Lasertec Discloses the Wavelength, Not the Units

The clearest statement of why the machine exists is a sentence in Lasertec's own integrated report about defects that used not to matter. As the leading edge moves from the 3nm generation toward 2nm and angstrom-class nodes, the company writes, minute defects on EUV masks that previously had no effect begin to affect device yield, so detecting them accurately becomes the key to supporting volume production of the next generation. High-NA EUV exposure tools, at a numerical aperture of 0.55, are described as close at hand. That is the entire argument for the inspection step, made by the party with the strongest interest in making it.

The word that carries the mechanism is actinic, which the company defines simply as using the same 13.5nm wavelength as the exposure tool. Three properties follow, per the company. Inspecting at 13.5nm gives far higher sensitivity than DUV sources at around 200nm. Phase defects buried inside the EUV mask's multilayer stack can be detected directly. And masks can be inspected with the pellicle in place. None of it is quantified in either report — no minimum detectable defect size, no capture rate, no side-by-side against DUV. The claim that the same wavelength sees what prints is asserted in adjectives.

The recent product cadence is about throughput rather than a new physical principle. ACTIS A300 arrived in 2023 with three times the inspection speed, three times what the report does not say, alongside URASHIMA, a high-brightness EUV light source that later took an industry society award and whose method the company does not describe. On 31 October 2025 it announced ACTIS A200HiT for wafer fabs, this one explicitly three times the speed of the A150, achieved through optimised optics and a redesigned inspection system, with no absolute throughput figure, no stated measurement conditions and no shipment date. A pellicle particle inspection system, PELMIS EPM200, was announced in July 2025, after the fiscal year closed.

Financially, there is less structure than outsiders assume. Lasertec reports one segment — the design, manufacture and sale of inspection and measurement systems — so segment sales do not exist as a disclosed object, and the company's own English is product category rather than segment. In the year to 30 June 2025 it recorded sales of ¥251,477 million (US$1,535.27 million), operating profit of ¥122,843 million (US$749.96 million), and an operating margin of 48.8%. Anyone asking for the revenue of the mask inspection business is asking for something the accounts do not contain.

There is exactly one place where the company splits the semiconductor line, and it splits it in adverbs. The quarterly supplement guides mask inspection to a little over 80% of semiconductor-related sales for the year to June 2026, blanks inspection to a little under 15%, wafer inspection to 5%. For the third quarter alone it reports a little over 60%, a little under 30% and 10%. Semiconductor-related sales are guided to ¥165.0 billion (US$1.01 billion), so a little over 80% of that is roughly ¥132 billion (US$806 million) of mask inspection on our multiplication, with the phrase a little doing work that no digit constrains. The category called other is divided the same way, flat panel display equipment a little over 65% of it across the year against 45% in the quarter, microscopes and the rest holding the balance. Three fractions per column, rounded to the nearest word, so the parts need not sum; no time series stands behind either column; and the doubling of the blanks weighting inside a single quarter goes unremarked. Acceptance timing would explain it. So would several other things the filing does not mention.

Series-level figures exist only in a stacked chart in the results presentation, where the bars are not labelled in tabular form. Read in the order the legend gives, ACTIS was about ¥114.8 billion (US$700.9 million) for the year to June 2025 and MATRICS about ¥50.6 billion (US$308.9 million), with other semiconductor equipment behind them; the chart rounds to hundreds of millions and its parts need not sum to the audited total. Services were ¥42,959 million (US$262.27 million), up 48.3%, lifting services to 17.1% of sales from a much smaller base four years earlier. Semiconductor-related equipment grew 11.7% and added ¥21,213 million (US$129.51 million) of the increase, more in absolute yen than the ¥13,989 million (US$85.40 million) services added; the company nonetheless credits services, meaning maintenance and upgrades, with driving both the increase and the move in gross margin to 59.0% from 50.3%. The growth rates point at the installed base. The yen still arrives mostly with machines.

