Generative AI software requires a semiconductor stack that is increasingly constrained at multiple levels: Nvidia dominates GPU designs.8 TSMC dominates advanced chip manufacturing.9 High-bandwidth memory is dominated by SK Hynix, Micron, and Samsung. But another critical bottleneck is one layer deeper: the equipment used to fabricate the most advanced chips. At the center of that other layer is a company known as ASML.
Based in the Netherlands, ASML (the acronym name is derived from Advanced Semiconductor Materials Lithography) is Europe’s most valuable technology company. It produces approximately 90% of all lithography tools and is the only supplier of extreme ultraviolet (EUV) systems, a critical manufacturing technology.29 ASML has a market value of approximately $677 billion, with more than $37 billion in FY 2025 revenues, $11 billion in profits, and 44,000 employees. But the company’s de facto monopoly is not based on a single breakthrough innovation. Rather, it results from decades of foresighted investment combined with extraordinary platform leadership and ecosystem management skills.11 No competitor has yet been able to reproduce the core technology or the full stack of system capabilities needed for EUV lithography.
Since 2019, following U.S. policy, the Netherlands has prevented export to China of the most advanced ASML machines. This ban has provoked in China a Manhattan Project-type of response. This column examines how ASML has come to dominate EUV lithography, why the company and the technology have been so difficult to replicate, and what China is doing about it.
Technology and Company Origins
Photolithography uses light to “print” billions of microscopic circuit patterns onto silicon wafers treated with photosensitive coatings. The technology resembles photography and dates back to the 1950s, invented for the U.S. military and commercialized by firms such as Fairchild Semiconductor and Texas Instruments. By the early 1980s, Geophysical Corporation of America (GCA) had as much as an 85% market share in lithography machines. By the later 1980s, however, Nikon and Canon overtook GCA (which shut down in 1993) and came to dominate the industry.19,38
ASML began in 1984 as a joint venture between Philips, the Dutch electronics conglomerate founded in 1891 (and an early investor in TSMC), and ASM International, a small Dutch semiconductor manufacturing equipment firm founded in 1964. The goal was to commercialize Philip’s in-house photolithography technology. However, ASML’s first product, the PAS 2000 (PAS stands for “Philips Automatic Stepper”), was unreliable and a failure.35 In 1988, ASML became independent. Its next product line, the PAS 5500, introduced in 1991, was highly successful, using a modularized architecture, precision optics from Carl Zeiss AG of Germany (based on a partnership started in 1983), and “near” ultraviolet light to produce denser chip designs.2,39 PAS 5500 in the mid-1990s added “Deep Ultraviolet” (DUV) capabilities, which the Japanese were also perfecting.
The challenge for European and U.S. lithography producers in the 1990s was to find a way to leapfrog Nikon and Cannon. One way was to establish an early leadership position in the next generation of the technology.12 DUV had physical limits and experts believed the shorter wavelengths and higher photon energy of EUV would be necessary to produce denser chips. The problem was how to generate and control EUV light for mass production. There were other candidate technologies (X-rays, electron beams, ion beams, and charged particles), but none offered the potential speed and cost advantages of EUV.17
Successful commercialization of EUV lithography is one of the greatest stories in the history of science, engineering, and business. That said, the success took roughly two decades, $11 billion in R&D, and a global effort coordinated by one firm—ASML.24 The history goes back to 1985, when Hiroo Kinoshita, a researcher at Japan’s NTT labs, adopted multilayer mirror research from the Soviet Union to project the first EUV images. He used Mercury-vapor lamps, which produced ultraviolet light of 365 nanometers (nm). Kinoshita’s presentation at a 1986 Japanese optics conference was met with skepticism but prompted other researchers around the world to study EUV lithography.13
By 1994–1995, ASML executives had concluded EUV was indeed the future of lithography. Martin van den Brink (later president and CTO), a technology executive with a strong physics background, had led development of the PAS 5500. He championed the effort to explore EUV and put another senior manager, Jos Benschop, in charge.13,34 In 1998, ASML and Zeiss formed a European industrial consortium named EUCLIDES for “Extreme UV Concept Lithography Development System.”13 Similar consortia appeared in the U.S. and Japan, but only EUCLIDES succeeded.
