One machine prints the world's chips — and China is closing in
Every time you unlock your phone, ask an AI a question, drive a modern car or tap to pay, you are leaning on a tiny chip. Almost every advanced chip on Earth is printed by a single machine, made by a single company, in a small Dutch town. It is the biggest bottleneck in technology — and the reason the world's two largest powers are fighting. Now Beijing says it has built one of its own. Chips are the brains behind nearly every electronic thing we own. And here is something very few people know: almost all of the world's most advanced chips depend on one machine, built by one company, in the Netherlands. No other country and no other company has managed to make that machine and sell it. Not the United States. Not Japan. Not China, despite spending more money trying than most nations spend on defence. That is why it has become one of the biggest bottlenecks in the world — and a big reason behind the technology fight between Washington and Beijing. Whoever can print the best chips can build the best phones, the best weapons and the best artificial intelligence. Everyone else waits in line. Control the machine, and you control who gets to build the future's brains. America worked this out and acted. Since 2019 it has pressed the Dutch government to stop the machine ever reaching China. Not one has been sold there. The bet was simple: without it, China could never reach the very smallest chips, and would be stuck one step behind forever. Beijing did not accept that. It poured in state money, put its biggest technology companies on the job, and — according to reports that are still unverified — switched on a working version of its own inside a guarded building in Shenzhen in late 2025. To judge whether that matters, you first need to see what the machine actually does. Then we will come back to the fight. Before anything else, understand this: a chip is made with light. Not blades, not drills, not tiny robotic hands — light. Every step of this story, every billion spent and every rule written, comes back to one beam and how thin it is. A chip is not carved like a sculpture. It is made by printing tiny patterns using light. Powerful light passes through a special template (called a mask), is focused by precision lenses, and falls on a silicon wafer coated with a light-sensitive material called photoresist. Wherever the light hits, the material changes. The unwanted parts are then removed. This process is repeated around 80 times, layer by layer, until billions of tiny electronic switches are built on a single chip. Those switches are called transistors. The more of them you fit into the same space, the faster and cooler the chip runs. In 1971 a processor held 2,300 of them. The one in your pocket holds tens of billions, in an area smaller than a postage stamp. Which creates the problem that has defined the last twenty years of engineering: the light got too fat. Trying to print a 5 nm wire with 193 nm light is like signing your name on a grain of rice with a marker pen. A nanometre — "nm" — is a millionth of a millimetre. It is the unit this entire industry now argues in, so it is worth holding on to: the width of the light you print with, and the width of the line you are trying to draw, are measured on the same ruler. And for years, the light has been far too wide for the job. The finest line a machine can draw comes down to one simple rule: thinner light draws thinner lines. Stronger optics help too. Try the controls below and see if you can make a 5 nm chip — the technology that powers modern smartphones. Slide the light down the list. Watch the lines go from smeared to sharp. Notice what happens the moment you step away from that last, thinnest light. With the older 193 nm light, no amount of clever glass gets you a clean line at the smallest sizes in one go. Chipmakers got around it by printing the same layer three or four times over, each pass nudged slightly, letting the overlaps carve out the fine edges. It works — and it costs a fortune. The newest light squeezes those four print runs back into one. That, in a sentence, is why one machine is worth hundreds of billions. Everything that follows — the money, the export bans, the raids on rival engineers — is a fight over 13.5 nanometres of light. Here is the catch nobody expects. No lamp, no bulb, no laser you can order gives out light this thin. It has to be ripped out of matter. And the way it is done sounds made up. A drop of molten tin, thinner than a hair, falls through an empty chamber. A small laser pulse squashes it flat. Then a colossal laser — thirty thousand watts, built in Germany — blasts the flattened drop into a gas hotter than the surface of the Sun. As that gas dies, it gives off a flash of the light we need, and a curved mirror scoops it up. In a working machine this happens fifty thousand times a second, all day, for years. Now the cruel part. This light is swallowed by everything — glass, air, even your breath. So there are no lenses anywhere inside the machine. The light is bounced along by mirrors, each one polished so perfectly that the biggest bump on its surface is about the height of a single atom. Even then, every mirror eats roughly a third of the light. Bounce it off ten of them and only a few percent of what you made survives. That is why the source has to be so absurdly overpowered. You build a star to deliver a candle. Watch it run · 4 min · opens on YouTube The falling droplets, the double laser pulse, the flash of plasma — in motion, at the speed the machine actually does it. It is widely called the most complicated machine humans have ever built — and it is not one invention. It is around thirty of them, forced to cooperate. If any single one fails, the whole machine fails. More than 5,000 suppliers feed it. It has over 100,000 parts. Picture a machine the size of a city bus and the weight of two lorries, wrapped in a white metal skin. There is no window and nothing to see: no glowing beam, no moving arms. What you notice instead