ASML, the Dutch company ChatGPT, Claude and the rest depend on

Today we talk about the most strategic company in the entire global tech sector, and almost nobody in Spain ever says its name. It sits in a Dutch town of a hundred and sixty thousand people called Veldhoven, on the outskirts of Eindhoven. It makes a single thing: extreme ultraviolet lithography machines, the devices without which you cannot produce a single one of the chips that hold up ChatGPT, Claude or Gemini. One machine costs between 200 and 380 million dollars and ships in four chartered Boeing 747s. The company is called ASML. And the Dutch government decides who it can sell to, not the company. My take is biased because I watched the ASML documentary and stayed fascinated for months. Form your own: if you don't hear the name within six months, it's because your usual press isn't covering the sector.

I've spent two years collecting clippings about ASML because it strikes me, without exaggeration, as the most fascinating piece in the whole AI chain. It's a company whose monopoly owes nothing to mergers, nothing to regulatory barriers, nothing to dirty play. It owes everything to forty years of applied physics and billions in R&D that no competitor has been able to replicate. And, as we'll see, it's also the cleanest example of how a strategic company can lose control over its own sales when exports become a diplomatic tool.

What ASML does, and why only ASML

ASML —short for Advanced Semiconductor Materials Lithography— was founded in 1984 as a small spinoff of Philips. For nearly two decades it was a bit player in semiconductor lithography, a field then dominated by Nikon and Canon. The bet that changed history was made in the nineties, when ASML decided to invest aggressively in a technology called Extreme Ultraviolet Lithography, EUV, which promised to allow circuits to be etched at scales below 7 nanometers, where conventional lithography no longer reached.

The bet was risky. Extreme ultraviolet light —wavelength of 13.5 nanometers— is absorbed by air and by practically every known material. Generating it means firing droplets of molten tin at 50,000 km/h with a high-power laser and collecting the resulting plasma with mirrors machined to atomic tolerances. Designing machines that do this stably, repeatably and economically is an engineering problem on the edge of the impossible. ASML took almost two decades to solve it, working with a consortium that included funding from Intel, Samsung and TSMC. The first commercial machines shipped in 2018.

As of today, ASML is the only company in the world that makes EUV machines running in commercial production. Nikon and Canon, its old rivals, dropped out of the race at different moments because of cost and complexity. China has spent years trying to develop a homegrown alternative through SMEE (Shanghai Micro Electronics Equipment) and has not yet managed to produce a comparable EUV machine. Industry estimates put SMEE somewhere between five and ten years behind the state of the art.

The price reflects this concentration. A standard EUV machine —generation NXE:3400 or NXE:3600— costs between 180 and 220 million dollars depending on configuration. The next generation, the High-NA EUV (NXE:5000 and EXE:5000), runs in the range of 350 to 380 million dollars. To put it in perspective, the price of a High-NA machine equals the full annual revenue of a mid-sized Spanish listed company. And ASML produces only a few dozen EUV machines a year, not hundreds: the complexity and the lead times on component manufacturing —especially the optical mirrors, made by Zeiss in Germany— cap output.

Who buys and how many there are

EUV machines go to the only three customers capable of paying for them and using them: TSMC, Samsung and Intel. These three manufacturers run the advanced nodes —5 nanometers, 3 nanometers and, since 2025, 2 nanometers— that require EUV without exception. Companies like GlobalFoundries, UMC and SMIC run earlier nodes that can still be made with conventional DUV lithography, not EUV.

ASML publishes the exact figures for installed machines in its annual reports. By the end of 2024, ASML had delivered around 200 NXE-generation EUV systems, plus a few dozen earlier generations already retired. The first High-NA machines began shipping in late 2024, with Intel as launch customer and TSMC and Samsung in line. The High-NA production pace is slow: around half a dozen units a year, according to the guidance the company shares with analysts.

This slowness is not strategic. It's structural. Each machine needs thousands of components produced by a highly specialized chain of European and U.S. suppliers. Any attempt to speed up production runs into a bottleneck at some second-tier supplier —Zeiss for optics, Trumpf for lasers, Carl Zeiss SMT for alignment, and other less familiar names.

Geopolitical control from Washington

Here comes the part that turns ASML into a geopolitical piece of the first order.