The nine months to March 2026 show what moves the line, and where the disclosure stops. Sales were ¥169,539 million (US$1,035.04 million), up 0.4%. Semiconductor-related equipment was ¥124,664 million (US$761.07 million), down 6.7%; services ¥42,063 million (US$256.79 million), up 35.5%; other ¥2,811 million (US$17.16 million), down 33.3%. Services added ¥11,011 million (US$67.22 million), equipment gave back ¥8,904 million (US$54.36 million), other ¥1,403 million (US$8.57 million), and what survives is the ¥704 million (US$4.30 million) that makes 0.4%. The environment paragraph immediately above describes robust demand for advanced chips on AI investment and more aggressive customer capital spending; the equipment number underneath it is 6.7% lower; the document does not join the two. Below the category line there is nothing: no unit counts in production, orders or sales, and no average selling price anywhere in these filings. Volume, price and machine mix cannot be told apart from what is published. Nor is that a habit of the downturn — in the record year to June 2023, when semiconductor-related equipment sales rose 72.1%, the increase was explained the same way, by listing categories.

Production moves first, and it is measured in the same units, being stated at selling prices. For the nine months it was ¥152,309 million (US$929.85 million), down 23.8%, with equipment production down 34.8% against equipment sales down 6.7%. Divide total production by total sales and the ratio runs 1.50 in the year to June 2023, 1.03 in the year to June 2025, 1.18 in the same nine months a year ago and 0.90 now — arithmetic of ours, from two of the company's own tables. What sits behind it barely moved. Work in process fell 3.3%, to ¥119,598 million (US$730.15 million); raw materials and supplies fell 1.0%, to ¥44,877 million (US$273.97 million). There is no finished goods line on the balance sheet at all, which is what a build-to-order machine business looks like in an account, and the two inventory lines together come to ¥164,475 million (US$1,004.12 million) on our addition — the balance sheet prints no inventories total — or 52.9% of assets. Output has been cut several times harder than revenue, and the stock behind it has barely moved.

Two lines in the same filings are called services and they are not the same line. On the product-category table, services were 17.1% of sales in the year to June 2025, 24.8% of the nine months against 18.4% in the same nine months a year earlier, and 34.4% of the third quarter alone. On the revenue-recognition note, revenue transferred over time was ¥35,197 million (US$214.88 million) of the nine-month total, or 20.8%, against 15.5% a year earlier. The four-point gap is definitional — a category count against a timing count — and neither document reconciles them. Full-year guidance puts services at ¥51.5 billion (US$314 million) inside total guided sales of ¥220.0 billion (US$1.34 billion), which is 23.4% on our division, alongside semiconductor-related sales of ¥165.0 billion (US$1.01 billion) and other at ¥3.5 billion (US$21 million); those three sum to the total exactly, which the rounded percentage table never does. The mid-term plan asks for 20% or more. Both definitions clear it, in a year the company guides sales down 12.5%.

Three customers account for most of it, and the last time the company named them the sum was 77.0%. In the year to June 2023, Taiwan Semiconductor Manufacturing Company Limited was ¥51,782 million (US$316.13 million), or 33.9% of sales; Intel Corporation ¥48,003 million (US$293.06 million), or 31.4%; Samsung Electronics Co., Ltd. ¥17,871 million (US$109.10 million), or 11.7%. Together ¥117,656 million (US$718.29 million) of ¥152,832 million (US$933.04 million), a sum the company does not itself print. The year before, the same three were 76.5% in a different order — Intel 31.6%, Samsung 24.1%, TSMC 20.8% — so the concentration held while the ranking inverted: TSMC's yen multiplied 2.75 times and passed Intel, and Samsung's fell 17.8% in absolute terms in a year total sales rose 69.1%. Disclosure convention reaches customers at about a tenth of sales or more, so the figure is a floor rather than a measure; whatever sits below the line is not named. In the report for that year, filed on 28 September 2023, the risk section ran to twelve items — market swings, research and development, key personnel, quality, intellectual property, acceptance timing, specialised parts and materials, overseas operations, currency, disasters, information security and other. Customer concentration was not one of the twelve. The nearest thing is acceptance timing, which states the same fact from the other end: the machines are expensive enough that a small number of them moving across a year-end moves the year.

