The Most Expensive Photograph Ever Taken
There is only one company on Earth that can make the machine that makes the machine that makes the chip that makes your phone think. That company operates out of Veldhoven, a town of 45,000 in the southern Netherlands, surrounded by flat farmland and cycling paths, and it is — by any reasonable measure — one of the most strategically consequential businesses in human history. ASML Holding N.V. does not design microchips. It does not fabricate them. It builds the lithography systems that print the impossibly fine circuitry onto silicon wafers, and at the bleeding edge of that discipline — extreme ultraviolet lithography, or EUV — it has no competitors. Zero. The $350-million-plus machines it ships, each requiring three Boeing 747s to transport, are the most complex devices ever commercially manufactured, containing over 100,000 parts, operating in a vacuum with mirrors polished to a flatness measured in atoms, and generating light at a wavelength of 13.5 nanometers by firing a laser at 50,000 droplets of molten tin per second. Every advanced chip powering AI inference, every GPU training the latest large language models, every leading-edge processor from TSMC, Samsung, and Intel — all of them pass through an ASML EUV system. The company's market capitalization has exceeded €250 billion. Its gross margins hover above 50%. And the entire enterprise began in a leaky wooden shed.
The paradox at the core of ASML is this: it is simultaneously the most monopolistic and the most dependent company in its industry. It has 100% market share in EUV lithography and roughly 90% of the advanced deep ultraviolet (DUV) segment, yet its existence depends on a supplier ecosystem so specialized and so fragile — Carl Zeiss SMT for mirrors, Cymer (now a subsidiary) for light sources, Trumpf for lasers — that a fire at a single subcontractor's facility could delay the entire global semiconductor roadmap. It sells to perhaps five meaningful customers, three of whom — TSMC, Samsung, Intel — account for the overwhelming majority of its revenue. It is a monopoly that lives at the mercy of an oligopsony. And now, caught between Washington's determination to deny China access to advanced chipmaking technology and Beijing's determination to acquire it, ASML has become the most geopolitically contested capital equipment company since — well, since ever.
By the Numbers
ASML at Scale
€28.3BNet sales, FY2024
51.3%Gross margin, FY2024
~44,000Employees worldwide
100%Market share in EUV lithography
€4.4BAnnual R&D spending (estimated)
~€250B+Market capitalization (early 2025)
150+Nationalities represented in workforce
1984Year founded
Born in a Shed, Saved by a Phone Call
The founding mythology of ASML is almost too perfectly cinematic to be true. In 1984, Philips — the Dutch electronics conglomerate that had given the world the compact disc, the cassette tape, and the Philishave razor — entered a joint venture with Advanced Semiconductor Materials International (ASMI) to develop lithography systems for the semiconductor industry. The resulting entity, ASM Lithography, was housed in a prefabricated shed on the Strijp T industrial campus in Eindhoven, adjacent to Philips' research laboratories. The roof leaked. The ambitions did not.
That same year, the small team shipped its first product: the PAS 2000 stepper, a machine that projected circuit patterns onto silicon wafers using ultraviolet light. The technology built on Philips' optical research dating to the early 1970s, and the market it entered was brutally competitive. Nikon and Canon, flush with precision optics expertise from their camera businesses, dominated lithography. GCA Corporation in the United States had pioneered the original wafer stepper. ASML was an afterthought — a small, underfunded Dutch startup in a market that did not need another entrant.
Growth came fast but profits did not. By 1985, the company had 100 employees and a new factory in Veldhoven. In 1986, it launched the PAS 2500 stepper with novel alignment technology and, critically, formalized a partnership with Carl Zeiss for optical lenses — a relationship that would become one of the most consequential supplier agreements in industrial history. But the late 1980s were savage. The global electronics industry slumped. ASMI, bleeding cash from its investment, decided to withdraw. Philips, undergoing its own vast cost-cutting program under pressure from declining semiconductor revenues, considered letting the venture die.
