One Company Makes Every EUV Scanner

Most supply-chain stories in electronics involve a handful of dominant firms. The extreme ultraviolet lithography story involves exactly one. Every EUV scanner in every leading-edge fab in the world comes from a single company in Veldhoven, in the Netherlands. There is no second source at any price and no alternative route to the smallest transistors except a slower, costlier one. I want to walk through why that is true, starting from the physics of printing small features, moving through how 13.5 nm light is actually made, and ending with what a monopoly on this one machine means for governments and foundries planning a roadmap.

The equation that sets the limit

Photolithography is the step that turns a circuit design into a physical pattern on a wafer. A scanner shines light through a patterned mask, called the reticle, then through a reduction lens onto a wafer coated with photoresist, a light-sensitive chemical. The resist is then developed to leave the pattern behind. Every later step in the fab, whether etch, deposition or implantation, works inside the geometry that lithography has already drawn. That makes it the rate-limiting physical constraint on how many transistors fit into a given area of silicon.

The governing relationship is the Rayleigh resolution criterion. The smallest feature a projection system can reliably print, the critical dimension, depends on three things: the wavelength of the light, the numerical aperture (NA) of the optics, and a dimensionless process factor called k1.

That leaves three levers, and only three. Shorten the wavelength, enlarge the numerical aperture, or push k1 down. NA is a function of the refractive index of the medium the light travels through and the angle of the cone of light the optics can collect. This is why immersion lithography, introduced in the deep ultraviolet (DUV) era, replaced the air gap between the last lens element and the wafer with a thin film of ultra-pure water. Water bends light more than air does, which lifts the achievable NA above 1.0, a threshold that is physically meaningless in an air-gap system.

The k1 lever has a hard floor. For single-exposure imaging it sits near 0.25, and that limit comes from diffraction itself, not from any engineering shortfall that better tools could fix. Most of the tricks that dominate a modern lithography process, such as off-axis illumination, phase-shift masks, computational proximity correction and multiple patterning, exist to push a real process as close to that floor as the photoresist and the pattern fidelity will tolerate. None of them is free. Each adds mask complexity, simulation time or extra exposure and etch cycles. The story of the last two decades is the industry repeatedly deciding to spend capital on wavelength and NA rather than keep paying the k1 tax indefinitely.

Why DUV ran into a wall

For roughly twenty years the industry held wavelength fixed and spent everything on k1 and NA. Argon fluoride immersion DUV, at a 193 nm wavelength and an NA of 1.35, became the workhorse of logic and memory from the 28 nm node through 7 nm. According to the book, those systems cost roughly $75 million each and reach a single-exposure resolution of about 38 nm. To print anything finer, fabs split one design layer across two, four or more separate lithography-and-etch cycles. The techniques go by the names LELE (litho-etch-litho-etch), SADP (self-aligned double patterning) and SAQP (self-aligned quadruple patterning), and the combined result achieves a pitch that no single 193 nm exposure could resolve.

Multiple patterning works, but its cost does not scale gently. Each extra cycle means another full mask and another coat-expose-develop-etch loop through the fab. Each also creates another chance for the pattern from one exposure to drift out of alignment with the pattern from the previous one. That overlay error compounds across three or four cycles on a single layer and gets harder to control as pitches shrink. By the time critical layers at 10 nm and 7 nm needed quadruple patterning, the combination of cycle time, tool count and yield risk had become an economic argument in its own right. I find that the useful way to read the arrival of EUV: it was not only an engineering advance, it was the point where a shorter wavelength became cheaper than continuing to slice a longer one.

EUV uses 13.5 nm light, roughly fourteen times shorter than the 193 nm of immersion DUV. The first-generation Low-NA platform, the TWINSCAN NXE:3600D, has a numerical aperture of 0.33 and a single-exposure resolution of around 13 nm. That is fine enough for the most demanding layers at the 7 nm, 5 nm and 3 nm gate-all-around (GAA) nodes without the multi-patterning cycle counts DUV would have needed. It costs roughly $180 million, more than double a DUV immersion scanner. For the specific layers where DUV would need triple or quadruple patterning, one EUV exposure can be the cheaper and far less overlay-prone path to the same pattern.

EUV did not replace DUV in the fab. Dry krypton fluoride systems, such as the TWINSCAN XT:1060K and XT:1460K at roughly 65 nm resolution, remain the workhorse for mature and legacy nodes. Immersion argon fluoride machines, led by the NXT:2050i at 1.35 NA, carry the advanced logic and memory work described above. A leading-edge fab's lithography floor is therefore a stack: DUV dry, DUV immersion, Low-NA EUV and increasingly High-NA EUV, each assigned to the layers where its point on the resolution-versus-cost curve makes sense. At 3 nm the book puts 20 to 28 of a wafer's 60 to 90 mask layers on EUV and the rest on DUV. Only a minority of layers on any leading-edge die ever touch an EUV scanner.

