In June, the U.S. Department of Commerce informed the Dutch company ASML that it suspects one of its extreme ultraviolet lithography machines has reached China, an accusation that ASML denied, and for which no supporting evidence has yet been made public. ASML is the sole global producer of the machines capable of making the most advanced microchips. Whether China obtained a full extreme ultraviolet lithography scanner, major subsystems, or just related components, the strategic question is the same: How much can China gain from physical access to these pieces of the technology?
The U.S. concern presumably spans across several dimensions, but one is particularly important. If a smuggled extreme ultraviolet machine or related components were to reach China, domestic lithography firms might extract design and production knowledge and attempt to reproduce it. Similar concerns emerged in December 2025, when, according to Western media, former ASML engineers working for a Chinese lab reportedly helped produce an extreme ultraviolet lithography prototype by reverse-engineering older machines and studying their components.
Whereas this concern is legitimate, there are important constraints as to what can be gained through direct access to complex technologies such as these machines. Access to an extreme ultraviolet lithography machine could potentially accelerate component- and subsystem-level learning and reveal some architectural choices. However, it cannot transfer all the systems-integration knowledge, the technical understanding and experience accumulated through trials, errors, and processes, the established relationship with suppliers, and the maintenance capabilities required to reproduce and operate the technology at competitive performance and scale.
Sign Up for Our Newsletter
ASML’s Machines
Born in 1984 as a spinoff from the Dutch electronics company Philips, ASML is a critical bottleneck for the AI revolution. It produces extreme ultraviolet lithography machines, which print the finest circuit patterns on the most advanced logic and memory chips, including those that power smartphones, data centers, and AI accelerators. ASML thus sits at the center of the technological competition between the United States and China, and of the export-control regime built around it.
At the leading edge, its position is unrivalled. For chips at five nanometers and below (the nodes used for the most advanced AI accelerators), no current alternative to ASML’s machines exists. It is true that for seven-nanometer chips the Chinese firm Semiconductor Manufacturing International Corporation has relied on older-generation deep ultraviolet machines, printing different patterns multiple times on the same wafer, a technique known as “multi-patterning.” But this approach is economically inefficient and difficult to scale, increases the chances of producing defective chips, and becomes progressively less competitive for the most critical layers at five-nanometer-class nodes and below.
The advantage of extreme ultraviolet lithography machines stems from the performance they can achieve. Whereas such performance is generally discussed in terms of the fine patterns they can print on a semiconductor wafer — the “nanometer nodes” that receive most attention — this is only part of the equation. To print such fine patterns, these machines must operate reliably within extraordinarily narrow tolerances, often smaller than atoms, across thousands of interdependent processes. There is little room for compromise: Small deviations in alignment, contamination, positioning, synchronization, speed, temperature, vibration, or other parameters will degrade yield and can compromise the whole process. These performance requirements, in turn, have led extreme ultraviolet lithography machines to become extraordinarily complex, among the most complex products in the world.
The discussion in the media about ASML has focused primarily (and understandably) on the extraordinary performance of individual components and subsystems of these machines: the laser that strikes tin droplets roughly 50,000 times per second to generate plasma some forty times hotter than the surface of the sun; the mirrors fabricated to picometer precision that, if scaled to the area of Germany, would allow for a deviation of one-tenth of a millimeter; the mask (reticle) carrying the pattern that accelerates as hard as a car reaching 100 kilometers per hour in a fraction of a second; and so forth. Producing each of these components and subsystems is indeed a technical hurdle that could constitute a major industrial challenge in its own right. But this is not the central challenge of producing an extreme ultraviolet lithography machine.
The central challenge is making sure that each of these components and subsystems achieves the required performance while operating within a tightly coupled system whose other components generate heat, vibration, contamination, electromagnetic interference, and other disturbances. That is, the heat and the vibrations resulting respectively from the laser and from the reticle, for example, should not interfere with any of the other components and subsystems. Given the narrow tolerances within which the other subsystems must operate, the problem is therefore not simply to develop the “right” component in isolation. It is to develop components that deliver their intended performance in a demanding environment without compromising the performance of everything around them. Making matters harder, the machine must hold these margins continuously, for months without interruption of high-volume operation, with perturbations such as vibrations from passing traffic capable of pushing it out of tolerance and hence requiring careful isolation and compensation. When those margins are violated, yield, throughput, or reliability can deteriorate rapidly.
