly mediates everything else, be a tenant from morning to night.

Scholars have started reaching for older words to describe all this, and the words are uncomfortable. The economist Yanis Varoufakis calls it “technofeudalism” and Europe a “vassal”; Cédric Durand sees capital shifting from making things to extracting rent from platforms; Couldry and Mejias describe the harvesting of human behaviour as a new “data colonialism”; Farrell and Newman map how control of the world’s digital chokepoints became, in their phrase, “weaponised interdependence.” Even Anu Bradford, who literally wrote the book on the “Brussels Effect,” now says its “high-water mark” has passed. You do not have to accept every flourish of the framing to feel its weight: a relationship in which one party supplies the platforms, the capital, the chips and the rules, and the other supplies the data, the customers and the rent, is not a partnership of equals. It is one of the oldest arrangements in economic history, wearing a touchscreen.

There is one accelerant left to account for, and it is the reason the whole ledger is about to get worse, fast. Every gap in this essay — capital, talent, energy, scale — is being multiplied in real time by a technology that rewards the leader and strip-mines the laggard with unprecedented speed. The final chapter of the diagnosis is about that accelerant: software that now costs almost nothing to make, deflates by an order of magnitude a year, and pours its returns into exactly the concentrated, well-capitalised, energy-rich hands Europe does not have. If the machine has run quietly for a decade, Software 3.0 is the foot going down on the pedal.

Everything you have read so far is the before. Ten chapters of gaps — capital, companies, talent, energy, scale, the quarter-trillion tribute — measured in a world that was, by the standards of what is coming, standing still. This chapter is about the multiplier. Artificial intelligence is not just another industry Europe is losing; it is a general-purpose accelerant that rewards, with brutal precision, exactly the five things Europe has spent this whole essay not having: scale, capital, cheap energy, data, and frontier compute. Pour it over a continent that is winning, and it compounds the lead. Pour it over a continent that is losing, and it compounds the loss. Software 3.0 is the foot going down on the pedal — and the car is pointed the wrong way.

Start with the deflation, because it is genuinely one of the most astonishing facts in the modern economy and it is almost universally misread. The cost of a unit of AI capability is falling by roughly an order of magnitude every year. A task that cost thirty dollars per million tokens on a GPT-4-class model in 2023 can be served by a “good enough” model for a few cents today — a fall of around 1,000× in three years at the cheap end, and still some 60× at the frontier. The instinctive read is that this democratises intelligence, hands it to the small and the scrappy, levels the field. It does the opposite, and the reason is Jevons’ paradox: when something essential gets radically cheaper, you do not use less of it, you use vastly more. Token demand is forecast to grow some 24× by 2030; enterprise AI budgets have risen nearly six-fold even as per-token prices fell about 65%. Cheap intelligence does not shrink the prize. It explodes it. Watch the two curves cross below.

And an exploding prize is won by whoever can pay to serve it — which turns the whole game into a contest of raw capital intensity, the one contest Europe has structurally opted out of. In 2026 the four largest American hyperscalers will spend, between them, on the order of $725 billion on AI and data-centre capital expenditure — up about 77% in a single year, and closer to $755 billion once you fold in xAI and the Stargate build-out. The whole of Europe’s comparable capex is around $60 billion. That is not a gap; it is a canyon — roughly twelve to one, and widening. Europe’s flagship answers are real and earnest and an order of magnitude too small: the EU’s “AI Gigafactories” initiative at around €20 billion, and the heroic, lonely figure of Mistral — France’s genuine frontier lab, valued at perhaps $23 billion in its 2026 raise. Mistral is the best Europe has, and it is roughly forty times smallerthan Anthropic, whose valuation crossed $965 billion in 2026, or OpenAI at $852 billion. Set the canyon to scale below; the European ledge all but disappears.

The capital gap becomes a capability gap with a short lag, and the capability numbers are the starkest in the essay. Of the notable frontier AI models released in 2025, the United States produced around fifty and China about thirty; Europe produced roughly three. The United States commands an estimated 74% of the world’s high-end AI compute; Europe holds something like 5 to 6%, and that share is falling, not rising. Every frontier-leading model since 2023 has come out of an American lab. This is what it looks like to be absent from the defining technology of the age not as a consumer — Europe consumes AI voraciously, mostly American AI — but as a producer. And it matters precisely because of the accelerant logic: a continent that does not make the frontier models pays rent on everyone else’s, forever, on terms it does not set.

