"There Is No Plan B"

In the last week of December 2023, the receiving dock of Intel’s D1X fab in Hillsboro, Oregon, began accepting the contents of three Antonov An-124 cargo flights, twenty trucks, and roughly 250 numbered crates. The largest crates came in by Antonov because nothing smaller could carry them. The shipment, taken together, weighed about 165 metric tons, which is the weight of two Airbus A320s. Inside, broken down into modules so it could be reassembled in a clean room in Oregon by 250 ASML engineers over the course of six months, was a single machine. Its name was the TWINSCAN EXE:5000. Its sticker price, depending on the source and the configuration, was between $370 million and $400 million. It was the first delivery of a new generation of lithography tool that ASML and its German optics partner Carl Zeiss had been developing for almost a decade. There was no second supplier. There was no domestic alternative. If the EXE:5000 did not arrive in Hillsboro, Intel did not get to print transistors at the Angstrom-scale node it had publicly committed to building. There was no Plan B.

That phrase, in different forms, had become the central piece of language in Veldhoven by the early 2020s. Peter Wennink, the soft-spoken Dutch accountant who had run ASML since 2013 alongside his engineering counterpart Martin van den Brink, used it with a kind of weary precision in interviews. He told a Bloomberg crew at the company’s Veldhoven headquarters in January 2024 that “without ASML, without our technology, that’s not going to happen,” referring to the AI buildout that was at that moment driving every order book in the industry. His successor, Christophe Fouquet, who took over in April 2024, was more blunt in a 2026 sit-down with TechCrunch: no competitor was coming. He could afford to say it because the math, in the only direction that mattered, was settled. By 2024 ASML had a 100 percent share of the global market for extreme ultraviolet lithography systems. Every chip in every modern smartphone, every AI accelerator in every hyperscale data center, every advanced node CPU and GPU at TSMC, Samsung, Intel, and SK Hynix had to pass through a machine made in a corner of the southern Netherlands or it could not exist.

The route by which ASML reached that position was, by 2024, an industry origin story. The Dutch firm had inherited Philips’s stepper bet in 1984, climbed past Nikon and Canon in the 2000s on the back of immersion lithography and a willingness to let TSMC and Intel co-invest in its R&D, and absorbed the surviving fragments of American optical lithography when it bought Silicon Valley Group in 2001. EUV itself had been the project of an industry consortium and a generation of physicists who refused to give up. The harder question, by the mid-2020s, was why, after thirty years of effort, ASML was the only company in the world that could build the resulting machine, and why that fact, more than any other technical detail of the chip industry, was the reason American export controls on China actually bit.

The answer began with what ASML did not build itself. Of the roughly 100,000 individual parts in a current-generation EUV scanner, the company manufactured perhaps 15 percent in-house, according to its own supply-chain disclosures and analyses by independent industry observers. The remaining 85 percent came from a network of about 5,100 suppliers, of whom ASML itself counted around 700 as product-related and roughly 200 as critical, single-source partners. The geography of that network was the first clue to why no government in the world could replicate it on demand. About 40 percent of the supplier spend, by ASML’s own annual reporting, sat in the Netherlands. Another 40 percent sat elsewhere in Europe, primarily Germany. Roughly 13 percent sat in the United States. The remaining sliver was in Japan and a handful of other Asian sources. Three of the most important nodes in that network mattered above all others: Carl Zeiss SMT in the small Swabian town of Oberkochen, Trumpf in Ditzingen near Stuttgart, and Cymer in San Diego, California. None could be substituted. Each was, in its own way, irreplaceable.

Zeiss sat at the optical heart of the problem. The Mo/Si reflective stacks the EUV machine required, polished to a smoothness measured in picometers, were the kind of artifact only one supplier on the planet had ever learned to build, and Zeiss SMT, the semiconductor optics arm of the 178-year-old Stiftung in Oberkochen founded by Carl Zeiss and Ernst Abbe in 1846, was that supplier. To deliver each mirror, Zeiss combined ion-beam figuring, in which charged particles slowly etched high points away one nanometer at a time, with interferometric metrology rigs the company had invented in parallel to verify the result. The polish on a single high-NA mirror could take months of cleanroom work.