What happened next is the kind of moment that separates companies that exist from companies that become essential. ASML's executives, desperate for funds and guided by what the company's own histories describe as "a strong belief in the ongoing R&D," reached out to Henk Bodt, a member of the Philips board of management. Bodt — an engineer by training who understood the strategic value of lithography even as the financial case looked terminal — persuaded his colleagues to extend one final lifeline. The investment held. Within months, ASML launched the PAS 5500, a breakthrough platform whose industry-leading productivity and resolution won the key customers the company needed to turn its first profit.
There is a beneficiary of that capital inefficiency, and that's us.
The PAS 5500 was more than a product. It was proof of concept — evidence that a company with fewer resources than Nikon and Canon could out-innovate them through architectural choices rather than brute capital. ASML's systems were designed around a modular architecture, sourcing best-in-class components from specialist suppliers rather than manufacturing everything in-house the way its Japanese competitors did. Zeiss made the lenses. ASML designed the systems around them. This decision — outsource the components, own the integration — would become the defining strategic principle of the company, and one of the great case studies in industrial organization.
The Architecture of Dependence
To understand why ASML won the lithography wars, you must first understand what lithography is and why it is so ferociously difficult.
A lithography system is, at its most elemental, a camera in reverse. Instead of capturing an image, it projects one — taking a circuit pattern drawn on a glass template called a reticle and shrinking it down to nanometer scale, then etching that pattern onto a photosensitive layer coating a silicon wafer. The light source determines the minimum feature size: shorter wavelengths allow finer patterns, which allow more transistors per chip, which is the engine that has driven
Moore's Law for six decades.
For most of the semiconductor industry's history, lithography used deep ultraviolet (DUV) light at wavelengths of 248 or 193 nanometers. Each successive generation of DUV systems pushed closer to the physical limits of what those wavelengths could resolve. By the early 2000s, the industry faced a wall: 193-nanometer light could not, through conventional means, print the features required for the next generation of chips. Two solutions emerged. One was immersion lithography — placing a thin layer of water between the lens and the wafer, which increased the effective numerical aperture and allowed finer patterning. The other was extreme ultraviolet lithography, which would use light at 13.5 nanometers — more than fourteen times shorter than DUV — to achieve a quantum leap in resolution.
ASML bet on both. And it won both bets.
The immersion story came first. In the early 2000s, Nikon and ASML were locked in fierce competition for the next-generation DUV platform. ASML, working with TSMC and leveraging a concept first proposed by a Taiwanese researcher, developed immersion lithography systems that fundamentally altered the competitive landscape. The TWINSCAN platform, which processed two wafers simultaneously — one being measured while the other was exposed — gave ASML a productivity advantage that Nikon could not match. By the mid-2000s, ASML had overtaken Nikon in market share for advanced lithography. Nikon never recovered. Canon retreated to the mature, less-demanding segments. ASML's share of the advanced DUV market climbed toward 90%.
But the real bet — the company-defining, industry-reshaping, geopolitics-altering bet — was EUV.
Twenty Years in the Wilderness
EUV lithography was not ASML's idea. It was, in a sense, nobody's idea and everybody's idea — a concept that emerged from research at the U.S. Department of Energy's national laboratories in the late 1980s and early 1990s, initially in the context of nuclear weapons research and space-based optics. Scientists at Lawrence Livermore, Sandia, and Lawrence Berkeley national laboratories realized that extreme ultraviolet light, generated by plasma sources, could theoretically be used to print semiconductor features far smaller than DUV could achieve.
The problem was that EUV light is absorbed by virtually everything — air, glass, conventional lenses. An EUV lithography system cannot use traditional refractive optics. It must operate in a near-perfect vacuum, and all its optics must be reflective: mirrors, not lenses, coated with dozens of alternating layers of molybdenum and silicon, polished to a smoothness where a single atomic imperfection degrades performance. The light source itself had to be reinvented. And the entire system had to achieve the throughput — the number of wafers processed per hour — that chipmakers demanded for economical mass production.
The U.S. semiconductor industry, through a consortium called EUV LLC formed in 1997, initially tried to develop the technology domestically. Intel, Motorola, AMD, and the national labs pooled resources. But the development costs were staggering, the technical challenges cascading, and the timeline stretched from years to decades. By the early 2000s, the consortium's work had produced fundamental breakthroughs in mirror coating and light source design, but no one had a viable commercial system.