Now put the two halves together, because this is the chapter’s whole argument. AI rewards scale — Europe is fragmented. It rewards capital — Europe exports its capital. It rewards cheap energy — Europe’s power costs double. It rewards data and frontier compute — Europe has little of either. Every single input that AI turns into compounding advantage is an input this essay has already shown Europe lacks. So the technology does not open a new front in the competition; it takes every existing gap — the capital drain, the brain-drain, the energy premium, the scale deficit — and runs a multiplier over it. Pull the accelerant lever below and watch the gaps you have already read about widen in real time.

A tide that lifts all boats still leaves you behind if your boat is tied to the dock. AI is not a rising tide. It is a current, and it runs toward scale, capital and cheap power — away from a continent that rationed all three.

It would be easy to end the diagnosis here, on the bleakest possible note: the machine has been handed a turbocharger and bolted it to the side that was already winning. And as description, that is correct. But the accelerant cuts both ways, and this is the hinge the final chapters turn on. A multiplier applied to a deficit widens it — but a multiplier is also the only thing that can close a gap fast, if you can get on the right side of it. The same deflation that lets a US hyperscaler serve a billion users cheaply also lets a five-person European startup do what once took five hundred people. The same Jevons explosion that rewards the compute-rich also creates more demand than any one bloc can serve. Europe is losing the AI race for reasons that are, every one of them, choices — and a technology that multiplies outcomes is, by definition, the most powerful tool ever invented for reversing a position fast.

There is a particularly cruel edge to this for Europe, because the one corner of software the continent genuinely leads is exactly the corner the accelerant threatens first. Europe’s enterprise-software champions — SAP, the German giant, and the per-seat business model it exemplifies — sell software by the user, by the month. But when a frontier model can do the data extraction, the support reply or the junior-developer task for a few cents of tokens, the per-seat fee starts to look like a toll the customer can route around, and the defensible ground migrates up to the things Europe does not own: the frontier model, the chips, the distribution. To feel the asymmetry that enforces it, hold one number in mind — a single US hyperscaler’s annual capital budget now exceeds the entire yearly wage bill of a mid-sized European country’s AI workforce. You cannot out-hire, out-build or out-spend that from a standing start, not without first deciding to try.

Put the capital gap in named-company terms and it stops being an abstraction. In a single year Microsoft alone is spending something like eighty to a hundred-and-twenty billion dollars on AI and cloud infrastructure; Amazon on the order of two hundred; Google and Meta tens of billions each; and the Stargate venture alone proposes to spend up to half a trillion dollars on data centres over a few years. Any one of those line items dwarfs the whole of Europe’s coordinated AI-infrastructure ambition. This is what it means to say the contest is decided by capital intensity: the unit of American investment is one company’s quarterly capex, and the unit of European investment is the multi-year, twenty-seven-country, much-debated programme. They are not the same kind of number — and pretending they are is how a decade goes missing.

There is one more turn of the screw, and it is the lane Europe most conspicuously left empty. As frontier models grew ruinously expensive, a parallel world of open-weight models — freely downloadable, cheap to run, good enough for most tasks — exploded to capture roughly half of enterprise inference. This was, in principle, Europe’s natural territory: open, sovereign, deployable on your own hardware, the very antidote to renting intelligence from a foreign hyperscaler. Instead the open-weight flood was led from China, by DeepSeek and Alibaba’s Qwen, and from American challengers like Meta’s Llama — and even the open models run overwhelmingly on American and Chinese cloud infrastructure. Europe had the strongest possible case for an open, sovereign-inference strategy, and the talent to execute it, and watched two other powers fill the lane while it debated the rules of the road.

The window in which this could be reversed cheaply is also closing, which is the part that should frighten European policymakers most. Each generation of frontier model costs more to train than the last — tens, then hundreds of millions of dollars, soon billions — and the set of organisations that can afford to sit at the very edge is narrowing toward a handful, all of them American or Chinese. A challenger two years behind today may be structurally unable to catch up tomorrow, not because its researchers are worse but because the price of a ticket to the frontier has risen past what any European entity will fund. The deflation makes intelligence cheap to use and ruinously expensive to make — and Europe, with all its savings, has so far chosen to be a buyer.

Europe is not doing nothing. The EuroHPC programme has stood up some of the world’s fastest public supercomputers — Germany’s exascale Jupiter among them — and the gigafactory plan would add five more. It is real, sovereign, publicly-owned compute, and it deserves credit. But weigh it honestly against the private American build-out and the scale problem reappears: a single US hyperscaler brings more accelerated compute online in a quarter than Europe’s flagship public machines hold in total, and it does so every quarter, funded by cash flows no public budget can match. Public supercomputers are necessary and they are not sufficient. The question is never whether Europe can build some sovereign compute. It is whether it can build enough, fast enough, to matter — and on current trajectories the honest answer is not yet.

