The EUV Monopoly: Why Every AI Chip on Earth Depends on One Dutch Company

There is a machine, roughly the size of a city bus and weighing about 150 tons, that has to exist before any advanced computer chip can be made. It costs upward of €150 million — the newest version, €350 million and up — ships in some 40 freight containers aboard multiple cargo planes, and takes a team of engineers months to reassemble on site. Inside, it draws circuit patterns using light so short and so difficult to produce that generating it means firing a laser at droplets of molten tin roughly 50,000 times a second.

Exactly one company on Earth makes this machine. It is not American, Taiwanese, or Korean — it is ASML, based in the Dutch town of Veldhoven. Every leading-edge processor in every AI data center, every flagship smartphone, and the advanced memory feeding the AI boom passes through ASML's equipment first. This article explains how one company came to hold the single tightest chokepoint in the global economy, how the business actually makes money, why its order book is one of the best leading indicators of the entire AI buildout — and where the monopoly is genuinely vulnerable.

The most complicated machine ever mass-produced

Chips are not carved; they are printed. Lithography is the printing step: a machine projects the pattern of a circuit onto a silicon wafer coated in light-sensitive chemicals, exposing the design layer by layer. The finer the pattern you can print, the smaller — and faster, and more power-efficient — the transistors. And the size of the smallest feature you can print is limited by one thing above all: the wavelength of the light you print with.

For two decades the industry used deep-ultraviolet (DUV) light at a 193-nanometer wavelength. To keep shrinking past a point, chipmakers needed something far shorter: extreme-ultraviolet (EUV) light at 13.5 nanometers, roughly fourteen times shorter. The problem is that nothing in nature emits usable EUV light cheaply, and EUV is absorbed by almost everything — including air and ordinary glass lenses. Solving it took one of the most improbable engineering efforts in industrial history.

Here is what ASML's machine actually does to make the light. A generator fires tiny droplets of molten tin across a vacuum chamber at about 70 meters per second. A high-power CO₂ laser hits each droplet twice — once to flatten it into a pancake, then again to vaporize it into a plasma reaching roughly 200,000 °C, which radiates a pulse of EUV light. This happens around 50,000 times a second. Because EUV can't pass through lenses, the light is steered by a series of mirrors made by Zeiss so precisely flat that, scaled up to the size of Germany, their largest imperfection would stand less than a millimeter tall. A single system contains over 100,000 parts. It is, by most accounts, the most complex device ever manufactured in volume.

How one company ended up alone

The strangest part of the story is that EUV did not begin in the Netherlands. It began as an American government project. In 1997, Intel, Motorola, and AMD formed a consortium called EUV LLC with three U.S. Department of Energy national laboratories — Lawrence Berkeley, Lawrence Livermore, and Sandia — under a roughly $250 million research agreement, one of the largest such partnerships in DOE history. The foundational science was American and publicly funded. Yet the company that turned it into a shippable product was the Dutch upstart that had licensed access to the research.

Turning lab physics into a factory tool took another two decades and tens of billions of dollars — so much that ASML's own customers helped pay for it. In 2012, in a program with few parallels in corporate history, Intel, TSMC, and Samsung together committed €1.38 billion to ASML's R&D and bought a combined 23% equity stake for €3.85 billion — the customers literally financing the supplier that would one day hold them captive. ASML also secured the two hardest pieces of the puzzle by locking up its supply chain: it relies on Zeiss for the optics and, in a deal completed in 2013, acquired Cymer — the maker of the laser light source — outright.

That is why the moat is not a single patent someone could invent around. It is an entire irreproducible ecosystem: 25-plus years of accumulated know-how, exclusive optics, an owned light-source business, and customer fab roadmaps built around ASML's specific machines. The two companies that could theoretically compete — Japan's Nikon and Canon — make only DUV systems and long ago chose not to chase EUV. In leading-edge lithography, ASML has no second source at all.

The two-engine business model

A monopoly on a €150-million machine sounds like it should print money smoothly. It doesn't — at least not from the machines alone. Equipment sales are lumpy and cyclical: chipmakers order in waves, and a single quarter's revenue can swing on whether a handful of tools shipped before or after the calendar rolled. ASML's stability comes from running two engines.