The United States has been pressing the Dutch government since 2018 to restrict exports of EUV machines to China. In 2019, the Netherlands decided not to grant export licenses for EUV machines to Chinese customers. That restriction has held ever since. China, as a result, has not been able to legally buy a single new EUV machine in more than six years.

The measure was widened in September 2023. Under renewed pressure from the Biden administration and, later, from the Trump administration in its second term, the Netherlands also agreed to restrict exports of the most advanced deep ultraviolet (DUV) lithography machines, specifically the TWINSCAN NXT:2050i and NXT:2100i models. Those machines, though less sophisticated than EUV, are needed to manufacture at 7 nanometers and below. The widened restriction means China is effectively blocked from producing 7-nanometer-and-below chips competitively, unless it manages to develop a domestic alternative.

China has tried to respond on several fronts. It has stepped up purchases of less-restricted DUV. It has developed multiple-patterning lithography techniques that allow manufacturing at 7 nanometers with DUV, albeit at lower cost-efficiency and yield. SMIC, the largest Chinese foundry, has managed to produce 7-nanometer chips for Huawei using that technique, proving that the ban slows but does not fully stop. The open question is whether China can reach 5 or 3 nanometers without EUV. The reasonable answer, according to analysts who know the physical details, is yes but at a cost several orders of magnitude higher, which puts it outside commercial competition.

The consequence for ASML is direct. China represented around 26% of ASML's revenue in 2023 before the harshest restrictions. That figure has fallen since, though with some inertia from contracts signed before the ban. ASML has publicly questioned the strategic usefulness of the restriction through statements from Peter Wennink, CEO until April 2024, and Christophe Fouquet, the current CEO. The company argues that the restriction slows but doesn't stop China, and that the cost to ASML outweighs the strategic benefit to the West. That position has not changed the decisions of the Dutch or the U.S. government.

The paradox of the customer with no control

Here's the part worth naming plainly, because it has implications beyond ASML.

ASML is a private company listed on Euronext Amsterdam and on NASDAQ. Its market capitalization has topped 300 billion dollars at its peaks. It's the most valuable European company by capitalization after Novo Nordisk. And yet it does not control who it can sell its most advanced products to. That decision is made by the Dutch government, partly under pressure from the U.S. government, with ASML having no formal voice in the licensing process.

This produces an odd situation in contemporary capitalism. The most strategic company on the European continent, with an effective monopoly in its technology, has lost commercial autonomy over a quarter of its potential market. The Netherlands, a country of 17 million people with a mid-sized economy, decides the exports of the most critical company in the global sector. That decision, in turn, is conditioned by the geopolitical weight of the United States. The chain of causation runs: the White House pressures The Hague, The Hague denies the license, ASML loses the customer.

There's nothing illegal in this scheme. There's a perfectly recognized right of export in every Western legal system. But the balance between national sovereignty, foreign policy and corporate freedom is, in ASML's case, particularly skewed. The company says so openly —Fouquet has repeated in several interviews in Bloomberg and the Financial Times that the restrictions fail to meet their goal and damage the European sector. The restrictions, however, do not ease.

What this means for the AI sector

To understand why ASML matters to a ChatGPT user in Madrid, you have to think back up the chain.

ChatGPT runs on servers that use Nvidia H100 or H200 GPUs. Those GPUs are made by TSMC in Taiwan, in its Hsinchu fabs. TSMC's fabs use ASML's EUV machines to etch the circuits. ASML's EUV machines use Zeiss mirrors from Germany and Trumpf lasers, also from Germany. The whole chain has four or five critical links, each one a different bottleneck.

If ASML stops delivering machines, in five years there are no new Nvidia GPUs able to compete with the current ones. If TSMC halts, the chain breaks within a year. If Zeiss has an industrial accident, ASML can't produce machines for months. Each of these risks is low on its own, but stacked up they're very real.

Unlike the Taiwan risk —which is geopolitical and exogenous—, the ASML risk is internal to the European space and, in theory, controllable by European decisions. The European Union passed the European Chips Act in 2023 for 43 billion euros precisely to reduce the dependence. A good chunk of that money is going into manufacturing —the Dresden fab, the Magdeburg fab (Intel, under review)—, but very little into protecting and diversifying ASML's supplier chain. That's a strategic decision the European debate has barely made visible.

The political question

This is personal opinion, but I hold it on the record of the file itself. ASML is the clearest example of a European company with structural power that the European Union has failed to turn into geopolitical leverage.