ASML entered the EUV arena in earnest in the late 1990s, licensing technology from the U.S. consortium and beginning its own development program. Nikon pursued EUV as well, but with less conviction and fewer resources. Canon opted out. What followed was a two-decade odyssey — the longest, most expensive, and most technically audacious R&D program in the history of semiconductor equipment.
🔬
The EUV Development Timeline
From lab concept to high-volume manufacturing
1988EUV concepts first explored at U.S. national laboratories.
1997EUV LLC consortium formed by Intel, AMD, Motorola, and U.S. national labs.
2006ASML ships first EUV prototype (alpha demo tool) to customer sites.
2010ASML ships NXE:3100 pre-production EUV system.
2012Customer Co-Investment Program: Intel, TSMC, Samsung invest in ASML equity and fund EUV R&D.
2013ASML acquires Cymer (EUV light source maker) for approximately $3.7 billion.
2017ASML ships NXE:3400B — first EUV system used in volume production.
The numbers tell a story of almost irrational persistence. ASML invested billions of euros over two decades, years before EUV generated meaningful revenue. The light source problem alone — generating enough EUV photons at sufficient power and stability to achieve production-worthy throughput — nearly killed the program multiple times. Cymer, the San Diego-based company that developed EUV light sources using laser-produced plasma (the tin droplet method), struggled for years to achieve the 250-watt power levels needed for mass production. ASML eventually acquired Cymer outright in 2013 for approximately $3.7 billion, a deal that brought the most critical and problematic component of the EUV system under ASML's direct control.
The mirrors presented their own order of impossibility. Carl Zeiss SMT, ASML's exclusive optics partner, had to produce mirrors with surface roughness measured in picometers — billionths of a meter. As NPR described it: if an EUV mirror were scaled up to the size of Germany, the largest bump on its surface would be less than a millimeter tall. These are, by some measures, the smoothest objects humanity has ever produced.
The Co-Investment Gambit
By 2012, ASML was in a peculiar position. It was the undisputed leader in DUV lithography, generating strong profits from its TWINSCAN platform, and the most advanced EUV development program in the world was underway in Veldhoven — but the costs were immense, the timeline uncertain, and the risk of failure non-trivial. The semiconductor industry needed EUV to continue shrinking transistors, but no single company could bear the development cost alone.
ASML's solution was as innovative financially as its products were technically. In July 2012, the company announced the Customer Co-Investment Program, an extraordinary arrangement in which its three largest customers — Intel, TSMC, and Samsung — would invest directly in ASML's equity and contribute funds to its EUV and 450mm wafer R&D programs. The structure was unprecedented in semiconductor equipment: customers buying equity in their sole-source supplier, effectively sharing the risk of a technology that all three needed but none could develop independently.
Intel committed €829 million for a 10% stake, with an additional €553 million in R&D funding. TSMC invested €838 million for a 5% stake, with €276 million in R&D funding. Samsung followed with €503 million for a 3% stake and €276 million in R&D funding. The combined investment exceeded €3.3 billion, with over €1.1 billion earmarked specifically for EUV development.
The program served multiple purposes simultaneously. It reduced ASML's financial risk during the most capital-intensive phase of EUV development. It locked in customer commitment — these were not contracts but equity positions, aligning incentives at the ownership level. And it sent an unmistakable signal to the broader market: the three companies most capable of judging whether EUV would work had put real money on ASML's ability to deliver it.
The corporate governance provisions were carefully drawn. Participating customers received limited information rights and the ability to nominate board observers, but no board seats and no veto power over ASML's strategy. Voting rights were capped. Standstill and lock-up provisions prevented any customer from building a controlling stake. ASML remained independent — funded by its customers, directed by its own management.
It worked. The capital injection, combined with ASML's own escalating R&D spending, funded the final push to production-ready EUV. When the NXE:3400B began shipping in volume in 2017, TSMC used it to introduce EUV into high-volume manufacturing at the 7nm node in 2019. Samsung and Intel followed. The bet had paid off — not just for ASML, but for the industry's roadmap.
