Essay · Hans Ley & Claude Dedo · 16 November 2026

Japan's Four Paths

What becomes of Europe's trump card when others learn to play without it

In May we called ASML Europe’s last trump card. At the top it has grown stronger since. At the lower end, Japan is working on four paths at once to make it superfluous. And ASML has answered one of them in a way worth listening to closely. An addendum to “ASML — Europe’s (Last) Trump Card”.

I. At the top: stronger than in May

In May it looked as if ASML’s newest machine, the high-NA scanner at around 350 million dollars, was being spurned by its most important customer. TSMC had repeatedly said the machine was too expensive and not needed for the next manufacturing generations. In September TSMC committed: high-NA will be used in volume manufacturing from 2030. Samsung wants to use the machine for memory as early as 2028, Intel for its 14A process.

Customers are now ordering more than ASML can deliver for 2027 and 2028. One bottleneck is the capacity of Zeiss in Oberkochen, the sole supplier of the mirrors. In July ASML raised its prices. A company that can raise prices while customers queue up is not afraid of its competitors. At the top, the trump card is stronger today than it was in May.

II. At the bottom: four paths

Thirty years ago Japan dominated the market for lithography machines. Nikon and Canon gave up early on extreme ultraviolet, the technology on which ASML’s monopoly rests, as too risky. Now the country is pursuing four paths at once.

The stamp. Canon prints circuits instead of exposing them. A pattern is cut into a quartz mould and pressed into a liquid layer on the wafer; a flash of light hardens it. The technology goes back to a company from Austin that Canon bought in 2014. On 26 September 2024 Canon delivered the first machine to the Texas Institute for Electronics, a research consortium in which Intel, AMD and Samsung take part. It achieves line widths of 14 nanometres. According to Canon chief Fujio Mitarai, the price is to be “one digit less” than an EUV scanner, and power consumption a tenth. Canon wants to sell ten to twenty machines a year. The printing group DNP plans to mass-produce stamp templates for structures of the 1.4-nanometre class from 2027. So far no chip factory has committed to the stamp for volume manufacturing. Every contact wears the mould, and every speck of dust becomes a defect.

The two mirrors. Tsumoru Shintake, a physicist at the Okinawa Institute of Science and Technology who previously built particle accelerators, asked in 2024 why an EUV machine has to be so complicated. In ASML’s scanners the light is guided over around ten mirrors; each swallows part of it, and only a small remainder reaches the wafer. That is why the machine needs a light source with an enormous energy demand. Shintake’s design manages with two mirrors in the projector and loses correspondingly less. In June 2026 he presented a second design for the high-NA class, with four mirrors. He says it should achieve structures of two to three nanometres “in a much cheaper way”. The price is a smaller exposure field. Shintake himself writes that he has to revert to a stepper, which exposes field by field instead of scanning. First devices are being built at the institute; a finished machine does not yet exist.

The price. Since April Nikon has had a new chief executive, Yasuhiro Ohmura. He is attacking ASML where it has no monopoly: in the older deep-ultraviolet lithography machines that still expose most layers even of the most advanced chips. ASML charges on average around 82.5 million dollars for them. Nikon wants to undercut that because it makes many parts in-house, and in 2028 it will bring a platform designed to fit into factories running ASML machines. The starting position is thin, though: according to company data, Nikon sold three older dry scanners and not a single immersion scanner in 2026.

The own factory. Rapidus was founded in 2022 by the government and eight Japanese corporations to make chips of the 2-nanometre class in Chitose on Hokkaido. The technology comes from IBM. In December 2024 the first EUV machine in Japan arrived; in July 2025 the first prototypes ran. Volume production is to start in the second half of fiscal 2027. State support now amounts to around 2.35 trillion yen, roughly 13 billion euros, and the state is the largest shareholder. No large customer has yet firmly committed. Rapidus works with ASML machines.

