The Sun photographed in extreme ultraviolet light, its corona glowing against black

The Machine Only One Company Can Build

In 1984 a Dutch joint venture started work in a Philips building in Eindhoven. By 2012 its great bet was producing seven wafers an hour and every sensible rival had left. For most of that time, patience and stubbornness looked exactly alike.

№ 013September 20269 min readPedrum Shokouhi

The Sun in extreme ultraviolet at 13.1 nanometers, seen from orbit. Light this short does not get through the air, so a chip machine has to make its own in a vacuum.

Gjalt Smit · Arthur del PradoStrijp TQ, Eindhoven, 1984

ASML opened for business on April Fools’ Day of 1984, as a joint venture called ASM Lithography, in the Strijp TQ building at Philips in Eindhoven, under a new general manager named Gjalt Smit, with forty-seven engineers who had been talked out of jobs for life. There had been fifty in the group. Philips offered them attractive terms, and they understood the offer well enough to call it asset stripping. Philips had not even wanted this partner: it had preferred the American giant Perkin-Elmer, regarded Arthur del Prado’s ASM as an upstart, and went back to him only after Perkin-Elmer signed with someone else and the group ran out of time. The machine the engineers carried across, the PAS 2000, needed its key technology redesigned, starting with a hydraulic wafer stage that brought oil into the cleanroom. The company soon outgrew its rooms and spilled into what ASML’s own timeline calls a leaky shed beside the Philips office, before moving to a new building in Veldhoven in 1985. That shed has since become scripture, and ASML’s own account of its founding uses the word mythos.

The first years gave nobody a reason to think about the far future. The lens partnership with Carl Zeiss began in 1986, but by 1988 the company was struggling badly enough that ASM pulled out, and a Philips board member named Henk Bodt persuaded his colleagues to lend a final helping hand. In 1986, at a scientific meeting in Japan that had nothing to do with anyone in Eindhoven, an NTT researcher named Hiroo Kinoshita reported that he had projected images using very short light bounced off multilayer mirrors. The logic was simple. Shorter light prints finer features, and the industry was then working with mercury lamps at 436 nanometers while Kinoshita was reaching down near 11. The catch was that nearly everything absorbs light that short, so a machine built on it would have to run in a vacuum on mirrors alone. “Unfortunately, the audience was highly skeptical of my talk,” Kinoshita said decades later. In America, reviewers of a Bell Labs proposal wrote that features that fine “would never be needed,” and Andrew Hawryluk of Lawrence Livermore presented a paper on the idea and went home, in his words, with “my tail between my legs,” vowing never to mention it again.

Andy Grove · Jos BenschopEUV LLC, 1997 to 2003

The first party to show real patience was American. In 1992 Andy Grove approved $200 million of Intel’s money for EUV. When federal support for the national labs ran down, Intel and its partners committed $250 million to them in 1997, with Intel agreeing to pay all of the labs’ EUV research costs through a consortium called EUV LLC. ASML hired Jos Benschop to start its program that same year, formed a European research group with Zeiss in 1998, and joined the American consortium in 1999. Canon and Nikon, who had led the world in lithography sales in the mid-1990s, stayed outside. The histories differ on how: one says Washington effectively blocked Japanese firms from taking part, another that Nikon declined amid the controversy and Canon was kept out. Either way, Intel pushed for the Dutch. ASML promised a factory in the United States as large as its Dutch one and 55 percent American components, and at least one careful history says it never met those terms. So a good part of what people now admire as ASML’s moat was handed to it by a foreign government and by a customer who badly needed a supplier. In 2001 it closed the purchase of Silicon Valley Group, which one history calls the last American lithography supplier, with Intel lobbying for the deal, and so, seventeen years after opening, ASML bought what had been Perkin-Elmer’s lithography business.

No lamp makes this light, so the machine has to manufacture its own. A generator fires droplets of molten tin, each about a third the width of a human hair, through a vacuum at seventy meters a second. A laser pulse strikes a droplet in flight and flattens it into a pancake, a stronger pulse vaporizes the pancake into a plasma of about 500,000 kelvin by ASML’s figure, and the plasma gives off light at 13.5 nanometers. Then the next droplet arrives, 50,000 times every second. Each mirror on the way to the wafer reflects only about 70 percent, so after a dozen of them less than 2 percent of the light may survive. Tin ions clogged the optics until engineers added hydrogen to sweep them away. Zeiss makes those mirrors, and its own illustration is that one enlarged to the size of Germany would stray from its intended shape by just 0.1 millimeters. Optics had been the most feared problem of all. Once Zeiss showed it could make them, ASML’s own account says, they quickly dropped off the list of critical issues, and the company’s attention shifted to the light source.