The Physics of Monopoly
ASML's monopoly in EUV lithography is not a market outcome in the conventional sense. It is a physics outcome. The technical barriers to entry are so extreme, the required supplier relationships so irreproducible, and the cumulative learning so deep that no competitor has been able to mount a credible challenge in over two decades of the technology's development.
Consider what a rival would need. First, a light source capable of generating 250+ watts of EUV power by firing a CO₂ laser at 50,000 tin droplets per second — technology that Cymer (now ASML-owned) spent decades and billions developing. Second, a set of multilayer-coated reflective optics from Carl Zeiss SMT, a division of the Carl Zeiss Group that has an exclusive, decades-long relationship with ASML and no commercial incentive to supply a competitor. Third, a high-power laser from Trumpf, the German industrial laser company that developed the drive laser specifically for ASML's EUV platform. Fourth, the systems integration expertise to assemble 100,000+ components into a machine that achieves nanometer-level precision while operating in a vacuum at production speeds exceeding 180 wafers per hour.
Each of these capabilities represents decades of accumulated learning. The mirrors alone require a fabrication process so refined that Zeiss reportedly takes months to produce a single set. Trumpf's lasers are custom-engineered for the EUV application. The hydrogen pellicle technology, the reticle handling systems, the computational lithography software that optimizes the patterning process — each is a deep specialization, and ASML has either developed it in-house, acquired the company that built it, or locked in an exclusive long-term supply agreement.
The tools are so expensive. A new EUV tool from ASML is probably $250 million. One tool.
Nikon, the only company that could have plausibly challenged ASML in EUV, effectively withdrew from the leading edge after losing the immersion lithography battle in the mid-2000s. It still sells DUV systems for mature semiconductor nodes and flat-panel display lithography, but its market share in advanced lithography has contracted to single digits. Canon never seriously pursued EUV. There is no Chinese, Korean, or Taiwanese equipment maker with a credible EUV development program, though Chinese efforts — notably at Shanghai Micro Electronics Equipment (SMEE) — have attracted attention and concern. SMEE's most advanced systems reportedly operate at DUV wavelengths comparable to technology ASML commercialized in the 1990s. The gap is measured not in years but in generations.
A San Francisco startup called Substrate has attracted venture capital and press attention for its attempt to reimagine chip manufacturing economics, but it is not building an EUV lithography system. The structural moat around ASML's core product is, for the foreseeable future, impregnable.
The Geopolitical Machine
If you wanted to design a company that would become the most geopolitically contested business on Earth, you would design ASML. A European company, headquartered in a NATO ally, with exclusive possession of the technology required to manufacture advanced semiconductors, selling primarily to customers in Taiwan, South Korea, and — until recently — China. The United States, which incubated the underlying science in its national laboratories and whose technology companies designed many of the chips ASML's machines produce, does not actually make the machines. It relies on a Dutch company to be the gatekeeper.
The inflection point came in October 2022, when the Biden administration imposed sweeping export controls on advanced semiconductor technology to China, targeting not just chips but the equipment used to make them. The Dutch government, under sustained pressure from Washington, subsequently implemented its own restrictions, effective January 2024, preventing ASML from shipping its most advanced DUV and all EUV systems to Chinese customers without export licenses.
The impact was immediate and asymmetric. In 2023, China had become ASML's largest market, accounting for 39% of sales in Q4 of that year as Chinese chipmakers rushed to acquire DUV systems ahead of the anticipated restrictions. This front-loading of demand created a revenue surge — net sales rose to €27.6 billion in 2023, a 30% increase from 2022 — followed by the prospect of a sharp decline in Chinese revenue as the new rules took effect.
The strategic dilemma for ASML is genuinely painful. China represents a massive and growing market for semiconductor equipment, driven by Beijing's determination to build domestic chip manufacturing capacity. ASML's DUV systems — not subject to the most restrictive export controls — still sell into China, but the trajectory of regulation is toward further tightening. Every expansion of the restricted list narrows ASML's addressable market while doing nothing to diminish China's motivation to develop indigenous alternatives. ASML's CEO, Peter Wennink, articulated the tension with characteristic bluntness: China won't accept being cut off from AI chips.