Making light that no laser emits

Producing EUV light is a substantially harder problem than producing DUV light, and the difficulty starts with a plain fact: no laser emits directly at 13.5 nm. ASML instead uses laser-produced plasma. A high-power carbon dioxide laser, delivering roughly 20 kilowatts of pulsed power, strikes a stream of molten tin droplets, around 50,000 of them fired per second. Each droplet is heated into a plasma hot enough to radiate across a broad spectrum that includes the 13.5 nm band the optics are tuned to collect. The book describes this as firing tens of thousands of laser pulses at tens of thousands of tin droplets every second, continuously, for the whole of a production run, inside a vacuum, at industrial throughput. There is no real precedent for that in earlier lithography generations.

Once the light exists, it cannot be steered the way DUV light is. No material passes 13.5 nm light without absorbing nearly all of it, which rules out every refractive lens a DUV or visible-light system would use. EUV scanners rely entirely on mirrors, built from multilayer coatings tuned to reflect a narrow band around 13.5 nm through constructive interference across many alternating thin films. Each mirror reflects only a fraction of the light that hits it, and the optical path involves several reflections between source and wafer. The source therefore has to be powerful enough to survive the cumulative loss and still deliver a usable dose. That is why source power, not just resolution, has been one of the industry's persistent engineering bottlenecks.

The absorption problem extends to the atmosphere. Ordinary air soaks up 13.5 nm light almost as fully as a solid would, so the whole optical path, from plasma source through every mirror to the wafer stage, is held under vacuum. That single requirement cascades into much of what makes the scanner enormous and mechanically exacting. Wafers, reticles and the mirror assembly all have to be moved, aligned and exposed in vacuum with the same sub-nanometre positioning precision a DUV system achieves in open air. The scanner also does not work alone. It is docked to a track system that spin-coats resist and an anti-reflective layer onto the wafer just before exposure and handles the post-exposure bake and development just after. The book names Tokyo Electron's CLEAN TRACK LITHIUS Pro EUV as the dominant platform in that role.

High-NA: paying again to avoid multi-patterning

Low-NA EUV was never going to be the final form. The resolution equation keeps asking for smaller features, and multi-patterning inside EUV is as expensive a fallback as it was inside DUV. ASML's answer is High-NA EUV, the TWINSCAN EXE:5000 and EXE:5200 platform. It raises numerical aperture from 0.33 to 0.55 through an anamorphic optical design built with Zeiss around roughly 8-metre mirror optics. Single-exposure resolution falls to around 8 nm, enough for the most demanding layers at the 2 nm, 1.4 nm (A14) and 1 nm GAA and complementary FET nodes without reintroducing EUV double patterning.

The jump could not be made by scaling up the symmetric optics of the Low-NA machine. An anamorphic design magnifies the mask pattern by a different factor along one axis than the other, instead of the uniform reduction used in Low-NA and DUV. That carries implications for reticle and exposure-field planning, and I should be straightforward that the book flags the size of that trade-off as outside its source set. I will not estimate it here.

The physical scale of the machine is hard to picture. The book puts a High-NA system at more than 150 metric tons, shipped in roughly 250 crates. Current production throughput on the EXE:5200B is 175 wafers per hour. A figure above 200 wafers per hour is a roadmap target, not something shipping today.

On price the sources give two figures: more than $350 million, derived from ASML filings, and roughly $380 million, from ASML's own product material. I read these as a working range of about $350 million to $380-plus million per system. List prices for something this configurable vary with the EXE:5000 versus EXE:5200 build, the options bundle and the date of the quote, and neither source breaks the number down. Against Low-NA's roughly $180 million, that is nearly a doubling.

The logic is the same as the earlier jump. Pay roughly double the previous generation's tool cost to buy back single-exposure patterning at a node where the alternative is double patterning, with its overlay risk and cycle time. Whether that trade keeps working beyond this generation depends on how expensive k1-driven multi-patterning at 0.55 NA becomes at volume. The industry has not had to answer that yet, because High-NA systems only began shipping around 2023 to 2024 and remain early in ramp.

One supplier, no alternative

Everything above converges on a structural fact. ASML Holding is the only company in the world that makes and sells EUV lithography systems, a verified 100 percent global share across both Low-NA and High-NA. The resolution physics, the light source and the anamorphic optics are not problems the industry has solved collectively. One firm has solved them, on a timeline and at a shipment volume that it alone controls.

Set against the rest of the wafer fab equipment oligopoly, ASML stands apart. In its own category, market share and market existence are the same number. The other four vendors compete against real rivals inside a shared technology base. A fab that cannot get an ASML EUV scanner has no alternative process route to the leading edge, only the slower and more expensive DUV multi-patterning path back to the same dimensions.