is the plumbing — thick bundles of cable, silver pipes, and steel vessels bolted to every side, running away into the floor and the ceiling. Every one of those pipes exists for one reason: to protect, cool, feed or steady a beam of light that no human eye can see, travelling through a chamber with no air in it. Engineers work around it in white suits and hairnets, because a single speck of dust in the wrong place is a ruined chip. The machine costs roughly the same as a jumbo jet, and it is delivered in around forty freight containers. Tap any violet dot to open up that part of the machine. At first glance it makes no sense. The most important machine in technology is not built in America, China or Japan. It is built in Veldhoven, a quiet town in the south of the Netherlands, by a company called ASML. The truth is more interesting than the surprise. The Netherlands did not invent everything inside the machine. What ASML did — over nearly four decades — was gather the best technologies in the world and force them to work together as one system. It began in 1984, when the electronics giant Philips and the chip-equipment maker ASM International set up a small joint venture. It was not a glamorous business. For years, ASML did one thing with total stubbornness: build printing machines that could draw a little finer than last year's. Visible light. Then ultraviolet. Then deeper ultraviolet. Then, for almost two decades, it poured billions into a problem most people thought would never pay off — while working with universities, national labs and hundreds of specialist suppliers. Because one breakthrough was never going to be enough. To make the next machine work, almost everything had to be reinvented at once: a completely new source of light And if any one of them failed, the whole machine failed. That is the part outsiders underrate. It was not a race to one invention. It was a race to make thirty inventions arrive at the same time, and agree to cooperate. ASML is often described as the Dutch company that controls the world. It is truer to call it the conductor. It designs the machine and makes the pieces work together — and then leans on a supply chain no single country could ever seize. Designs the machine and makes 100,000 parts behave as one. The mirrors. Flat to within an atom. Months of polishing for each one. The giant laser that hits every tin droplet, twice, 50,000 times a second. The light-sensitive coatings, the masks and the blank silicon the light writes on. Light-source technology, measuring tools, precision electronics and design software. US-led rules ban shipments of the newest machine to China. Not one has ever been legally sold there. People assume the secret is a document in a safe in Veldhoven. It isn't. The knowledge that matters cannot be written down. It lives in the hands of engineers who have failed ten thousand times and remember exactly how. Zeiss spent decades learning to polish and coat glass to within the width of an atom. There is no second supplier on Earth. Recreating that isn't a purchase order — it's recreating a career. When to drop the tin, how hard to hit it, how the leftover tin slowly dirties the mirror — none of it comes out of a textbook. Every fix in that chamber came from a machine that broke first. More than 5,000 suppliers, each holding one strange little skill: a seal that holds a vacuum, a table that glides without touching anything, a coating recipe. Replace any one of them and you start years of testing again. The first prototypes ran in the 2000s. It took until roughly 2019 before the machine was reliable enough to be trusted in a real factory. Fourteen years of debugging is the product. The monopoly isn't a patent. It's a memory. If one company makes the only tool that matters, that company is not really a company. It is a chokepoint. And chokepoints get taken over by governments — not by seizing them, but by telling them who they may sell to. The lock went on in stages, and each turn of the key was tighter than the last. 2019 — the licence dies. The Dutch government, under sustained American pressure, quietly declines to renew ASML's export licence for its newest machine to China. No dramatic announcement. A machine simply stops being available. To this day, not one has been legally delivered to a Chinese factory. 2022 — the rules go wide. Washington moves beyond the machine to everything around it: other advanced equipment, the American engineers who service it, the software used to design the chips. The clever, brutal move is cutting off spare parts and support. A machine you cannot service is a machine with an expiry date. 2023–24 — the allies fall in. The Netherlands and Japan match the American rules. The restrictions creep downwards, from the newest machine to the best of the older 193 nm ones — because Washington realises the old workhorse is doing more than it was supposed to. The theory of victory was elegant. Deny the machine, and a rival is stuck. They can print 14 nm chips, maybe scrape to 7 nm at terrible cost, and then physics simply refuses. No cutting-edge chips. No cutting-edge AI. Permanent second place, enforced by nature rather than by policy. The wall held. What nobody priced in was that a wall is also an instruction. Export controls told Beijing precisely where its weak point was, exactly how much it was worth, and handed it a decade of guaranteed political will to fix it. Money that would never have survived a normal cost-benefit review became untouchable. And China went shopping for everything that wasn't banned — becoming, for a stretch, ASML's single biggest customer, buying up every older and mid-range machine it was still allowed to own. The story is usually told as one dramatic breakthrough. It isn't. It's four separate climbs, happening at the same time — and only one of them is the machine everyone talks about. In September 2023, Huawei launched a phone with a 7 nm processor inside it, built by SMIC — a Chinese chip factory with no access to the newest machine, and no prospect of getting