And the binding constraint, underneath the capital and the chips, turns out to be the most physical thing of all: power. Training a frontier model is, in the end, a question of how many megawatts you can point at a data centre for months on end — and on that axis Europe’s two-to-threefold electricity-cost disadvantage from Chapter Seven becomes an AI disadvantage directly. The Chinese lab DeepSeek stunned the field in late 2025 by training a competitive model for a reported few million dollars and releasing it open-weight, proof that the frontier is not only about brute spend. But the broader trend runs the other way: training compute is scaling more than fivefold a year, and the three leading American labs alone command something like sixty per cent of the world’s frontier compute. The accelerant’s fuel is electricity and silicon, and Europe rationed its access to both.

Europe is, to its credit, finally trying to build models of its own. EuroLLM, trained on Barcelona’s MareNostrum supercomputer to speak all the EU’s languages, and the OpenEuroLLM consortium spanning nine countries, are genuine attempts at a sovereign, open, multilingual alternative to the American and Chinese frontier. They are also, by the standards of the firms they mean to answer, tiny — tens of millions of euros against training runs that cost hundreds. And the constraint bites deeper down the chain than money: the advanced packaging that stitches AI chips together, TSMC’s so-called CoWoS capacity, is sold out into 2027, the overwhelming majority pre-bought by American firms. You cannot rent your way to the frontier when the queue for the only factory that can build it is already full, and you are nowhere near the front.

Reduce the whole contest to one ratio and it is brutal. On the most consequential input of all — high-end AI compute — Europe holds under five per cent of the world’s, some seventeen times less than the United States, and even the optimistic projections have that share peaking near eight per cent later this decade before sliding back. Europe is responding at the only scale it can muster collectively: EuroHPC has announced something like thirty-five new AI supercomputers across twenty-three nations. It is a real and serious effort, and it is still being lapped, because the American build-out is private, profit-funded and relentless while the European one is public, budgeted and negotiated. Compute has become the new oil, and Europe is bringing a carefully-costed national programme to a market the hyperscalers are flooding with their own cash.

The scale of the appetite is hard to hold in the mind. Global data centres are on course to consume more than a thousand terawatt-hours of electricity in 2026 — about as much as the whole of Japan — and the curve bends sharply upward from there. Satya Nadella, Microsoft’s chief executive, drew the conclusion bluntly: economic growth, he said, will be “directly correlated” to the cost of the energy that powers AI. Read that sentence from a European chair and it is a verdict. The continent has chosen, through Chapter Seven’s long list of decisions, to make its energy among the dearest in the developed world — which means it has chosen, without ever quite saying so, to make itself one of the most expensive places on Earth to build the defining industry of the century. The accelerant runs on watts, and Europe rationed its own.

To feel how far the frontier has run, look at a single American cluster. Elon Musk’s xAI built a data centre in Memphis it calls Colossus, scaled toward two gigawatts of power and on the order of half a million Nvidia GPUs — one company, one site, commanding more concentrated AI compute than entire European nations, stood up in months. There is nothing in Europe remotely like it, and on current trajectories there will not be: the continent’s flagship answer, Mistral’s sovereign cloud outside Paris, is measured in tens of megawatts and thousands of chips — a creditable startup against an industrial juggernaut. The gap is not that Europe lacks AI ambition. It is that its largest ambition is roughly the size of an American afterthought.

And even Europe’s answer is built on borrowed silicon. Mistral’s own buildout — a €1.2 billion data-centre programme spanning a live Paris site and a Swedish facility for 2027, on the order of 13,800 Nvidia GPUs — runs, like roughly four-fifths of all the AI accelerators in Europe, on chips designed in California and fabricated in Taiwan. Members of the European Parliament have named the obvious bind: the very “AI gigafactories” meant to buy sovereignty deepen the continent’s dependence on a single American supplier, because there is no European alternative to Nvidia to buy instead. It is the chokepoint paradox nested inside its own remedy — a sovereign cloud sovereign in everything except the one component that matters most.

The cruelest statistic in the whole essay may be this one. Europe has more monthly users of large-language-model chatbots than the United States does — some 133 million, roughly twice the American base — and almost every model they use was built in America. Europeans are, per head, among the most enthusiastic consumers of AI on Earth and among the most negligible producers of it: the continent files perhaps 3% of the world’s new AI patents against America’s seventy. This is the chokepoint paradox rendered in a single behaviour — a market that adopts the technology avidly, pays for it eagerly, generates the data that improves it freely, and owns none of it. To be the best customer of an industry you hold no stake in is not a position of strength. It is the very definition of a market, as opposed to a competitor.