The company behind that position is large and growing. ASML reported 32.7 billion euros in net sales and 9.6 billion euros in net income for full-year 2025. For the second quarter of 2026 it reported 9.3 billion euros in net sales and 2.918 billion euros in net income, a 31.3 percent net margin on a 54.0 percent gross margin. It employed 44,175 people across 143 nationalities at the end of 2025.

The scale took four decades. ASML began in 1984 as a joint venture between Philips and Advanced Semiconductor Materials International, shipping the PAS 2000 wafer stepper that same year. It entered its defining optics partnership with Carl Zeiss in 1986, a relationship that still supplies the mirror systems for both Low-NA and High-NA EUV. It became a fully independent listed company in 1995, pioneered immersion lithography and the dual-stage TWINSCAN platform, and invested roughly two decades of R&D in EUV before it became commercially viable. Few competitors, and few national semiconductor strategies, could match that horizon from a standing start.

A queue measured in dozens of machines

The order book shows how tight capacity remains. The book reports typical lead times of 18 to 24 months from purchase order. ASML has guided to shipping more than 60 EUV systems in 2026, including 65 Low-NA units against 44 in 2025, plus roughly 10 High-NA scanners. At that volume each customer allocation is strategically consequential rather than routine. Intel has been reported as the first to ship a high-volume logic product using High-NA EUV, aimed at its 14A node, and SK hynix is expected to deploy two High-NA systems for memory. The first wave of High-NA capacity is going to a handful of named customers, not spreading across the industry.

The chokepoint has layers

ASML's EUV source technology traces to Cymer, the laser-produced-plasma specialist it acquired in 2013. Its mirror and lens systems are still developed jointly with Zeiss under a partnership that is now four decades old. So the monopoly itself rests on a small number of hard-to-replace technology relationships and is not fully vertically integrated. For any government or company hoping to build sovereign or diversified access to leading-edge lithography, the chokepoint is not just ASML. It is ASML, plus the Cymer-derived source it owns outright, plus the Zeiss optics partnership it does not.

ASML also does not sell the scanner alone. It supplies a metrology and inspection portfolio that measures pattern accuracy from R&D through mass production, bundled with computational-lithography software under what it calls holistic lithography. The reasoning is that a scanner's printed resolution is only as useful as the measurement system confirming that it landed where the design intended. ASML's 2025 strategic investment in and partnership with Mistral AI, aimed at embedding AI models into its systems and R&D workflows to improve yield and performance, extends that logic. Computational lithography, the software that pre-compensates a mask pattern for the distortions a scanner will introduce, is becoming a software layer of the chokepoint, not only a hardware one. Negotiating access to ASML's roadmap increasingly means negotiating access to scanner, metrology and software together.

Why export controls attach to this machine

This concentration is exactly why ASML sits at the centre of the export-control architecture around advanced chipmaking. The book states that advanced EUV systems, along with etch and deposition tools capable of enabling sub-14 nm logic or high-layer-count 3D NAND, fall under strict multilateral export controls coordinated among the United States, the Netherlands and Japan, under extensions of the Wassenaar Arrangement framework.

The logic is simple. A technology concentrated in one company hands its home government and allies a single, well-defined point through which to regulate who can build the most advanced chips anywhere. If there were three suppliers in three jurisdictions, controls would leak. With one, they are enforceable. The book is careful on one point that I will keep: the specific detail of which system configurations are licensed to which destinations, and how those thresholds have shifted, sits outside its source material and is flagged as unverified. What it does establish is the structure underneath the policy. ASML's monopoly is a commercial chokepoint on scanner supply and also a policy lever with global reach, because nobody, regulator or competitor, has an alternative supplier to route around.

What I take away

Lithography is the one step where the physics gives no room to negotiate. Resolution improves only through shorter wavelength, larger aperture or a lower k1 that has already hit its diffraction floor. The industry chose wavelength first, then aperture, and each move roughly doubled the price of the tool.

Second, EUV is a system, not a light bulb. A 20 kW CO2 laser, 50,000 tin droplets a second, all-reflective multilayer optics and a vacuum path have to work together at production throughput. That explains why the number of suppliers is one, and why 100 percent share here is a description of the technology and not a pricing choice.

Third, the practical constraint for anyone planning a leading-edge roadmap is not money but the queue. About 65 Low-NA and roughly 10 High-NA units in a year, with an 18 to 24 month lead time, means allocation decisions are made years ahead. For policy, the same fact is what makes export controls workable. For India and any other country without a leading-edge fab, the lesson is that access to advanced lithography is negotiated between a few governments and one company, and that is the position from which any domestic ambition has to start.