In 2000, convinced they could overcome the scientific hurdles, ASML created a small team to build a prototype EUV stepper. ASML also benefitted from U.S. military R&D after it bought the last surviving U.S.-based lithography company, Silicon Valley Group (SVG), also in 2000.19 The prototype took six years, completed in 2006. ASML worked closely with key suppliers and universities as well as IMEC, the Interuniversity Microelectronics Centre in Belgium, which served as a laboratory and equipment testing ground for the European semiconductor industry. In 2009, ASML opened a new EUV R&D facility at its headquarters in Veldhoven, Netherlands.13
Another important milestone occurred in 2012. Three key customers—Intel, along with Samsung and TSMC—agreed to become investors in ASML to share the R&D costs and help resolve the technical challenges.13 Intel’s investment of $4 billion (a 15% stake—several times larger than TSMC and Samsung) was especially significant.18 Intel was the world’s largest semiconductor manufacturer and this decision signaled a major commitment to EUV. ASML began shipping EUV prototypes in 2010 but needed help to produce a commercially viable system, which it did by 2013. Machines perfected for mass production took until 2017 and were priced at $100 million.3,29
Despite its early backing of EUV, Intel decided to delay commercial introduction until the next generation because of the high cost. Unfortunately, this delay (until 2025) caused Intel to fall behind TSMC and Samsung in advanced manufacturing capabilities. It also has led to severe financial and reputational difficulties, from which Intel is only now starting to recover.1,7 Samsung would go on to ship the first EUV-enabled product (using what is referred to as a “7nm” node process) in August 2019 for its Galaxy Note 10 smartphone.13 TSMC shipped its first EUV-enabled products (also with the 7nm node process) in October 2019 for multiple customers, including Huawei in China.30
Why ASML’s EUV Machines Have Been So Difficult to Copy
In semiconductor manufacturing, to shrink circuit feature sizes depends fundamentally on the wavelength of light used in lithography: shorter wavelengths enable finer patterns and denser packing of transistors. For decades, the industry has relied on DUV systems operating at 193nm, and these machines are still widely used. EUV reduced the potential wavelength to 13.5nm. This is a dramatic improvement, though mass production required significant advances in a range of complementary technologies.14
For example, EUV light cannot pass through air or conventional glass, which eliminates traditional lens-based approaches. Instead, ASML’s EUV machines rely on highly specialized multilayer mirrors, fabricated with sub-nanometer precision by Zeiss, to reflect and focus the EUV light. Generating and controlling EUV light is another complex physics problem. ASML uses a high-powered laser produced by TRUMPF GmbH, a German company, to strike microscopic tin droplets at 50,000 times per second. This creates a plasma about 200,000 degrees centigrade (40 times hotter than the surface of the sun) that emits the required wavelength.23 The laser module alone is said to contain 457,329 parts.19
The wafer masks are another critical component. Semiconductor foundries create the masks but work closely with electronic design automation (EDA) tools companies such as Cadence and Synopsys. The chip manufacturers use EDA tools to generate data for the 50 or so alternating layers of silicon and molybdenum in each mask, which must be perfectly compatible with the EUV optics, light system, and the stepper’s numerical aperture.10 To make sure each EUV machine can print the mask patterns and produce functioning semiconductor devices also requires special test engines and many rounds of simulations, done by the EDA firms and the foundries.33 Other firms handle final assembly and packaging of the semiconductor devices, with processing steps almost as complicated as the core EUV technology.6
The challenge for potential competitors is that no one component defines performance of the whole system. The light source, optics, wafer, mask, and control software all come from different specialist firms, and their modules must all function as a tightly coupled system and operate in a vacuum under extreme thermal, mechanical, and computational constraints. Each component has required scientific and engineering advances in different fields, achieved over decades, in plasma physics, precision optics, mechanical engineering, materials science, and software engineering. ASML also holds the rights to thousands of patents that protect its technology.5