Such systems integration, however, is closer to art than to codified science. Much of the knowledge required to achieve reliable performance is accumulated through repeated testing, production failures, supplier interactions, debugging, and hands-on experience over decades inside ASML and its suppliers. This is the core task at which ASML excels.
To meet these performance requirements and remain within the stringent margins for error, the machine’s technical complexity has grown by orders of magnitude, as a result of the increase in the number and of the sophistication of its components, including advanced sensors that detect tiny variations inside the machine in real time, and the corrective systems that act on them. An extreme ultraviolet system is composed of over 100,000 components at the system level — and roughly 700,000 once the parts inside its subsystems are counted — sourced from about 5,000 suppliers, including Zeiss mirrors and Trumpf lasers. ASML co-develops them with these specialized firms, working with them through design and development, providing the technical specifications each component must meet and refining those specifications until a working equilibrium is reached. This massive complexity explains the cost of an extreme ultraviolet lithography machine, on the order of $180 million, with the newest high-numerical-aperture extreme ultraviolet machines costing roughly $380 million.
As a result of these dynamics, it is reasonable to be skeptical that by inspecting an original unit, obtaining its blueprints, or hiring foreign talent, China will be able to replicate these machines within a policy-relevant timeframe. Industrial espionage, reverse engineering, and talent poaching can reveal some architectural choices and, potentially, accelerate the development of a particular subsystem. But they cannot supply all the process knowledge that ensures commercially viable yield, reliability, and throughput. Such knowledge spans multiple technical and operational domains, is tacit, path-dependent, and distributed among countless engineers, managers, scientists, specialized workers, and technicians working in different units of multiple organizations from different countries. Observing or disassembling a machine, obtaining its designs, or hiring individual engineers can transfer only a fraction of the necessary knowledge and understanding: Many processes and procedures are not going to be written down, and no single individual possesses all the knowledge to reproduce an entire system. And in fact, China has so far failed to catch up not only in these machines but also in the previous generation, immersion deep ultraviolet machines from the late 2000s. In contrast to some recent reactions, China is not going to “mass-produce” these machines any time soon: with 5 units scheduled for delivery in 2026 to multiple semiconductor producers, the machines are still at the development stage. Because extreme ultraviolet capability improves through accumulated operating experience, and because ASML has already introduced a new high numerical aperture generation of machines, China might be chasing a moving target as it struggles to master earlier generations.
The Ecosystem and Maintenance Challenges
Even if China produced a functional scanner prototype, it would still face significant limits on the supply chain that makes these machines valuable.
A fully functioning extreme ultraviolet lithography machine cannot produce a single leading-edge chip on its own. It depends on a highly concentrated ecosystem of materials and equipment. Photoresists –the light-sensitive chemicals in which the circuit pattern is actually formed – are a prominent example. Qualified resists for high-volume manufacturing are supplied predominantly by a small group of Japanese-headquartered companies — including JSR, Tokyo Ohka Kogyo, and Shin-Etsu Chemical — even as non-Japanese developers such as Lam Research and Dongjin Semichem pursue alternatives.
Replacing an established supplier is difficult because machine resists must meet demanding and partly conflicting requirements — such as in resolution, sensitivity, and line-edge roughness. Resists must also meet stringent standards for defectivity and outgassing: Molecules released during exposure can contaminate sensitive scanner optics. Consequently, a new resist must undergo extensive qualification with the exposure tool and the surrounding process — including coating, development, etching, and inspection — before it can be adopted in high-volume manufacturing.
The same pattern is reflected across the supply chain. The coaters and developers that apply and process the photoresist are a near-monopoly. Tokyo Electron holds essentially 100 percent of the market for extreme ultraviolet photoresist coaters and developers. Mask blanks, the defect-free reflective substrates onto which circuit patterns are written, are made by two or three companies in the world, including Hoya and AGC. Extreme ultraviolet-grade metrology and inspection is similarly concentrated. KLA Corporation leads the broad process-control market, while Lasertec holds a near-monopoly on the actinic mask-inspection tools that operate at the same 13.5-nanometer wavelength.
China is of course trying to develop its own supplier ecosystem. The problem is that these inputs are co-developed with the machine and specified to it. The photoresist is formulated for a particular exposure dose, wavelength, and outgassing behavior, the mask blank for a particular flatness and defect density, the tin for a particular purity. Machine and consumables are tuned to one another over years, with similar atomic-scale tolerances. Even if China reproduced the scanner and set out to supply its own resists, masks, and chemicals, it would confront the integration problem a second time, now spread across the entire chain: It would have to develop a matched set of materials, tools, processes, and specifications capable of operating together reliably.