And the frontier is already moving past the screen into the physical world, where Europe is, if anything, further behind. The race to build humanoid robots — general-purpose machines that could transform manufacturing and logistics — is being run by America’s Figure and Tesla and a phalanx of Chinese firms; Figure’s robots have already logged months on a BMW production line, which is to say that the German carmaker’s glimpse of the robotic future runs on American hardware and American AI. There is no European entrant of comparable scale. The same holds in AI-for-science: the protein-folding breakthroughs that will reshape medicine came out of a London lab — owned, as ever, by Google. Whichever way the technology turns next, from chatbots to robots to drug discovery, Europe keeps arriving as the customer, the test site, or the acquired — almost never the owner.

One pair of numbers captures the whole asymmetry. In 2025 American private investment in AI reached some $286 billion; the entire European venture-capital industry — not its AI spending, its whole annual venture pool, across every sector — deploys roughly sixty to eighty billion dollars a year. America invests more in artificial intelligence in a single year than Europe’s venture industry deploys, across every sector, anywhere, full stop. You cannot compete for the defining technology of the century out of a venture industry a fraction the size of your rival’s spending on that one technology alone. It is not that European investors are timid, though some are; it is that the pool they draw from is structurally too shallow, because the savings that should fill it are, as the capital chapter showed, parked in deposits and bonds and American markets instead. The accelerant runs on capital, and Europe brought a teaspoon to a flood.

That is the knife-edge on which this essay now balances, and it is the right place to turn from diagnosis to prescription. The machine is real, the tribute is real, the accelerant is real. But nothing in any of these eleven chapters was an act of God. It was capital that chose safety, regulators who chose process, governments who chose fragmentation, and a continent that chose, again and again, to manage its decline with dignity rather than risk an undignified attempt to reverse it. The last question is the only one that matters: what would Europe have to actuallydo — not resolve, not report, but do — to take the key it is holding and change the lock?

After eleven chapters of loss it would be reasonable to expect this one to be a eulogy. It is not, and the reason is the single most important fact in the entire essay: Europe is not poor, and it is not weak, and it is not, whatever the figures suggest, out of the game. It is a continent that holds a genuinely extraordinary hand and has spent fifteen years declining to play it. The tragedy of the chokepoint paradox was never that Europe lacks the cards. It is that it holds the one card no one else can replace — and keeps it in its pocket. This chapter is about the card, the moves, and the first real signs that the continent has finally started to reach for the table.

Begin with the card itself, because it is almost absurdly good. In the small Dutch town of Veldhoven sits ASML, a company worth around €650 billion that holds a 100% monopoly on extreme-ultraviolet lithography — the machines without which no advanced chip on Earth can be made. Not most chips: none of the leading-edge ones. TSMC cannot print a 3-nanometre processor without ASML; neither can Samsung, nor Intel, nor anyone in China, at any price, because there is no substitute and no second source. This is not a vulnerability. It is the most concentrated point of leverage in the entire global technology supply chain, and it is European. Around it Europe holds the rest of a formidable hand: €33.5 trillion in household savings, a single market of 450 millionpeople, 22% of the world’s scientific papers, and €2.5 trillion of manufacturing. The title of this essay is not a metaphor. Europe holds the key. Turn it below.

So if the assets are real, the obvious question is what to do with them — and here is the part that should be either infuriating or liberating, depending on the hour: there is no mystery. The plans are written, costed, and in several cases already signed. Mobilise the savings through a real Savings and Investments Union, and the €10-11 trillionsitting in low-yield deposits — and the roughly €300 billion a year that currently leaks abroad — could fund European companies instead of American ones. Complete the single market, whose internal barriers still amount to tariff-equivalents of around 54% on goods and 95% on services, and the prize is on the order of €2.8 trillion of GDP; Enrico Letta’s “One Europe, One Market” roadmap has even put a deadline on it — December 2027. Back the frontier through InvestAI’s €200 billion mobilisation and its five planned AI gigafactories. Fix the grid and the power price. Buy and build European. Pull the levers below — and then look at the gauge that matters.

That gauge is the whole problem in one number. As of early 2026, barely 11 to 15% of the Draghi report’s recommendations had been fully implemented. The savings-union measures are mostly non-binding suggestions to member states; the single-market roadmap is a list of intentions with a deadline it is already behind. This is the chokepoint paradox in its purest form: the diagnosis is correct, the prescription is written, the patient has the money for the medicine — and the prescription sits un-filled on the counter, because filling it requires twenty-seven governments to act together against their own short-term incentives, and that is the one thing the machine is built to prevent. The levers are not missing. The hand on the levers is.