The company is, in a sense, a hostage to its own indispensability. The U.S. and its allies use ASML as a chokepoint because ASML is the only chokepoint that works — the sole source of EUV systems means that denying China access to ASML is functionally identical to denying China access to leading-edge chipmaking. But this geopolitical utility comes at a cost to ASML's business. Lost Chinese revenue is real. And the very act of weaponizing ASML's monopoly creates the strongest possible incentive for China to break that monopoly — to invest whatever it takes, over whatever timeframe, to develop alternative lithography technologies.
Just read the papers, there are chip shortages everywhere. We have seen a significant spurt in terms of customer demand.
Veldhoven's Impossible Machine
Step inside the cleanroom at ASML's Veldhoven campus and you enter a world of controlled paranoia. The manufacturing floor where EUV systems are assembled is one of the cleanest environments on the planet. Technicians wear full-body suits, work in laminar airflow, and handle components with gloves over gloves. A single fingerprint on the wrong surface can cause significant damage to a machine worth hundreds of millions of euros.
The EUV machine itself defies concise description. Here is one attempt: A CO₂ laser, developed by Trumpf and generating roughly 40 kilowatts of power, fires pulses at a stream of tin droplets — 50,000 per second — falling through a vacuum chamber. Each droplet, hit first by a pre-pulse that flattens it into a pancake shape, then by the main pulse, explodes into a plasma that emits extreme ultraviolet light at 13.5 nanometers. This light is collected by a curved mirror, directed through a series of reflective optics manufactured by Carl Zeiss SMT (each mirror reflecting about 70% of the light that hits it, so the total light reaching the wafer is a fraction of what was generated), and focused through a mask containing the circuit pattern onto the photoresist-coated silicon wafer below.
The wafer sits on a stage that moves with nanometer precision. The system exposes a portion of the wafer, then steps to the next position, then the next, hundreds of times per wafer, processing over 180 wafers per hour. Each wafer may undergo dozens of exposure steps as different layers of the circuit are built up. The entire process operates in a vacuum because EUV light is absorbed by air. The mirrors must be kept at stable temperatures to prevent thermal distortion. The tin debris from the exploding droplets must be continuously cleaned from the optical path.
It is, by reasonable consensus, the most complex commercial machine ever built. ASML delivered approximately 50 of its highest-specification EUV models in 2022, alongside roughly 350 DUV and other systems. Each EUV system takes approximately a year to assemble and deliver.
⚙️
Anatomy of an EUV System
Key subsystems and their suppliers
| Subsystem | Function | Supplier |
|---|
| EUV light source | Generates 13.5nm EUV light via laser-produced plasma | Cymer (ASML subsidiary) |
| Drive laser | 40kW CO₂ laser that fires at tin droplets | Trumpf (Germany) |
| Projection optics | Multilayer reflective mirrors that focus EUV light | Carl Zeiss SMT (exclusive) |
| Wafer stage | Precision positioning of wafer during exposure | ASML (in-house) |
| Reticle handling | Loading and positioning of mask/reticle | ASML (Linkou site) |
| Computational lithography software |
The Man Who Sold Machines and Bought Futures
ASML has had only a handful of CEOs in its four decades, and the two who matter most for the company's current form are Eric Meurice and Peter Wennink.
Meurice, a French engineer who had spent years at Intel and then at chip design company STMicroelectronics, became CEO in 2004 and served until 2013. He arrived at a moment when ASML was winning the immersion lithography battle against Nikon but the EUV program was devouring capital with no clear endpoint. Meurice's great contribution was strategic discipline: he pushed the TWINSCAN dual-stage architecture that gave ASML its DUV productivity edge, drove the Cymer acquisition, and — perhaps most consequentially — designed the Customer Co-Investment Program that brought Intel, TSMC, and Samsung into ASML's equity structure. He understood that EUV was not a product development problem but an ecosystem financing problem, and he solved it by turning his customers into investors.