Setting production aside, there remains the problem of keeping the machine running. An extreme ultraviolet system must be calibrated and adjusted continuously, and it depends on an uninterrupted flow of spare parts, software updates, and hands-on servicing from ASML. ASML embeds a workforce of roughly 10,000 customer-support engineers inside its customers’ fabs around the world, working twelve-hour shifts day and night — what one ASML executive describes as the company’s “front-line troops.”
This expertise cannot be transferred in its entirety with the hardware. A customer-support engineer trains for close to a year, rotating through ASML’s centers in Taiwan, the United States, and Germany, before being cleared to work on a machine unsupervised. Diagnosis is often embodied and intuitive rather than codified. In one account, an engineer caught an incipient failure by feeling a barely perceptible distortion in a cooling hose — a warp that, had the hose burst, would have brought the entire tool to an immediate halt.
Implications and Recommendations
ASML’s machines achieve performance that is difficult to fathom, so much so that the machines have been the subject of documentaries and investigations. Reliably and systematically achieving that performance requires an extraordinarily complex machine, comprising individually complex components sourced from a complex ecosystem of suppliers. This triple complexity is what limits China’s ability to take shortcuts, and points to some important implications for policy.
First, talent is a critical bottleneck. China is recruiting engineers and scientists with experience. Such knowledge will at best help with the development of critical components and subsystems without showing how to make the whole machine work. Allied governments should nonetheless develop retention and counter-recruitment efforts to ensure that, even at the level of components and subsystems, access to critical individuals is restricted.
Second, a system of hundreds of thousands of components contains countless individual points of failure, and each one is also a point of leverage. The United States and its allies should take advantage of those chokepoints — such as the actinic mask-inspection tools and the photoresists — and calibrate controls and protective effort there, so as to force Chinese espionage, research, and talent efforts to disperse further.
Third, China’s drive to acquire foreign technology through espionage, talent recruitment, and state-directed acquisition is well documented and central to its industrial strategy. But espionage is not free. It requires effort, and effort imposes trade-offs. Western governments should consider turning that reliance against itself, distracting key engineers and scientists and inducing them to spend their scarce time on activities that yield little — above all, the pursuit of technologies that cannot in practice be reproduced. In other words, if China is determined to copy, Western governments have an interest in ensuring it spends its effort copying rather than building the accumulated experience that catching up actually requires.
Fourth, because recreating this capability is so difficult, leapfrogging, or investing in a next-generation technology where no incumbent has an advantage yet, becomes China’s most rational bet. There are already candidates under active investigation in China: alternative extreme ultraviolet light sources based on particle accelerators and on steady-state microbunching (a concept pursued at Tsinghua University, now moving from laboratory demonstration toward a dedicated facility), which would sidestep the tin-droplet plasma source that ASML and Cymer have spent years perfecting.
None of these are close to displacing these machines today, and each faces integration and ecosystem hurdles of its own. But each also resets the experience curve on which ASML’s lead depends. Allied governments should therefore treat frontier alternatives, not just the existing chokepoints, as a domain of competition, and work to lock China out of them early.
Over the past ten years, analysts, decision-makers, and scholars have come to realize the importance of technological competition, a realm that, for a long time, was largely neglected. That neglect imposed severe costs on many Western countries, especially with respect to production. Yet, production remains critical for our society. Understanding how production works, the challenges it poses, the dynamics it triggers, and the peculiarities it generates remains critical for analysts, scholars, and decision-makers.
Write for Cogs of War
Andrea Gilli is a lecturer at the University of St. Andrews in Scotland and has provided consulting services to the Office of Net Assessment of the U.S. Department of Defense.
Mauro Gilli is a professor of Military Strategy and Technology at the Hertie School in Berlin and has provided consulting services to the Office of Net Assessment of the U.S. Department of Defense.
Francesco Fimiani is a former research assistant at the Center for International Security of the Hertie School, and a freelance analyst covering military procurement, emerging technologies, and European defense.
Nicolas Deriquehem is a former research assistant at the Center for International Security of the Hertie School and currently an intern at the United Nations Office to the African Union.
**Please note, as a matter of house style, War on the Rocks will not use a different name for the U.S. Department of Defense until and unless the name is changed by statute by the U.S. Congress.
Image: Midjourney