That capability did not exist anywhere else. Nikon and Canon had fine optics groups, but their lens-grinding traditions had been built around transmissive glass for cameras, microscopes, and DUV steppers; neither had a comparable reflective EUV stack. American optics firms had thinned out catastrophically in the 1980s and 1990s as the lithography customer base disappeared overseas. By the 2010s Zeiss SMT effectively was the world’s EUV optics industry. ASML knew this and spent a decade tightening the relationship into something close to a corporate merger. In November 2016, the two companies announced that ASML would take a 24.9 percent stake in Carl Zeiss SMT GmbH for one billion euros in cash, with another 760 million euros committed over six years to underwrite Zeiss’s R&D and capital expenditures for the next-generation high-NA optics. The deal closed in the summer of 2017, with the bulk of the investment flowing to Oberkochen. From that point on, the people who built the most precise mirrors on Earth were partly owned by the company that put them inside lithography systems.

Trumpf was the second pillar. The Ditzingen-based laser maker, family-owned by the Leibinger clan since 1923, had spent decades building industrial CO2 lasers for cutting and welding metal. In the 2000s it had been pulled into an entirely different problem. Generating EUV light at industrially useful power required vaporizing tin into a plasma forty times hotter than the surface of the sun, and the only practical way to do that was to fire a sequence of laser pulses at a stream of microscopic molten droplets falling through a vacuum chamber. The pulse rate had to be 50,000 hertz. Each droplet, about thirty micrometers across, had to be hit twice by an infrared CO2 laser pulse: once with a low-power “pre-pulse” that flattened the droplet into a wider pancake, and once with a high-power main pulse that turned the pancake into plasma. The main pulse needed average power in the tens of kilowatts and peak power above a megawatt. Trumpf’s engineers built it by chaining five amplifier stages onto a small seed laser, multiplying the input ten thousand times. Peter Leibinger, the firm’s vice chairman, told interviewers that ASML and Trumpf functioned like a single corporation rather than two. “Only Trumpf can build the lasers needed for EUV lithography,” the company stated bluntly in its own technical literature. As of the mid-2020s that remained true.

Cymer was the third pillar, and the one ASML had simply absorbed. Founded in 1986 by two University of California, San Diego physicists, Robert Akins and Richard Sandstrom, Cymer had spent two decades building the excimer laser sources used in deep-ultraviolet steppers. In the 2000s, in cooperation with Trumpf and ASML, it had pivoted to the harder problem of laser-produced plasma EUV sources, the integrated system that fed the tin droplets, fired Trumpf’s CO2 pulses, captured the resulting 13.5 nanometer photons in a graded multilayer collector mirror, and delivered them to the scanner optics. By 2012, Cymer’s prototype source was running at about 50 watts under research conditions and ASML had concluded that it could not afford a single point of failure outside its own balance sheet. In October that year ASML announced an agreed acquisition for 1.95 billion euros, a roughly 60 percent premium to Cymer’s market price. The merger closed in May 2013. Cymer became a wholly owned subsidiary of ASML, its San Diego campus continuing to operate as the world’s only EUV light-source factory and one of the only places where the integrated tin-droplet, dual-pulse, plasma-generation system existed at all.

These were the three load-bearing pillars. Around them hung dozens of smaller but equally specialized suppliers: VDL ETG in Eindhoven, which built the wafer stages capable of accelerating at three g while holding nanometer position; Schott in Mainz, which supplied the ultra-low-expansion glass blanks Zeiss polished into mirrors; Berliner Glas, which ASML eventually bought outright in 2020; Veeco for the ion-beam deposition equipment that laid down the multilayer coatings; Pfeiffer Vacuum and Edwards for the pumps that held the chamber at 10^-7 millibar. Each supplier had spent a decade or more learning to deliver one specific subsystem to one specific customer. Each, in turn, had its own supply chain of specialty alloys, rare-earth magnets, semiconductor-grade neon, and tin pellets. ASML’s own internal estimate, repeated by van den Brink in a 2024 retrospective, was that the full ecosystem employed something on the order of 100,000 people across multiple continents. ASML directly employed about 44,000 of them. The rest worked for the other companies in the network, doing things ASML did not know how to do.

The reason no one else had this network was, in part, that no one else had been allowed to. When the United States Department of Energy admitted ASML to the EUV-LLC consortium in 1999 and excluded Nikon and Canon, the two Japanese firms lost legal access to the foundational IP on which the rest of the EUV stack would be built.