And yet — and this is why the chapter is not a eulogy — something has changed in the last eighteen months that fifteen years of competitiveness reports never managed. The continent has started, haltingly and unevenly, to move. Germany rejected Palantir over sovereignty. A sovereign software suite, Euro-Office, launched; Denmark, Austria and France began migrating their public sectors off Microsoft. Mistral raised three billion euros at a twenty- billion valuation and crossed four hundred million in revenue. Europe pledged to spend like it means it on defence. Letta’s roadmap got signed. None of these is sufficient. All of them are real. Walk the field below and watch when it came alive.

Read those green shoots carefully, though, because they share a single, sobering root: almost every one was watered by fear, not foresight. Germany did not reject Palantir because of an economic argument; it did so because an American administration had made the cost of dependence suddenly, viscerally real. Europe is re-arming because Russia is on its border and the American guarantee feels conditional. The data-sovereignty awakening followed a war-crimes prosecutor’s email going dark. The recurring, uncomfortable finding of this entire essay is that the slow economic version of the threat — the bloodless capture documented in every chapter — never moved Europe at all; only the sharp, frightening version did. The open question, the one the next five years will answer, is whether a continent can learn to act on the quiet emergency before it becomes a loud one.

Return, finally, to the savings, because they are the whole argument in microcosm. Of Europe’s household wealth, something like €10 to 11 trillion sits in bank deposits earning next to nothing — not invested, not at risk, not building anything, simply parked. In the same years, the continent’s own competitiveness reports put the annual investment shortfall at €800 billion and climbing toward €1.4 trillion. The money to close the gap is not missing; it is sitting in current accounts, and a third of a trillion euros of it leaks abroad every year to be invested by someone else, in someone else’s companies, for someone else’s returns. A continent does not get more self-evidently rich, or more self-evidently timid, than that: the deepest savings pool on Earth, declining to fund its own future, then wondering why the future is owned elsewhere.

None of which means the only move is to out-spend the un-out-spendable. The smarter prescriptions aim not at matching America gigafactory for gigafactory but at the lanes where Europe could actually win. Edge inference — running AI cheaply on local, low-power chips rather than in vast foreign data centres — plays to European strengths in semiconductors and embedded systems; companies like the Dutch-founded Axelera are already chasing it. Open-weight, sovereign models hosted on European soil answer the dependence head-on. And the single most powerful lever requires no new technology at all: redirect even a fraction of the €33 trillion in European savings — through a pension-allocation rule, a public guarantee, a Savings and Investments Union with actual teeth — and the missing growth capital appears, because it was never truly missing. It was parked.

It would be dishonest to end on easy optimism, because the obstacle is real and it is structural. The reason the savings stay parked and the single market stays fragmented is not that no one has noticed; it is that every fix requires twenty-seven governments — each with its own banks to protect, its own champions to favour, its own electorate to answer to — to surrender a piece of control at the same moment. A genuine capital-markets union threatens national financial centres; a real single market threatens protected national firms; a serious industrial policy means picking winners across borders. Each is individually rational to block, and collectively catastrophic to keep blocking. That is the true lock on the door — not American power but European collective inaction — and it is the one no foreign capital can pick for Europe, and none can keep shut either.

And there are real things to back, named and fundable today, if the capital would only turn up. Draghi’s own report called for a €50-billion European Deep Tech Fund; the seeds it would water already exist. Axelera and Germany’s Black Semiconductor are building European AI silicon; IQM in Finland and Pasqal in France are among the world’s leading quantum-computing firms; a EuroStack movement has published a manifesto for a modular, Europe-owned computing stack from chips to cloud to applications. None of these is a fantasy; each is a company or a coalition with customers and a roadmap, starved only of the patient, late-stage capital Europe exports by the hundreds of billions every year. The way out is not a mystery waiting to be discovered. It is a decision, already drafted and costed, waiting to be funded from savings that already exist.

One reform in particular has gathered momentum because it attacks the fragmentation head-on: “EU Inc”, a proposed pan-European legal entity — a single, optional “28th regime” company form a founder could incorporate under once and operate across all twenty-seven member states, instead of refounding in each. Thousands of European founders and investors have signed on, because it targets the precise friction that sends them to Delaware: not a lack of talent or ideas, but the absurdity of a “single market” in which building a company that works everywhere means complying everywhere. It is small, technical and unglamorous, and it is exactly the kind of fix that matters — because the chokepoint was never built by one grand decision, and it will not be dismantled by one either. It will be dismantled, if at all, by a hundred boring reforms enacted with unfamiliar urgency.

