There is a fact about the modern world that almost nobody outside the semiconductor industry knows, and that almost everybody inside it treats as the single most important fact of the AI age. Every advanced chip on Earth — every Nvidia accelerator training a frontier model, every processor in every iPhone, every high-performance logic chip in every data centre — is printed using light from a machine that exactly one company in the world knows how to build. Not one company that does it best. One company that does it at all. If that company stopped shipping machines tomorrow, the entire trajectory of computing would freeze within a few years, and there is no second supplier to call.
The company is ASML. It is Dutch. And the story of how a near-bankrupt 1980s spin-off from Philips became the most strategically important manufacturing company on the planet is the story of the chokepoint at the centre of the twenty-first century.
What the Machine Does
To make a chip, you print patterns onto silicon — the circuits that become transistors. The finer the patterns, the smaller and more numerous the transistors, and the more powerful the chip. For decades this was done with ordinary ultraviolet light. But there is a physical limit: light cannot resolve features much smaller than its own wavelength. To print smaller, you need shorter wavelengths. And below a certain point, the only usable light is extreme ultraviolet — EUV — at 13.5 nanometres.
Generating EUV light is absurdly difficult, and ASML’s method sounds like science fiction. A tiny droplet of molten tin is fired into a vacuum chamber. A laser pulse flattens it into a pancake. A second, more powerful pulse vaporises it into a plasma so hot it emits extreme ultraviolet light. This happens about fifty thousand times every second.
Fifty thousand times a second, a laser vaporises a droplet of tin into plasma hotter than the surface of the sun. The light it emits prints the circuits that think.
Then comes the second impossibility. At 13.5 nanometres, EUV light is absorbed by almost everything — including glass. You cannot focus it with lenses; they would simply swallow it. So the entire optical system works by reflection, using mirrors of a precision that beggars description. ASML’s mirrors are the flattest, smoothest objects ever manufactured by human beings: if you scaled one up to the size of Germany, the largest bump on its surface would measure a fraction of a millimetre. The light bounces off these mirrors, passes through a patterned mask, is optically shrunk by a factor of four, and lands on the silicon wafer — printing features 3 nanometres across, about twenty-five thousand times thinner than a human hair. Without EUV, a 5-nanometre chip would need something like a hundred separate patterning steps. EUV collapses that complexity. It is not a convenience. It is the only economically viable path to the leading edge.
The Underdog Origin
None of this was inevitable. ASML began in 1984 as a joint venture spun out of Philips — an afterthought, really, housed initially in a leaky shed next to a Philips building. Its first product was a flop. By 1988 the company was on the brink of collapse. It was saved less by commercial logic than by conviction: a Philips board member argued that Europe needed a stake in semiconductors, and kept it alive.
The decision that made ASML was contrarian. Its Japanese rivals, Nikon and Canon, were vertically integrated — they made their own optics, their own motors, everything in-house. ASML went the opposite way: it outsourced the optics and the motors and focused on system architecture, final assembly, and relentless optimisation. German engineers warned it would lose all control over its technology. They were wrong about the consequence. The modular approach meant ASML’s machines could be repaired on-site, component by component, and kept running longer than the competition’s. In an industry where a single hour of fab downtime costs a fortune, reliability beat raw precision. When IBM chose ASML over the Japanese competition, the trajectory began to bend.
The Bet on EUV
The decisive move came in 1997, when ASML joined EUV LLC — a public-private partnership built around the US national laboratories at Lawrence Livermore, Sandia, and Berkeley. EUV was, at the time, an unproven technology many believed would never work in production. ASML invested massively in a technology with no guarantee of ever functioning.