If the United States had an American ASML, it would use it to negotiate concessions on a thousand fronts with its allies and its rivals. It does so systematically with Boeing, with Lockheed, with Microsoft. If China had a Chinese ASML, it would use it as a diplomatic and commercial tool. It does so with TikTok, with Huawei, with CATL. Europe has a European ASML and limits itself to complying with the restrictions dictated by the bigger ally. Not out of bad faith, but out of institutional inability to articulate an autonomous industrial policy at the continental scale.

The discussion about what Europe can or cannot do with ASML is a discussion about what Europe is in the contemporary international system. If Europe wants to be a strategic actor, ASML is the operational test: use its power, negotiate the restrictions from a position of its own, demand reciprocity for the weight it contributes. If Europe accepts staying a customer of an order set elsewhere, ASML is the example of the ceiling on that acceptance.

The concrete figure that closes it: according to ASML's annual report for fiscal year 2024 —published in March 2025—, the share of Chinese orders in the total fell from 26% in 2023 to 41% in 2024 (it rose temporarily because of contracts predating the September 2023 restriction that were executed during the year), and is projected to drop to the 20% range in 2025 as the legally signed contracts are fulfilled and the tap shuts off. That roller coaster, entirely outside ASML's control, is the numerical embodiment of the situation I've been describing.

Definitions

Lithography: the process of transferring a pattern onto a surface. In semiconductors, it's used to etch circuits into silicon wafers.

EUV (Extreme Ultraviolet): a lithography technology that uses extreme ultraviolet light at a 13.5-nanometer wavelength. It allows circuits to be etched at scales below 7 nanometers. It's exclusively ASML's as a commercial product.

DUV (Deep Ultraviolet): the technology that preceded EUV, with wavelengths of 193 to 248 nanometers. Still valid for nodes above 7 nanometers and usable, with multiple-patterning techniques, for smaller nodes at a cost and yield penalty.

High-NA EUV: a new generation of ASML EUV machines with widened numerical aperture (NA = 0.55), allowing manufacturing at 2 nanometers and below. The first commercial deliveries began in late 2024.

Export control: a legal mechanism by which a government restricts the international sale of certain products on national-security grounds. It's the tool applied to ASML with respect to China.

References

Chris Miller, Chip War: The Fight for the World's Most Critical Technology (Scribner, 2022). Central chapters on the historical development of EUV lithography and ASML's position.

ASML, Annual Report 2024 (ASML Holding NV, March 2025). Official figures on deliveries, market by geography and projections.

Marc Hijink, Focus: The ASML Way (Ambo|Anthos, 2023; English edition Maven Publishing, 2025). An institutional history of the company written by the NRC Handelsblad correspondent.

Financial Times, sector coverage of ASML and export restrictions (2022-2025). Recurring analysis of the positions of Wennink and Fouquet.

Nederlands Ministerie van Buitenlandse Zaken, official documentation on export licenses for lithography equipment (2019-2024). The regulatory framework of the restriction.

U.S. Bureau of Industry and Security, Export Administration Regulations on semiconductors (Federal Register, 2022-2024). The U.S. regulatory framework that conditions the pressure on the Netherlands.

Centre for Strategic and International Studies, Mapping the Semiconductor Supply Chain (csis.org, periodically updated). A visual map of ASML's position in the global chain.

Going deeper

Vili Lehdonvirta, Cloud Empires (MIT Press, 2022). A framework for understanding how the concentration of digital infrastructure runs through European companies with little institutional weight.

Graham Allison, Destined for War (Houghton Mifflin Harcourt, 2017). A broad geopolitical framework for understanding the U.S.-China dynamic that conditions ASML.

Dan Wang, Breakneck: China's Quest to Engineer the Future (W. W. Norton, 2025). A recent analysis of China's industrial bets in semiconductors and the efforts to reproduce ASML's technology.

European Commission, European Chips Act – Implementation Reports (2023-2025). Documentation of the European program and its limited attention to ASML's supplier chain.

You may also like

• Why OpenAI wouldn't exist without a factory in Taiwan
• Why the United States banned selling chips to China
• Nvidia is worth more than Spain's economy
• DeepSeek and the end of the American monopoly

Elsewhere

• Revamping Europe's chips strategy: indispensability, not self-sufficiency (Bruegel)
• Semiconductors — Center for Security and Emerging Technology (Georgetown)