The first engine is new systems — EUV (over €150 million each, and €350 million-plus for the newest “High-NA” generation) plus the older, cheaper DUV tools that still print the majority of layers on any chip. The second, quieter engine is Installed Base Management: the service contracts, spare parts, and performance upgrades on every machine ASML has ever sold. Once a tool is in a fab, it generates high-margin recurring revenue for a decade or more — and because the installed base only grows, this engine climbs regardless of where the equipment cycle sits.

The importance of that second engine showed up plainly in ASML's Q2 2026 results. The company beat its own guidance on both revenue and gross margin — and it explicitly credited the beat not to selling more machines, but to record installed-base service revenue of €2.8 billion. The annuity, not the lumpy hardware line, carried the quarter. That is the structural reason a company selling one of the most cyclical products in technology can post remarkably steady results.

The other half of the model is a lesson in pricing power. In Q2 2026 ASML shipped 91 lithography systems, but only 16 of them were EUV — and those 16 machines accounted for 57% of all system-sales revenue. The other 75 tools split the remaining 43%. A single EUV machine sells for roughly an order of magnitude more than a mainstream DUV tool, which is monopoly economics in one chart: a handful of units, the majority of the value.

Where ASML sits in the AI chain

To see why ASML matters far beyond its own share price, follow the money. The AI buildout starts with the hyperscalers — Amazon, Microsoft, Alphabet, and Meta — guiding to roughly $700 billion of combined capital spending in 2026. That money buys NVIDIA GPUs and the custom accelerators Broadcom builds, each one wrapped in stacks of high-bandwidth memory from Micron and its rivals. But every one of those chips — the logic and the memory alike — has to be manufactured. And manufacturing them means running them through ASML's lithography machines. Every path in the AI supply chain eventually terminates at the same Dutch tool.

This position makes ASML an unusually good leading indicator. Fabs order lithography tools twelve to twenty-four months before the capacity actually comes online, so ASML's order book is a preview of where the whole chain is heading. And its end-use mix reveals which customers are spending. Through 2025, logic chipmakers like TSMC dominated ASML's system sales. Then the memory makers — scaling up for AI's voracious appetite — came off the sidelines. Memory's share of ASML's system sales roughly doubled, from about 30% in late 2025 to nearly half by mid-2026.

That shift is the memory supercycle made visible one layer up the chain. When memory prices exploded and Micron, SK Hynix, and Samsung printed record quarters, the skeptic's question was whether the boom was real capacity expansion or just a price spike. ASML's order book is part of the answer: the memory makers aren't only raising prices, they're buying the tools to build more — and paying ASML for them.

What the order book is saying now

In July 2026, ASML did something it rarely does: it tore up its own full-year forecast mid-stream. Having guided in April to €36–40 billion of 2026 revenue, management raised the number to €43–45 billion just three months later — a roughly €6 billion upgrade, and a pace of growth (around 35% over 2025) that is extraordinary for a company of this size. The third quarter alone was guided to €11–12 billion, up about 23% from the second. The stated reason: order intake through the first half of the year was, in the CEO's word, “extremely strong.”

Then came the louder signal. A guidance raise is a forecast; capacity is a commitment. ASML announced it is expanding its own factory output by about 30% for both EUV and DUV immersion tools for 2027 — and studying another 30% on top for 2028. A supplier only pours that much concrete when customers have committed to buying what comes out of it. This is the AI buildout being underwritten in physical capacity years ahead, and it is the strongest forward signal in the whole report.

The quarter carried one more milestone worth understanding. ASML's next-generation High-NA EUV machine — the €350-million tool that shrinks the smallest printable feature from 13nm to 8nm — reached its first high-volume production win. Intel qualified it on select layers of its 18A manufacturing process, now running in high-volume production in Oregon at reported yields around 85%. For years the open question about High-NA was whether it would work economically at scale; a real product shipping on it is the most concrete answer yet, and it de-risks the next decade of ASML's roadmap.

The bear case: a monopoly is not an immunity

None of this makes ASML invulnerable, and the honest way to hold the story is to name what could break it.