It took more than two decades. EUV did not reach high-volume manufacturing until around 2019. ASML spent the better part of thirty years and tens of billions of dollars turning a laboratory curiosity into the most complex machine ever built. Nikon and Canon, watching the cost and the risk, effectively gave up. By the time it worked, ASML was the only company left standing — and the entire semiconductor industry had no choice but to come to Veldhoven. This is how monopolies of genuine substance are made: not by buying out competitors, but by being the only one willing to spend thirty years and a fortune on a problem everyone else considered too hard. The moat is not legal or financial. It is the accumulated, compounding, almost un-copyable knowledge of how to make light from tin and bounce it off mirrors flat enough to print the circuits that run civilisation.
Why This Is the Real Bottleneck
The AI Job Thesis Fails at Physics made the case that the AI revolution is constrained by physics — by the need to build chips, power plants, and data centres faster than is physically possible. ASML is the sharpest point of that constraint. Every leading-edge chip that goes into an AI accelerator is printed by an EUV machine. The number of EUV machines that exist is limited by how many ASML can build — a few dozen of the most advanced units per year. You cannot buy your way around this. You cannot print more chips than the EUV installed base allows. And you cannot build more EUV machines than ASML produces.
So the ultimate ceiling on AI compute growth is not Nvidia’s design cycle or TSMC’s fab construction. It is the throughput of one company’s assembly line in the southern Netherlands. When that piece calculated a 28-year buildout at aggressive growth rates, the EUV production rate was one of the hidden variables underneath that arithmetic. ASML is the bottleneck inside the bottleneck — and it is the single most important industrial asset Europe possesses.
What It Means
The deepest moats are built on time and knowledge, not capital. ASML’s monopoly cannot be bought or copied because it is made of thirty years of accumulated engineering distributed across thousands of people and a supply chain of specialist firms. A rival with unlimited money could not replicate it in under a decade, probably not two. This is the same kind of moat as Bitcoin’s seventeen years of hashrate: not a feature that can be cloned, but a history that must be lived through. Time is the one input no amount of capital can accelerate past a point.
And Europe’s problem has never been capability. It has been the failure to recognise and wield it. The single most critical manufacturing chokepoint in the global technology supply chain — the one machine without which the AI age cannot proceed — is European. Not American. Not Chinese. Dutch. This is the thesis of The Last Axiom in miniature: capability aligned across borders by complementarity is power — if it is seen, named, and used. The continent that built the one machine has not yet understood what it means to own it.
Flight Log — Dispatch from Altitude
The jet engine on the wing of my A320 is one of the few machines in the world that rivals an EUV scanner for difficulty of manufacture. A modern turbofan runs with turbine blades spinning in gas hotter than the melting point of the metal they are made from — kept solid only by intricate internal cooling channels and ceramic coatings, each blade grown as a single crystal of metal because ordinary grain boundaries would fail under the stress. Perhaps a handful of companies on Earth can make them. The knowledge is guarded, accumulated over decades, and very nearly impossible to replicate from scratch.
When I push the thrust levers forward on the runway, I am trusting an object that only a few places in the world know how to build. And I have always found something reassuring in that — not the scarcity itself, but what it represents: that some things are hard enough, and important enough, that they can only be made by institutions willing to spend decades getting them right. The jet engine is not a commodity. Neither is the EUV scanner. Both are monuments to the kind of patient, compounding, un-glamorous engineering that no quarterly earnings call rewards and no shortcut can replace.
Europe makes both. The jet engines on European aircraft, the EUV machines that print the world’s chips — these are not relics of a faded industrial past. They are present-day proof that the hardest manufacturing on Earth still happens here. The continent so often described as a museum is, in fact, the place where the single most important machine of the AI age is assembled, in a town most people have never heard of, by engineers most people will never meet.
I fly aircraft built on European engineering. I write this blog on chips printed by a European machine. The capability is not gone. It was never gone. What is missing is not the engineering — it is the recognition of what the engineering means, and the will to wield it as the strategic asset it has quietly become. The one machine sits in Veldhoven. The future runs through it. And almost nobody, flying over the Netherlands at 38,000 feet, has any idea that the most important factory in the world is somewhere below them in the dark.