Technicians working on the large primary mirror of the Hubble Space Telescope
The Hubble primary mirror being ground at Perkin-Elmer in 1979. A few years later Philips would prefer the same company to ASM.
Grant WillsonSPIE keynote, February 2005

Benschop had once said EUV would be in volume production by 2006. In 2006 it produced two prototypes, one for imec in Belgium and one for a nanoscale college in Albany, and a year earlier the keynote stage at the field’s main conference had turned on the idea altogether. “No one will ever print a wafer for profit with EUV,” said Grant Willson, a University of Texas professor and a co-founder of Molecular Imprints, which made a rival technology. DARPA ended its lithography program that same year, a billion dollars and fourteen years in. Backers in 2001 had talked about $15 million to $20 million a tool. The first pre-production system shipped to a Samsung research site in South Korea in 2010 and reached first light on Christmas Eve, and by the following year ASML was reporting an average selling price for that model of about €40 million. “They practically crucified us at SPIE,” Benschop remembered, because customers kept holding up the old source roadmap against what the source actually did. When ASML agreed to buy its light-source supplier, Cymer, in October 2012, the release gave the state of play in its own numbers: up to 11 watts at customer sites, good for about seven wafers an hour.

Eleven watts

EUV source power in watts, against the 250 that volume production needed.

250 W, the target October 2012, customer sites: up to 11 W 11 W Oct 2012 customer sites October 2012, laboratory: 30 W 30 W Oct 2012 lab 2016, laboratory: 200 W 200 W 2016 lab July 2017, demonstrated: 250 W 250 W July 2017 demonstrated 250 W, the target 11 W Oct 2012 customers 30 W Oct 2012 lab 200 W 2016 lab 250 W July 2017 shown
Eleven watts meant about seven wafers an hour. Two hundred and fifty meant 125. ASML published the first figure itself, in the release announcing it would buy its light-source supplier.
Canon · NikonThe exits, 2009 and 2011

It is tempting to call the people who quit fools, and I don’t think they were. Canon had put its money on X-ray lithography and said in late 2000 that it would stop. Nikon had bet on electron projection and dropped it in 2005, after Intel bought $90 million of its bonds in part to keep it committed to EUV. Nikon still delivered an EUV tool to the Japanese venture Selete in 2007. In May 2009 a Nomura analyst, hearing that Nikon had paused some of its EUV tools, wrote that “ASML’s virtual monopoly may well become a real one,” and that September EE Times reported from industry sources that Canon’s EUV efforts were fading along with its high-end lithography, which Canon partly disputed. Nikon left EUV systems in 2011, citing in a later investor deck “the enormous investments required.” Its 2013 annual report still said EUV “has not proceeded as planned” and pointed to 450-millimeter wafers instead. Put that sentence beside the eleven watts and look for the fool in it. A sensible board quits there. I think I would have quit there, and I would have written a very good memo explaining why.

Intel · TSMC · SamsungCustomer Co-Investment Program, 2012

What kept the bet alive was the customers, and the structure of that rescue is the part an investor should study. On July 9, 2012, with EUV tools under evaluation and ASML short of cash, the company announced a program that ended with Intel, TSMC and Samsung taking 15, 5 and 3 percent of the shares and agreeing to fund €1.38 billion of research over five years. They paid €3.85 billion for stock that was largely non-voting, and ASML handed all of that money back to its other shareholders. Cymer followed for €1.95 billion at announcement, because ASML told the SEC it did not believe Cymer had the resources to finish the source alone, and in 2016 came 24.9 percent of Carl Zeiss SMT for €1 billion. Of Intel’s €829 million research commitment, €553 million was for 450-millimeter wafers, the direction Nikon had pointed toward. In November 2013, sixteen months after the program was announced and following its customers’ decision, ASML paused that work, and the money went to EUV.

The watts arrived late. By 2016 ASML had 200 in the lab, and in July 2017 it could say it had demonstrated 250 rather consistently, the level that meant 125 wafers an hour, while one of its executives joked that when he worked at Cymer the goal was always next year. In 2019 what ASML counts as the first commercial EUV product reached the public, Samsung’s Galaxy Note10, and at a workshop that year Benschop took out his phone. “I once said naively that EUV would be in volume production in 2006,” he recalled telling the room, thirteen years late by his own count, and then he waved it at them. Asked later to sum the whole thing up, he laughed. “You could say that we were either brilliant or just stubborn.” Chris Miller, who wrote the book on all this, names the quality “a stubborn willingness to invest,” and I think he is right. ASML’s annual report for 2025 calls it the world’s only manufacturer of EUV systems, and it recognized revenue on 48 of them that year, though Reuters had reported in December 2025 that a prototype in a Shenzhen laboratory was generating EUV light and “has not yet produced working chips.”

I think Benschop’s line is the most honest thing anyone has said about the company, because he refuses to pick. IBM and NTT evaluated X-ray lithography seriously too, before abandoning it over masks that could not be made to work, and that patience is remembered as waste. I have written before about the planes that came home, and this is one of them: a survivor whose bullet holes we now read as character. The best I can offer is two tests, held loosely. ASML’s stubbornness was narrow. It kept improving the deep ultraviolet machines that paid the bills, shipping its first immersion system in 2003, and it dropped what customers stopped wanting. And the people who would have to buy the thing put money down while the number read eleven. Even that does not come out clean, because the customer who believed longest collected the least. Intel made its first bet in 1992 and, after stumbling in its own factories, did not start using EUV until 2023.