III. The answer from Veldhoven

ASML has responded to Shintake’s second design. A spokesperson told the Dutch trade journal Bits&Chips that ASML and Zeiss had evaluated similar designs, some even with better performance. But they “haven’t generated chipmakers’ interest due to the reduced resolution” and the smaller exposure field.

That is probably true, and it is revealing. ASML thereby confirms that simpler and cheaper machines could exist, and that they were not built because the largest customers did not want them. For a factory exposing tens of thousands of wafers a month around the clock, throughput and image quality count for more than price. Bits&Chips itself adds that a cheaper tool with lower throughput might still find a market for smaller production runs.

This is the pattern we described in “The View in the Rear-View Mirror”, this time at the market leader itself. Whoever builds a machine to the wishes of the best customers leaves the margin open: the university lab, the smaller country, the maker of specialty chips who cannot afford 200 million dollars. The question whether it could be done more simply was not asked by the monopolist but by a physicist who came from another field.

IV. What this means for Europe

Our finding from May still holds: ASML is Europe’s strongest lever, and at the top it will not weaken in the coming years. But a monopoly lasts only as long as customers believe there is no alternative. Japan is working to undermine that belief at the lower end, and it is doing so with state strategy, money and patience. Canon worked on the stamp from 2014 before the first machine was delivered.

Europe owns the machine everyone needs, but no factory of its own that makes leading-edge chips with it. In 2023 Germany promised Intel subsidies of 9.9 billion euros for two factories in Magdeburg, a project of around 30 billion euros. In July 2025 Intel pulled out. Japan has so far put around 13 billion euros into Rapidus, more than Germany had offered Intel. The difference lies not only in the sum. Germany wanted to pay part of the construction costs of an American corporation and would have owned nothing in return. Japan is the largest shareholder of its own factory. Japan buys the machine from ASML, builds its own factory with it and at the same time looks for three ways not to need it one day. Which of the two takes its dependence more seriously is not in question.

Europe holds the trump card. Japan is learning to play without it.

Read on: ASML — Europe’s (Last) Trump Card · The View in the Rear-View Mirror · From Both Shelves

Hans Ley & Claude Dedo (Anthropic) — Nuremberg, 16 November 2026.

Sources. Bits&Chips, “TSMC embraces high-NA EUV”, 8 September 2026; ASML and TSMC, joint press release on large-format photomasks, 8 September 2026; Bits&Chips, “ASML raises tool prices on soaring chip ASPs”, 27 July 2026; EE Times, “ASML’s EUV Expansion Runs Through Zeiss”, October 2026. · Canon, press release on the delivery of the FPA-1200NZ2C to the Texas Institute for Electronics, 1 October 2024; TrendForce and Data Center Dynamics, October 2024 (price, power consumption, sales target). · Tom’s Hardware, “Japan’s DNP targets 2027 mass production of 1.4nm nanoimprint templates”, December 2025. · Bits&Chips, “Japanese researcher proposes simpler high-NA EUV optics design”, 25 June 2026 (statements by Shintake and the ASML spokesperson); Tsumoru Shintake, “High-NA In-Line Projector for EUV Lithography”, arXiv 2508.00433. · SPIE, Photomask Japan 2026, Proceedings Vol. 14275 (status of the OIST project). · Tom’s Hardware, “Nikon plans to undercut ASML on price to win back chipmaking lithography customers”, May 2026; Bits&Chips, “Nikon aims to win back ArF litho business”, 1 June 2026. · Tom’s Hardware, “Rapidus fab roadmap examined”, 2026; Igor’sLAB and Semiecosystem on Rapidus funding, April 2026; TrendForce, March 2026. · Igor’sLAB, “Aus die Maus: Intel streicht die Mega-Fabrik in Magdeburg endgültig”, July 2025; Deutsche Wirtschafts Nachrichten and Der Aktionär (dpa), July 2025 (9.9 billion euro commitment, project volume around 30 billion euros).