Nikon kept trying anyway. Through the 2000s and into the early 2010s, the Tokyo firm maintained an EUV development program out of its Kumagaya operations and showed prototype tools at the Society of Photo-Optical Instrumentation Engineers conferences. By the middle of the decade the math had stopped working. The cumulative cost of carrying an EUV program against a single end customer that no longer wanted the product, against a partner ecosystem that had concentrated entirely around ASML, against a patent thicket Nikon could not freely operate in, became unsustainable. Around 2014 Nikon’s annual reports began to drop EUV from their forward roadmaps. In quiet acknowledgments to the Japanese trade press, the company conceded that the financial and supplier disadvantages were no longer surmountable. Canon, which had bet earlier on a different post-DUV path called nanoimprint lithography, had effectively withdrawn from EUV competition years before. By 2016 the field consisted of one company. By 2024 the same was still true.

That singularity was what made the next part of the story possible. In late 2018 and early 2019, in a series of meetings at the Dutch embassy in Washington, officials from the U.S. Department of Defense and the State Department began telling their Dutch counterparts that an EUV machine ASML had agreed to ship to a Chinese customer, eventually identified by the Nikkei and Reuters as Semiconductor Manufacturing International Corporation, must not leave Veldhoven. The license had been granted by the Dutch government in 2018 under standard dual-use rules. By the new American argument, allowing it to ship would hand China a capability that no other tool could substitute. Secretary of State Mike Pompeo personally lobbied Prime Minister Mark Rutte through the spring of 2019. According to subsequent reporting by Reuters, White House officials shared a classified U.S. intelligence assessment with Rutte during a June 2019 visit to the West Wing. On June 30, 2019, ASML’s Chinese export license expired without renewal. The eight-week window in which the Dutch government would normally have considered an extension elapsed in silence. ASML never shipped an EUV system to China. As of the mid-2020s, it still hadn’t.

The decision was striking precisely because it could be enforced. Many export controls in the modern era have been described as porous, easily routed around through transshipment, gray-market resale, or licensed local copies. EUV was not. There were no parallel manufacturers in Russia, Israel, or India. There was no Korean or Taiwanese imitator. There was no Chinese domestic option, despite Beijing’s repeated declarations that one was being built. A sanctioned customer who could not buy from ASML simply could not buy. The U.S. government had finally found a chokepoint in the chip supply chain that did not require constant policing because no second source existed to be policed.

The lesson, learned in 2019, became the template. Through 2022 and into 2023, Washington pressed the Dutch and Japanese governments to broaden the export net beyond the EUV machines themselves. The next prize was ASML’s most advanced deep-ultraviolet immersion scanners, the TWINSCAN NXT:2000i and the related 2050i and 2100i series, which were not as capable as EUV but were sufficient, in the hands of a careful fab, to print logic and memory at process nodes well below 10 nanometers. On June 30, 2023, the Dutch government announced new rules requiring an export license for those models, with the regulation taking effect on September 1. ASML disclosed in early 2024 that the Dutch government had, on its own initiative, partially revoked an existing 2023 export license for shipments of NXT:2000i and 2050i tools to China, cutting off deliveries that had previously been authorized. In September 2024 the rules were broadened again, drawing in additional DUV models that had previously been outside the scope. By the time Wennink retired in April 2024, the regulatory perimeter around ASML’s product line had become the central diplomatic question of the chip industry.

Wennink himself was not a quiet partisan of the policy. In a December 2022 NRC Handelsblad interview that Marc Hijink, the paper’s longtime ASML reporter, would later draw on for his 2024 book Focus, Wennink told the reporter that ASML had already given up enough. Existing controls had cut his customers off from EUV in China, were taking U.S. companies’ competitors out of the market more cleanly than they were affecting U.S. firms, and the policy logic at this point was running ahead of the data. He repeated the point on Bloomberg in early 2023 and on Dutch radio in 2024 after his retirement, when he described the U.S.-China chip dispute as “ideological” rather than fact-based and predicted that it would persist for decades. None of those interviews, however, contradicted the underlying premise that the policy was technically enforceable. They contested whether it was wise. They did not contest whether it was possible. It was possible because of the supply chain ASML and Zeiss themselves liked to describe as “two companies, one business,” and because of the historical fact that there was, at the most advanced node, no other tool.