EUV machines weigh approximately 400,000 pounds, are the size of a double-decker bus, and require three Boeing 747s to transport.20 ASML charges around $200 million for less sophisticated versions and $400 million for the most advanced versions (“2nm” and denser node processes). Each system contains more than 100,000 modules from some 800 suppliers, which have thousands of additional suppliers in their supply chains. ASML’s strategy has been to outsource 90% of the modules and act as the R&D leader and system integrator.20,28 This strategy leveraged the R&D and capabilities of many outside firms, but has limited ASML’s control over manufacturing capacity. In 2025, ASML shipped just 48 EUV systems out of 327 ASML lithography machines of all types.16 The company plans to ship 60 EUV systems in 2026 and 80 annually in future years.15
EUV machines and advances in chip design have enabled the semiconductor industry to continue on the path laid out by Moore’s Law and produce increasingly advanced microprocessors, GPUs, and memory chips. The price tag limits the market for EUV machines but, as Intel discovered, due to their power, precision, and speed, using EUV to make the densest chips has become cheaper than workarounds with creative three-dimensional chip designs and DUV equipment.34
The cost and complexity of ASML’s EUV machines help explain why the industry’s leading manufacturing firms—Intel, TSMC, and Samsung—chose not to develop EUV lithography themselves but instead to invest in ASML. The cost and complexity also help explain why companies that mastered DUV lithography—particularly Nikon and Canon—gave up on EUV. They also fell behind ASML in advanced DUV machines, where ASML holds between a 60% and 85% market share.20,31 It probably had some impact that the Japanese did not have direct access to U.S. military and national laboratory research, which ASML did through its SVG acquisition.19
Future Competition: Spotlight on China
ASML illustrates a unique type of monopoly—a dominance rooted not in a single proprietary invention but in a technology platform and ecosystem strategy used to integrate a complex system of components and skills provided by hundreds of partners, suppliers, and customers. ASML subsidiaries also maintain tight control over critical technologies: HMI—mask pattern verification; Cymer—light sources; and Berliner Glas—optics.25 This platform-ecosystem structure is useful to understand why ASML’s main customers became investors and why competitors face a challenge that looks increasingly like a nation-state effort. The Russians have been trying to develop EUV technology for years but appear to be as much as two decades behind ASML.4 The Chinese are more advanced and investing heavily, spending an estimated $50 billion per year on semiconductor manufacturing technology.25
China’s leading foundry, Semiconductor Manufacturing International Corporation (SMIC), founded in 2000 as an attempt to build a TSMC-like foundry in China, had been a large ASML customer.19 However, SMIC and other Chinese firms since 2019 have been prohibited from importing not only EUV machines but also the latest DUV machines produced by ASML as well as Canon and Nikon. SMIC and other foundries have extended DUV technology by using “multi-patterning” to add extra lines on the masks with multiple passes. Though costly and slow, this approach has enabled the production of 7nm and 5nm node devices, which SMIC has done mainly for Huawei.10,27
To produce its own EUV machine, China launched a project that is supposed to be a state secret. As reported by Reuters, in early 2025, former Chinese ASML engineers completed a prototype at a Shenzhen laboratory using components from older ASML machines available in secondary markets. The prototype is operational and undergoing testing. Observers believe it will take until at least 2030 for commercial production.22,26,37
China’s effort differs from simple reverse engineering. The Central Science and Technology Commission is overseeing the project, with Huawei as the system integrator. The Shanghai Institute of Optics and Fine Mechanics (SIOM) as well as several other private companies are involved as well. They are using solid-state lasers for EUV light generation, an approach ASML determined was inadequate for mass production. The Chinese still need Zeiss mirrors (also banned for export to China) or a substitute. In addition, Huawei is developing a Laser-Induced Discharge Plasma (LDP) technology that uses high voltages to generate extreme ultraviolet light.32 LDP should use less energy than ASML’s tin-based approach but still generate the 13.5nm EUV light.36
Perhaps the biggest challenge for China is to produce EUV machines that are suitable for mass production, though initial cost is probably a secondary concern. For example, for 7nm chips using DUV technology, SMIC reportedly had yields less than 50% and wafer costs 40% to 50% more than TSMC, but at least it was able to make these advanced devices.32 Another challenge is that the target is moving: ASML and its customers are already introducing technology for sub-2nm fabrication.21,29 At some point, the laws of physics will limit ASML’s approach and a better technology will emerge, but we are not there yet.