And the roster of fundable European deep tech is longer and stranger than the gloom suggests. Isar Aerospace is building rockets in Germany; The Exploration Company, split between Munich and Bordeaux, is building reusable space capsules; Proxima Fusion is chasing a working fusion stellarator with German state money behind it; IQM and Pasqal are selling real quantum computers to real customers. None is a sure thing; several will fail. But the seedbed is not empty — it is, if anything, unusually rich — and what these companies share is not a shortage of ambition or engineering but the same missing input as everyone else in this essay: the patient, scaled, late-stage European capital that would let them grow up at home instead of being bought, or starved, before they can. The way out is standing right there, incorporated and pitching. It is waiting to be funded.

It helps to be honest about the size of what would be required, because it is the part the green-shoot stories tend to skate over. Draghi’s own figure for the investment Europe needs is around €800 billion a year — close to 5% of the continent’s GDP, sustained for years. To feel the scale: the Marshall Plan ran at roughly one per cent of American output; the Apollo programme peaked near half a per cent; what Draghi asks for is several times the intensity of either, not for a few years but indefinitely. That sounds impossible until you remember that Europe has already done the once-unthinkable thing the plan requires: in 2020 it agreed to borrow jointly, some €750 billion through NextGenerationEU, breaking a taboo that had stood for the union’s entire history. The machinery for collective ambition exists. It has been switched on exactly once, in a pandemic, and switched off again. The question the next decade answers is whether it takes a catastrophe to turn it on — or whether, just this once, foresight might be enough.

And the templates for what such a mobilisation looks like are not exotic; America itself built them. The Interstate Highway System wired a continent together with public money over a decade; DARPA seeded the internet, GPS and a dozen foundational technologies by funding the research private capital would not touch; the whole American semiconductor and internet edge was, at root, a story of patient public investment that private dynamism then compounded. Europe keeps imagining that sovereignty is something you legislate. The American example says it is something you build — with state money, at national scale, over decades, accepting the failures along the way — and then let the market run with. Europe has the savings to fund a dozen DARPAs. What it has lacked is the conviction that the future is worth buying outright rather than renting by the year.

France is the test case for whether ambition is enough, and the early returns are sobering. No government has tried harder: the Tibi programme has corralled some thirteen billion euros of institutional money toward European tech; Mistral is building a sovereign AI cloud outside Paris; Hugging Face, the Paris-founded “GitHub of machine learning,” hosts millions of open models as a genuine counterweight to the American labs; the country runs on cheap, clean nuclear power that ought to be a decisive AI advantage. And yet French startup funding still roughly halved in the first half of 2025 after a snap election spooked investors, and the flagship champions remain a fraction of their American rivals. If the European country doing the most can be knocked off course by a single election, the lesson is not that the strategy is wrong. It is that the strategy needs to be bigger, steadier, and shared across the whole continent — immune to any one nation’s politics — or it will keep being undone faster than it can compound.

The most encouraging sign is also the most recent. In June 2026 the European Commission unveiled a sweeping tech-sovereignty package — on the order of a hundred and twenty billion euros for semiconductors, two hundred billion for data centres, a hundred billion for cloud and AI, and a smaller but pointed open-source pillar — a tacit admission that owning the code may matter as much as owning the metal — alongside a “Scale-Up Europe” fund aimed squarely at stopping high-potential startups from leaving. The figures, for once, are within sight of the scale of the problem rather than an order of magnitude below it; the diagnosis has plainly landed. Whether this proves the turning point or merely the latest, largest plan to be “filed under noted” depends entirely on execution — on whether twenty-seven governments fund it, sustain it across elections, and resist the thousand temptations to water it down. The plan is finally the right size. The only remaining question is the one this essay has asked from its first page: whether Europe will do it, or merely resolve to.

And before despairing that Europe simply cannot build champions, recall that it has done exactly that, spectacularly, when it chose to. Airbus began in 1970 as an improbable cross-border consortium — French, German, Spanish and British — backed by patient state money against a near-total American monopoly in large passenger jets. For years it was mocked as a subsidised vanity project. Today it is one of only two companies on Earth that can build a wide-body airliner, routinely out-selling Boeing, employing tens of thousands across the continent and exporting to the world. Airbus is the existence proof the whole prescription rests on: when Europe pools its resources across borders, backs an industry with public capital for the decades it takes to mature, and refuses to be scared off by the early losses, it can build a world-beater from nothing and hold it. The continent plainly does not lack the ability to do for AI, chips and energy what it once did for aerospace. It lacks, so far, the will to try at that scale more than once a generation.

And Airbus is not a lonely exception; it is the headline of a longer list Europe rarely reads back to itself. Galileo, the EU’s satellite-navigation system, now delivers civilian positioning accurate to twenty centimetres — better than the American GPS it was told it would never need to build. CERN gave the world the Web. Erasmus has put eighteen million young Europeans across one another’s borders since 1987, knitting a continent together one student at a time. The euro is the world’s second reserve currency; the single market is the largest trading bloc on Earth. Every one of these is a thing Europe was told it could not, or need not, do — and did, by pooling sovereignty and spending public money with patience across borders. The lesson could not be plainer, or more encouraging: the machine that produces European failure and the machine that produces European triumph are the same machine. It depends entirely on which way Europe decides to point it.