Wennink, a Dutchman with an accounting background who had served as ASML's CFO for more than a decade, succeeded Meurice in 2013 and led the company through the final EUV development push, the technology's entry into high-volume manufacturing, and the onset of the U.S.-China semiconductor conflict. Where Meurice was the architect of the EUV financial structure, Wennink was its political navigator — managing the relationship with Washington while maintaining ASML's commercial relationships in Asia, defending the company's independence against geopolitical pressures that threatened to turn it into an instrument of Western foreign policy. His public remarks — pragmatic, occasionally sardonic, always precise about the numbers — defined ASML's external voice for a decade.
In April 2024, Wennink was succeeded by Christophe Fouquet, a French physicist and long-serving ASML executive who had led the company's EUV business and its holistic lithography strategy. Fouquet's appointment signaled continuity — he was an internal candidate, steeped in the technology, with deep relationships across the customer base. His challenge is the next frontier: High-NA EUV.
High-NA: The Next Impossible Thing
Even as ASML's current EUV systems achieve full market adoption, the company is already shipping the next generation. High-NA EUV — where NA stands for numerical aperture, the measure of a lens system's ability to gather light and resolve fine details — increases the numerical aperture from 0.33 in current EUV systems to 0.55, enabling roughly 60% finer resolution. The first High-NA system, the EXE:5000 (also known as the Twinscan EXE:5200), began shipping to lead customers in 2023-2024.
The price tag is staggering: reportedly north of $350 million per system, with some estimates approaching $400 million. The machine is even larger than its predecessor. The new mirror system from Carl Zeiss SMT, with its larger anamorphic optics, pushed the boundaries of what was physically achievable in mirror fabrication. The EXE:5000 is designed to enable chipmaking at the 2-nanometer node and below — the frontier where TSMC, Samsung, and Intel are locked in a three-way race.
High-NA is the continuation of a pattern that has defined ASML for forty years: investing in the next generation of the seemingly impossible while still reaping the returns from the current generation. The company's R&D spending runs at approximately €4 billion per year, or roughly 15% of revenue — an extraordinary figure for a hardware company, more typical of a pharmaceutical firm at the frontier of drug discovery. The returns, when they arrive, are enormous: EUV system revenue has grown from €4.5 billion in 2020 (31 systems) to many multiples of that figure as adoption has broadened.
The Installed Base: Where Machines Become Relationships
One of the most underappreciated aspects of ASML's business model is what happens after a system ships. An EUV lithography system is not a product you buy, install, and operate independently. It is a machine that requires continuous ASML involvement — software upgrades, maintenance, parts replacement, performance optimization — for its entire operational life, which can span a decade or more. The installed base of ASML systems at customer fabs worldwide represents a growing annuity stream of service and upgrade revenue.
In 2021, when the global chip shortage was at its most acute, ASML's CEO noted that the company had rushed margin-rich software upgrades to customers as a stopgap measure to boost fab output before additional hardware could be delivered. These upgrades — computational lithography enhancements, productivity optimization packages, metrology software — are high-margin, recurring revenue streams that compound as the installed base grows.
The service business also creates profound switching costs. When a chipmaker builds a fab around ASML lithography systems, every subsequent technology node, every process optimization, every yield improvement effort is built on ASML's platform. The computational lithography software — which ASML develops at sites in Silicon Valley, Shenzhen, and elsewhere — is increasingly critical: as features shrink below the wavelength of the light used to print them, the gap between the intended pattern and the physical pattern on the wafer must be corrected computationally. This software is tightly integrated with ASML's hardware, and its value grows with the complexity of the manufacturing process.
ASML has also expanded its portfolio through acquisitions. The 2016 acquisition of Hermes Microvision (HMI), a Taiwanese company specializing in e-beam inspection systems, added a critical capability: the ability to inspect wafers at the nanometer scale to detect defects that optical inspection cannot see. HMI's products — the eScan series, including the world's first multi-beam wafer inspection tool — are increasingly essential as transistor geometries shrink beyond the resolution limits of conventional defect detection.