• Export controls. ASML cannot freely sell its most advanced tools to China, and the effect is visible in the numbers: China fell from 36% of system sales in late 2025 to just 14% by Q2 2026. So far, AI-driven demand elsewhere has more than filled the gap — ASML raised guidance through the China decline — but a permanent loss of the world's largest chip-consuming market is a real ceiling.
• The cycle is deferred, not repealed. ASML's leading-indicator position cuts both ways. If hyperscaler capex ever rolls over, tool orders would crack before almost anything else in the chain. Semiconductor equipment has always been cyclical; the services annuity cushions a downturn but does not cancel it.
• Customer concentration. A handful of chipmakers — TSMC, Samsung, Intel, SK Hynix, Micron — are essentially the entire order book. A stumble at any one of them (an Intel foundry setback, a memory glut) removes a large block of demand at once.
• High-NA economics. The Intel 18A win is real, but High-NA adoption is early. If the €350-million tools don't deliver enough value to justify their price across more customers, the next growth leg slows.

Could anyone break the monopoly?

The most-watched question in the industry is whether ASML's monopoly can last — and as of mid-2026, the challengers remain real but far behind.

China is trying hardest, out of necessity. Blocked from buying ASML EUV tools, Huawei and SMIC are reportedly developing a home-grown EUV machine using a different light-generation method (laser-induced discharge plasma rather than ASML's laser-produced plasma), with trial production reported and mass production targeted around 2026. But the gap is large: reported prototypes generate on the order of 100–150 watts of EUV power against ASML's ~600-watt benchmark, which translates directly into slower, more expensive wafers. China's most advanced shipping lithography tool is still a 28nm-class DUV system — generations behind the leading edge.

Canon is pursuing a genuinely different path called nanoimprint lithography, which stamps patterns rather than projecting light and costs a fraction of an EUV machine. It has real promise for cost-sensitive and memory applications, but the industry verdict so far is blunt: it is not ready for high-volume leading-edge logic, and no major foundry has committed it to advanced production. For now it is a complement to EUV in niche uses, not a replacement.

So the falsifiers to watch are specific and datable: a credible EUV alternative reaching competitive power and throughput; ASML quietly shelving those 2027–28 capacity expansions (which would signal demand softening); a broad cut in hyperscaler capex guidance; or a stumble at one of the few customers that make up the order book. Absent those, ASML's position is as close to structural as anything in technology gets — a company whose product is a precondition for the entire modern economy, made by no one else.

How to research this chain yourself

ASML is best understood not in isolation but as the base of a single connected chain: the hyperscalers funding the buildout (Microsoft, Alphabet, Meta), the chip designers the money buys (NVIDIA, Broadcom), the memory makers scaling for it (Micron), and ASML underneath all of them. Read any one link in isolation and you miss the picture — the same dollar of AI spending flows through every one of them, which is also why they share the same downside if the wave ever rolls over.

Sources

• ASML Q1 & Q2 2026 results, investor presentations and press releases (reported Apr 15 and Jul 15, 2026 — revenue, gross margin, guidance, end-use/region mix, capacity-expansion plans, dividend and buyback)
• High-NA EUV reaches first high-volume logic product (Intel 18A) — ASML/Intel, Jul 15, 2026
• ASML Customer Co-Investment Program press releases (Jul 2012) and the Cymer acquisition completion (May 2013), via ASML news and SEC 6-K filings (Intel/TSMC/Samsung €1.38B R&D + 23% equity stake for €3.85B)
• Construction Physics: How ASML Got EUV and CSET (Georgetown): Tracing the Emergence of EUV (EUV LLC founded 1997; DOE national-lab consortium)
• ASML: how EUV light is generated (tin droplets, dual laser pulse, 13.5nm) · machine weight/logistics via Tom's Hardware and Data Center Knowledge
• Tom's Hardware: High-NA EUV ~$380M, order book, ramp
• The Diplomat: China's EUV progress (Jul 2026) · China LDP EUV power vs ASML benchmark, via Global SMT & TrendForce
• SemiAnalysis on Canon nanoimprint lithography · IEEE Spectrum and Tom's Hardware on NIL status