And the proof is not only in the grand pan-European projects; it is in the pockets where the continent already behaves the way the whole of it could. Sweden, a country of ten million people, has produced some forty unicorns — the highest count per head in Europe, with the alumni networks behind Spotify and Klarna still minting more. Estonia, smaller still, has the densest startup ecosystem in Europe per capita and the second-highest unicorn rate per head in the world after Israel, built on a state that digitised itself completely. These are not resource miracles; they are the ordinary output of good education, early capital, light rules and a decision to treat technology as a national project. Where a European place has chosen to compete, a European place has competed. The deficit was never in the genes. It is in the choice.

And Europe need not even invent the playbook, because three other latecomers wrote it in living memory. Taiwan went from licensing an obsolete American chip process in 1973 to spinning out TSMC — today the maker of ninety per cent of the world’s most advanced chips — in fourteen years, by way of a single determined state research institute. South Korea spends over five per cent of its GDP on research and pairs its industrial giants with decades of patient state financing to hold a fifth of global chip production. Israel, a country of nine million, turned a military signals-intelligence unit into a startup conveyor belt that has produced more than ninety unicorns and the highest venture investment per head on Earth — one of them, Wiz, sold to Google in 2026 for thirty-two billion dollars. None of these is bigger, richer or better-educated than Europe; every one simply decided, and then sustained the decision across governments and decades. The difference between them and Europe is not capacity. It is the willingness to choose a hard thing and refuse to let go of it.

And the next races have not yet been lost — they are being run right now, and Europe is actually on the track. In quantum computing, Finland’s IQM and France’s Pasqal are among the genuine global leaders, installing real machines into European supercomputing centres. In fusion, Munich’s Proxima Fusion is building a two-billion-euro stellarator demonstrator with Bavarian and Max Planck backing, alongside the giant ITER reactor rising in the south of France. In space, Ariane 6 is finally flying a steady cadence and a clutch of new launch startups — Isar Aerospace, Rocket Factory Augsburg — are reaching for orbit. The science is there; the nerve is flickering to life. But watch the early-warning lights, because they are the same ones as ever: IQM and Pasqal are both eyeing Nasdaq listings, and in 2025 Europe managed around nine orbital launches to America’s hundred and ninety-plus. Europe is in the next races. Whether it stays in them, and on its own terms, is the question this whole essay has been about.

The hardest obstacle is structural, and it has a name: unanimity. On the decisions that matter most — tax, foreign policy, the deepest fiscal integration — a single member state can veto the other twenty-six, and they do: there have been dozens of national vetoes in the past decade and a half, Hungary alone casting some twenty-one, often to extract unrelated concessions. The treaties even contain a mechanism to escape this trap — the “passerelle” clause — but invoking it itself requires unanimity, and it has been used essentially once since 1992. So Europe is a great power in a straitjacket of its own stitching: it has the population, the wealth and the science of a superpower and the decision-making speed of a condominium association where any one resident can halt the roof repair. When people ask why Europe cannot simply do the things this essay prescribes, this is the honest answer. The will is not the only thing missing. So, by design, is the ability to act on it without everyone agreeing at once.

One of the more elegant ideas to come out of the soul-searching is Enrico Letta’s, who proposed that Europe needs a “fifth freedom” to sit beside the famous four of goods, services, capital and people: the free movement of research, knowledge, innovation and education across the continent’s borders. It sounds abstract until you notice how un-free those things currently are — a discovery made in one member state, a researcher trained in another, a spin-out incorporated in a third, each crossing borders a single market was supposed to have erased decades ago. The fifth freedom is really just the recognition that Europe’s greatest untapped asset is not a pot of money but the ability to let its own brilliance circulate and combine without friction. It is, like almost everything in the prescription, less a matter of building something new than of finally removing the obstacles Europe long ago, and quite deliberately, built around itself.

History offers Europe a precise and unflattering mirror. Leadership in trade and technology has changed hands before — from Venice to the Dutch Republic to imperial Britain — and the shape of each handover is the same: the incumbent, grown rich, loses its nerve for the new thing, drifts from making into rentier finance and the curation of past glories, and is overtaken by a hungrier rival still willing to build. It is not a comforting precedent, because in every case the decline felt, from the inside, like stability. But the mirror cuts both ways: each of those powers was eventually succeeded by another that chose to build, which means the role is never retired, only reassigned. The stakes were put most memorably at the November 2025 Berlin digital-sovereignty summit co-hosted by Chancellor Merz and President Macron, in a single line from Ralf Wintergerst, head of Germany’s digital-industry association: “If Europe does not want to become a museum of technology, we must ramp up investment significantly.” The museum is one available future. It is not, yet, the only one — and which it becomes is still, narrowly, a matter of choice rather than fate.