Our vision is to enable affordable microelectronics that improve the quality of life. To achieve this, our mission is to invent, develop, manufacture and service advanced technology for high-performance lithography, metrology and software solutions for the semiconductor industry.
The Onshoring Windfall
The global semiconductor supply chain's geographic concentration — roughly 85% of ASML's machines shipped to just three markets in 2020 (Taiwan, South Korea, and China) — became a source of strategic anxiety for the United States and Europe after the COVID-era chip shortage exposed the fragility of the system. The result was an unprecedented wave of government-subsidized fab construction: the U.S. CHIPS Act (providing $52 billion in subsidies), the European Chips Act (targeting €43 billion in public and private investment), and similar programs in Japan and elsewhere.
Every new fab needs lithography equipment. TSMC's $40 billion+ Arizona complex, Intel's Ohio fabs, Samsung's Texas plant — all of them will require ASML systems. The onshoring trend represents a structural tailwind for ASML that is largely independent of the semiconductor demand cycle: even if chip demand were flat, the duplication of manufacturing capacity across new geographies would drive equipment orders for years.
Wennink described this dynamic with characteristic economy: "It's the decoupling of a worldwide ecosystem. There is a beneficiary of that capital inefficiency, and that's us." The capital inefficiency he referred to is real — building redundant fab capacity in multiple countries is wasteful from a pure efficiency standpoint — but it creates a structural demand floor for ASML that didn't exist in the pre-CHIPS Act world.
The AI Catalyst and What Comes Next
The emergence of large language models and generative AI since 2022 has supercharged demand for the most advanced chips — and, by extension, for the machines that make them. AI training requires massive quantities of high-performance logic chips (GPUs, TPUs, custom accelerators) and high-bandwidth memory chips, both of which rely on leading-edge lithography. ASML's Q4 2023 order bookings surged to a record €9.19 billion, more than tripling from the prior quarter, driven overwhelmingly by demand for EUV systems from customers building out AI chip capacity.
The AI demand cycle is layered on top of the structural onshoring trend and the ongoing secular growth of semiconductor content per device. Whether it is 5G networks, electric vehicles, data centers, or edge computing, the trajectory of silicon demand is upward. ASML's position as the sole supplier of the most critical equipment in the value chain means it captures a disproportionate share of every dollar invested in advanced chip manufacturing capacity.
But the dependence on a small number of customers — TSMC alone is estimated to account for a significant plurality of ASML's EUV revenue — creates concentration risk. If TSMC delays or cancels a node transition, ASML's order book feels it immediately. If Intel's foundry ambitions falter, one of three expected High-NA customers weakens. The AI boom is real, but the semiconductor industry has always been cyclical, and ASML's revenue has historically tracked those cycles even as the long-term trend is upward.
The question that hangs over Veldhoven is whether the current moment — AI-driven demand, onshoring investment, geopolitical urgency — represents a permanent step-change in ASML's trajectory, or a cyclical peak that will be followed by the same digestion period that has characterized every previous semiconductor supercycle. ASML's own guidance has been carefully calibrated: a "transition year" in 2024, followed by what the company has described as "very significant" growth in 2025 and beyond.
A Fingerprint on a Mirror
There is a detail from the BBC's visit to ASML's Veldhoven cleanroom that crystallizes the entire enterprise. Bram Matthijssen, a technician assembling one of the company's latest EUV systems, described his working conditions: "There are moments where we have to wear gloves over gloves to make sure we don't leave any fingerprints, to make sure we don't bring any extra dust into the machine. A single fingerprint... can cause significant damage to the machine."
A fingerprint. The oils from a human hand, deposited on a mirror polished to atomic smoothness, in a vacuum chamber where light at 13.5 nanometers bounces between surfaces that must maintain their shape to within fractions of a nanometer — a fingerprint can degrade the performance of a $350 million machine producing the chips that power the global economy. That is the scale at which ASML operates. That is the margin for error in which it has built a monopoly. Every chip in every AI data center, every smartphone, every autonomous vehicle — all of them trace their existence back to a cleanroom in Veldhoven where a technician puts on gloves over gloves, and the machine begins its impossible work.