Which leaves the man at the keynote. Canon bought Molecular Imprints, the company Grant Willson had co-founded, in 2014, and nanoimprint became its route back toward the leading edge of chipmaking, while Nikon’s 2022 management plan listed EUV-related components as a growth driver, parts for the machine it had walked away from. In September 2024 Canon made what was reported as its first official sale of a nanoimprint tool, ten years after the purchase, to a research institute in Texas. Ten years into its own bet, ASML had two prototypes in the field and had been told from a stage that no one would ever print a wafer for profit. I would love to tell you that one of these is patience and the other is stubbornness. That is the kind of sentence you only get to write afterward.

Notes

  1. ASML’s founding, the forty-seven engineers, Perkin-Elmer, the PAS 2000 and the sheds are from Jorijn van Duijn, “ASML’s founding story,” ASML, April 3, 2024, and ASML’s online history timeline, which also gives the leaky shed in Eindhoven, the 1985 move to Veldhoven, the start of the Zeiss partnership and the events of 1988.
  2. Kinoshita’s 1986 talk and Hawryluk’s reception are told in Sander Hofman, “Making EUV: from lab to fab,” ASML, March 30, 2022, and in SPIE’s Photonics Focus, May/June 2023. Sources differ on the society that hosted the talk. The 11-nanometer figure is from Rachel Courtland, IEEE Spectrum, October 31, 2016.
  3. The Bell Labs review, Grove’s $200 million, the 1997 commitment, the Japanese firms being blocked, ASML’s 1999 terms, the SVG lobbying, the description of SVG as the last American supplier and the IBM and NTT X-ray work are from John VerWey, Tracing the Emergence of Extreme Ultraviolet Lithography, CSET, July 2024. That Nikon declined, that Canon was prevented and that ASML never met its terms are from Brian Potter, “How ASML Got EUV,” Construction Physics, November 20, 2025, which also names Ultratech Stepper as another American licensee. The sale of Perkin-Elmer’s lithography division to SVG in 1990 is from ASML’s Wilton history, 2023.
  4. Droplet speed, rate and plasma temperature are from ASML’s “The hottest thing about ASML machines.” ASML elsewhere describes the plasma as 40 times hotter than the Sun’s surface, which does not square with that figure, so I have used one. Droplet size and the hydrogen are from MIT Technology Review, October 27, 2021; mirror reflectivity from IEEE Spectrum, 2016; the Germany comparison from Simon Bihr, Carl Zeiss SMT, 2022 EUVL workshop presentation.
  5. Grant Willson’s remark is from “EUV litho headed for failure, say SPIE keynoters,” EE Times, March 1, 2005. DARPA’s exit and the 2001 price talk are from the CSET report; first light from “Making EUV.” ASML gave the NXE:3100’s average selling price as €42 million in its second-quarter 2011 results and €40 million per system in its third-quarter results, both filed with the SEC. CSET’s higher figure of $120 million could not be traced to its cited source.
  6. The 11 watts, seven wafers an hour and the 2012 deal terms for Cymer are from ASML’s release of October 17, 2012, filed with the SEC. The 30-watt laboratory figure is from the same release, the 200-watt figure from IEEE Spectrum, 2016, and the 250-watt report from EE Times Asia, July 18, 2017.
  7. Canon and Nikon: CSET; John Walko, “Nikon said to be delaying EUV lithography development,” EE Times/EDN, May 15, 2009; Mark LaPedus, “Canon struggles in litho as tool prices soar,” EE Times, September 18, 2009; Nikon, Annual Report 2013; Nikon, IR Day 2022 presentation, page 47. Canon denied it had left 193-nanometer lithography.
  8. The co-investment program is from ASML’s releases of August 5 and October 17, 2012, a later 6-K, and the 20-F filings for 2012, 2013 and 2016, which also give the 450-millimeter pause, Intel’s commitment and ASML’s view of Cymer’s resources. The Zeiss stake is from the joint release of November 3, 2016.
  9. Benschop’s remarks and the account of the optics are from “Making EUV.” The 2025 figures are from ASML’s Form 20-F filed February 25, 2026. The Shenzhen prototype is from Reuters, “How China built its ‘Manhattan Project’ to rival the West in AI chips,” December 17, 2025, read as republished by The Standard of Hong Kong. Miller is quoted in MIT Technology Review, April 1, 2024. Intel’s 2023 start is from CSET.
  10. Canon’s 2014 acquisition and its first official sale are from “Canon finds first customer for nanoimprint tool,” Bits&Chips, September 30, 2024.

P.S.I’m Pedrum Shokouhi. I’m obsessed with how companies get built, funded, and run. So I write about them, take apart their strategy like an investor, and build my own. This is where I share what I’m learning.