The sheer mechanical fact of the machine sat behind all of it. By 2024 the standard production EUV scanner, ASML’s Low-NA NXE:3800E, listed at roughly 180 million euros, took twelve to eighteen months from order to delivery, and shipped at a rate of forty to fifty units per year. The High-NA EXE:5000 that arrived at Intel in late 2023 was the next leap: about 0.55 numerical aperture instead of 0.33, capable of resolving features 1.7 times smaller, weighing 165 tons, and pricing at 350 to 400 million dollars. ASML’s order book in early 2024 ran to about ten high-NA systems already in queue, distributed across Intel, TSMC, Samsung, and SK Hynix. Each took six months of cleanroom installation by 250 dedicated engineers, who arrived on site in shifts and stayed until the machine could pass acceptance testing. Each replacement unit, if a fab destroyed one, would have to be ordered eighteen months in advance, manufactured from a supply chain whose inputs flowed through Oberkochen, Ditzingen, San Diego, Eindhoven, and a hundred smaller towns, and delivered in three Antonov flights. There were no field-swap spares. There was no warehouse stock. There was no second factory in Arizona or Hsinchu where an emergency unit could be pulled off a line.

The ASML view of this circumstance, as Marc Hijink summarized in 2024, was that the company had become two contradictory things at once. On the one hand, it was a kind of utility, a machine supplier that the rest of the chip industry expected to be available always and everywhere, neutrally and without regard to who was buying. On the other hand, in the Washington view, it had begun to look like an arms manufacturer, the maker of an instrument that decided which countries got to build the substrate of artificial intelligence and which did not. ASML’s leadership, raised in the engineering culture of Eindhoven and Oberkochen and oriented around customers in Hsinchu and Hillsboro, did not particularly want to be either. The Pentagon’s planners, by contrast, had concluded that ASML was the most consequential company in the world for their immediate strategic problem. Both interpretations were correct. The difference between them was a difference in policy temperament, not in fact.

Behind the policy was a simpler, harder reality. A nation-state with a hundred billion dollars and a decade could not build an EUV machine from scratch. It could not buy one secondhand. It could not lease one. It could not acquire the capability through a joint venture, because the capability did not exist except inside the heads and the supplier registries of a network of companies the Dutch government had jurisdiction over. The components that mattered most were embedded in family-owned firms in Swabia, in a state-funded foundation in Oberkochen, in an American campus that had been Dutch since 2013, and in the institutional knowledge of a few thousand engineers who had spent careers learning to coax 13.5 nanometer photons out of a stream of tin droplets. Money could not build that ecosystem on demand. Time could, in some theoretical future of fifteen or twenty years, but only at a rate of progress that the policy clock did not allow. The actual cost of replicating ASML’s position, by van den Brink’s own offhand estimate at industry events, was somewhere in the territory of a quarter trillion dollars and a generation of focused effort. No government in the world had committed to that program at the scale required, including the one that had been most publicly trying to.

This was what made the export controls work. Not the wording of the regulations. Not the political will of any particular administration. Not the cooperation of the Dutch or the Japanese, although that cooperation was necessary. The reason American policy could lean on a single Dutch company and through it bend the production decisions of every advanced chipmaker on the planet was that the policy was leaning, ultimately, on a fact. The fact was that the most important machine in the chip industry was made in only one place, by one company, with one supply chain, and that no parallel chain existed. The chokepoint was real. It had been built, accidentally and over forty years, by a series of decisions in Eindhoven, Oberkochen, Ditzingen, San Diego, and Washington that no single planner had ever fully designed. By the time the planners noticed it, it could not be unbuilt.

Wennink, asked variations of the question many times in his last years as chief executive, gave variations of the same answer. If something happened to the Veldhoven campus, he would say, the leading edge of the chip industry would not slow. It would stop. Other parts of ASML’s footprint could carry on with mature lithography for some time. But for the most advanced node, the node that fabricated the substrate of every modern AI buildout and every smartphone application processor, there was no second supplier and no warehouse stock and no friendly rival who could be persuaded to step in. The point was not bravado. It was a statement of inventory. The questioner who pushed for reassurance was, by 2024, expected to do the harder thing instead, which was to absorb the answer and consider what it implied about the structure of the industry that had grown up around the machine. Outside ASML’s headquarters, on most evenings, the Veldhoven cleanrooms ran their usual three shifts, assembling the next batch of EUV scanners in twelve-month cycles for the half-dozen customers who, between them, would decide what computing the world got to do for the rest of the decade.