one. Washington had assumed this was impossible. It wasn't. It was just expensive. The trick is simple to describe. If your light is too fat to draw a fine line in one go, you draw the same layer several times over, each pass nudged slightly, and let the overlaps carve out the sharp edges. With enough passes, you can beat your own light. What you cannot beat is the arithmetic — every extra pass costs money, costs time, and ruins more chips. A rough model, built to show the shape of the problem — not real factory data. The true survival rates are among the best-kept secrets in the industry. Push it down to 3 nm and watch the right-hand column fall apart. That is the honest position China is in today: not blocked, but taxed — paying several times the cost, taking several times as long, and throwing away a brutal share of everything it makes. You can win a phone launch that way. You cannot win a decade. So Beijing went after the machine itself — and not the way a market does. There was no investor round, no plan to turn a profit. The state named the target, poured in tens of billions through its national chip funds, and pointed an entire industrial base at the problem: Huawei's engineers, a Shenzhen toolmaker called SiCarrier, established equipment firms like SMEE, Naura and AMEC, plus the university labs. This is the part Western analysts keep underrating. A private company would have killed a project like this after the third impossible year. A state-run campaign does not have that failure mode. It has the opposite one: it cannot stop. Here is the interesting bit. China does not seem to be building a copy of the Dutch machine. Its researchers are chasing two other ways to make the same light — and both keep turning up in Chinese patents and research papers. A cheaper spark. Instead of firing a giant laser at falling tin, you run a powerful electric spark between two pieces of tin and let the spark itself create the light. It is cruder, and historically it has not been bright or clean enough. But it needs no giant German laser — and the giant German laser is exactly what nobody will sell them. A ring instead of an explosion. The stranger, more ambitious idea. Rather than blowing up tin at all, you use a particle accelerator: send electrons racing round a ring, pack them into tight bunches, and they give off the light continuously — a tap you never switch off, rather than a flashbulb fired 50,000 times a second. It is a university idea, linked to work at Tsinghua, and it is nowhere near a factory. But if it ever worked, it would not be catching up. It would be a different machine altogether — one ring, possibly feeding a whole factory. Neither has been proven at factory scale. But together they point to the thing that should worry Western officials more than any prototype: China has stopped trying to copy the Dutch answer, and started hunting for its own. The less romantic channels matter too. Over a decade, Chinese buyers stockpiled older machines and second-hand equipment from Nikon and Canon, then stripped them for tables, sensors and optical parts. Reporting suggests the Shenzhen prototype is a hybrid — new home-grown light-source work bolted onto reverse-engineered pieces pulled from machines that were bought perfectly legally, before the door shut. US officials have said they fear restricted equipment may have reached China through middlemen; ASML has publicly denied that its newest machines ever did. And as we established, blueprints aren't the bottleneck — people are. Reuters reported in December 2025 that the programme systematically recruited senior engineers away from ASML in Europe and TSMC in Taiwan, at pay far above the market rate. Whether or not every detail holds up, the strategy is obvious and sound: you cannot steal knowledge that lives in someone's hands. You can only hire the hands. The West protected the machine. The knowledge walked out on its own two feet. Say the reports are right, and a working machine has been switched on in Shenzhen. What would still stand between that machine and a chip in a phone? Tick off what a challenger has actually solved, and watch the readiness meter move. This is the honest shape of the story. Cracking the light is a real, enormous scientific achievement — and it is also the part that most resembles a laboratory. The other 80% is measurement, chemistry, reliability and waste: thousands of small, boring wins, stacked up over years. Most independent analysts put 2030 as the earliest realistic date for China to be making these chips in volume, at home. Which leaves the world in a strange place. The machine is still one of a kind. The ban still holds. And yet the belief underneath the ban — that this was a wall, not a hill — has quietly stopped being true. The question has shifted from whether to when. That is the whole story, and it turns on a beam of light thinner than a virus, made by exploding tin, in a room where the air itself has been thrown out.The machine almost nobody has heard of
A printing press, but for atoms
Pick your light. Watch it print.
You cannot buy this light. You have to make a star.
Inside the machine that prints chipsStarts at 1:38 · plays on YouTube 180 tonnes, the size of a bus
How did a small Dutch firm end up holding the world by the throat?
mirrors smoother than a single atom
chambers emptied of air
new sensors, new software, new materialsNo country makes this machine
ASML
Carl Zeiss
TRUMPF
JSR · TOK · Shin-Etsu
Cymer · sensors · software
Export controls
Why a blueprint would not help you
You can't buy the mirrors
You can't guess your way through the plasma
You can't hire the whole network
You can't skip the years of failure
How the West locked the door
How China is catching up
1. The workaround that already works
One shot with the new light, or four with the old?
One clean pass
The old light, four times over
2. The mobilisation
3. Trying a different kind of light
4. The parts, and the people
How the chip war got here
A prototype is not a factory
Click it and Unblock the Notifications
Sign in with Google