The scale of ambition is, at least, finally being named at the top. In early 2026 President Macron called for Europe to invest €1.2 trillion a year across the green, digital and defence transitions — a figure that would once have been unthinkable and is now merely daunting. It needs to be, because the trend it must bend is stark: Europe’s share of world output has fallen from around 28% in 2010 toward a projected 17% by 2050, a trajectory on which, by mid-century, no European state would sit among the world’s largest economies at all. Numbers that large can read as rhetoric. But they are the honest measure of the gap between managing a decline and reversing it — and the whole argument of this essay is that the second course is still, expensively and narrowly, available.

We opened with a paradox and we can now close it. Europe is the richest savings bloc on the planet and Wall Street’s most reliable source of capital. It trains the engineers, writes a fifth of the science, and owns the one machine in the entire chip supply chain that cannot be replaced. By every measure of raw capacity it should be a peer of the United States and a rival to China. And instead it ships its savings, its founders, its companies and a quarter-trillion euros a year to the firms that out-compete it, and calls the arrangement stability. This was never a failure of genius. Europe has the genius. It is a failure of nerve, and across twelve chapters the failure turned out to have names, dates, and line items — every one of them a choice, which means every one of them reversible.

The cost of continuing as we are is not abstract, and Draghi named it plainly: without radical change, he warned, Europe faces a “slow agony” of decline, an “existential challenge” to the European model itself. The arithmetic agrees. On current trajectories Europe’s share of world output, already down to around a seventh, slides below a tenth by 2050 — a continent shrinking from a pole of the world economy into a museum of it, prosperous and admired and increasingly beside the point. Jean Monnet, who built the union out of the rubble of the last time Europe destroyed itself, left a line that reads differently now: there is, he wrote, “no future for the people of Europe other than in union.” He meant it as the path out of war. It is turning out to be the path out of decline as well — and Europe keeps, politely, declining to take it.

And the clock is not neutral. Europe is ageing fast: on current trends the working-age population falls in twenty-two of the twenty-seven member states by 2050, with a projected shortfall of tens of millions of workers, even as the bloc’s share of world output slides toward a tenth. A continent that is shrinking, ageing and ceding the technologies of the future all at once does not have the luxury of a slow, dignified adjustment; the demographic arithmetic turns every decade of delay into a steeper climb. The phrase that haunts Europe’s own strategists is that the continent risks becoming an open-air museum — a beautiful, beloved place to visit, exhibiting the achievements of a civilisation that has lost the will to add to them. The whole of this essay is an argument that this outcome is not fated. But it is, now, thedefault — the thing that happens if nothing is decided.

And yet the most important thing that happened in the writing of this essay is that the question stopped being academic. In 2025 the United States, under a new administration, reminded Europe in a sequence of shocks — tariffs aimed explicitly at its tech rules, a pause in intelligence to Ukraine, sanctions that knocked a war-crimes prosecutor off his email, open talk of taking Greenland by force — that the dependence catalogued in these pages was not a comfortable economic arrangement but a lever, and that the hand on it could no longer be assumed to be friendly. By the spring of 2026, polls found that barely one in ten Europeans still regarded the United States as an ally. The fright did what fifteen years of competitiveness reports could not: in November 2025, twenty-three governments and a roomful of companies met in Berlin and pledged the first real money toward digital sovereignty. Whether that proves the turning point or just the loudest meeting yet, no one yet knows. But the paradox has, at last, been felt as well as measured — and the things that are finally felt are sometimes, at last, the things that get done.

A key in the pocket opens nothing. The whole of Europe’s tragedy, and the whole of its hope, is that the lock was never the problem. The hand was.

That is, in the end, the strange consolation of the chokepoint. A continent that had simply been out-built could do nothing but try to build faster and lose. But Europe has not been out-built; it has been out-nerved. It holds ASML, the savings, the market, the science — the key is real and it is in European hands. Turning it does not require a miracle, a new technology, or American permission. It requires twenty-seven governments and a few hundred million people to decide, just once, that managing a dignified decline is the more frightening option. Washington owns the lock. But the lock is useless to everyone, including Washington, if the key is never turned — and the key, the one no one else on Earth can cut, has been European all along. The only question left in the whole affair is the one this essay cannot answer, because it is not a question of data: will Europe turn it?