Geography moved as well, and the file that carries it and the statement that carries it do not quite agree. For the year to June 2025, on a customer-location basis, the United States was the largest region at ¥67,938 million (US$414.76 million), ahead of Taiwan at ¥65,268 million (US$398.46 million) and Korea at ¥53,645 million (US$327.50 million), with Japan at ¥20,687 million (US$126.29 million); against sales of ¥251,477 million those are 27.0%, 26.0%, 21.3% and 8.2% on our division, and the overseas ratio, which the company does print, is 91.8%. Nine months later the order has changed again: Korea ¥49,185 million (US$300.27 million), up 69.3%, ahead of the United States at ¥44,562 million (US$272.05 million), up 1.4%, with Taiwan down 43.8% to ¥31,401 million (US$191.70 million) and Europe down 43.9% to ¥5,630 million (US$34.37 million). Add the quarters in the company's own highlights file and the regional totals come out a million or two below the figures in the quarterly statement, the same rounding gap that shows up in the cash flow lines; nothing turns on it, except that two Lasertec documents are two documents. China is not a row in the six-way table at all; asked about a piece of American legislation, the president put it at about 6% or 7% of last year's sales, which is the only size given, orally, for a country the split does not name. The filing states that sales are classified by the customer's location, which need not be where the tool is installed. Below the top line the geography stops. With one reported segment, the note carries sales by region and no profit at all, and in the year to June 2023 the regional split of tangible fixed assets was omitted outright on the stated ground that over 90% of it sits in Japan. Regional profitability is not something an outside reader can check.

Then there is the part of the same filing that runs the other way. Sales rose 17.8%. Orders fell 61.4%, to ¥105,226 million (US$642.41 million). Backlog fell 31.6%, to ¥315,945 million (US$1,928.85 million) — about 1.26 years of sales, against a book-to-bill of 0.42 versus 3.58 three years earlier, both ratios ours rather than the company's. Orders are reported net of cancellations and changes, which the company states plainly. On the presentation chart, ACTIS orders for the year came in at negative ¥10.5 billion (negative US$64.1 million), against ¥143.3 billion (US$874.8 million) the prior year. Both directions are in the same document.

The order book decomposes further than the sales line does. Of that ¥105,226 million of orders, equipment was ¥55,814 million (US$340.74 million), down 76.3%, while services were ¥45,489 million (US$277.71 million), up 30.9% — services 43.2% of everything ordered, against 12.7% the year before. The backlog leans the other way, 94.1% equipment, and the whole of its decline was equipment: minus ¥147,151 million (US$898.36 million) of machines against services up ¥2,530 million (US$15.45 million). The identity behind all of it holds in every year that can be checked, the years to June 2020 through June 2025, backlog change equalling orders less sales to within a million yen; the first year in the file has no prior balance to check it against. Then the series stops. Quarterly orders and quarterly backlog were last printed for the quarter to June 2024; every quarter since is a hyphen in the company's own highlights file, with no note attached, and the product-level split of orders went dark at the same time. The half-year report has no production, orders and sales chapter, and the quarterly release carries none either.

What replaces the series is a product commentary with no quantities in it. MATRICS is described as continuing strong, on a target of roughly one and a half times the previous year's orders of about ¥45 billion (US$275 million); asked whether next year steps up to ¥100 billion (US$611 million), the president said the orders had not been counted carefully enough yet to answer. The first ACTIS A200HiT was ordered in the December quarter and several have been ordered since, to more than one customer, for the N2 generation and beyond, as evaluation tools that move to volume use if the evaluation passes, shipping from the third calendar quarter of 2026. Blanks inspection is strong in both product lines and especially one of them. MATRICS X712, for nodes at 90nm and above, is aimed at replacement demand for competing machines the company says have been discontinued, in a market it estimates at several tens of billions of yen (a few hundred million dollars) over multiple years, at a shortest inspection time of 22 minutes a mask. The phrase High-NA does not appear in the quarterly supplement at all, and nothing in it says whether the HiT machine addresses the 0.55 aperture the integrated report calls close at hand. The supplement files ACTIS under EUV mask inspection and stops.

What is left is a liability line, a range, and an inventory figure that the company says corresponds to something it no longer prints. Advance payments were ¥50,626 million (US$309.07 million) at the end of March, down ¥13,762 million (US$84.02 million) over nine months; with deferred revenue of ¥7,734 million (US$47.22 million), the two together fell ¥16,113 million (US$98.37 million), and the company says the money is received around shipment and does not offer it as a stand-in for backlog. In the results presentation the ¥164.4 billion (US$1.004 billion) of inventory is described as corresponding to the backlog, with nothing wrong in its contents, which is the closest anyone comes to sizing the number that stopped being published. On 30 April it raised the order forecast for the year to June 2026 to ¥200–240 billion (US$1.22–1.47 billion) from ¥170–220 billion (US$1.04–1.34 billion), the floor up ¥30 billion (US$183 million) and the ceiling up ¥20 billion (US$122 million). First-half orders were about ¥70 billion (US$427 million), which leaves ¥130–170 billion (US$794 million–1.04 billion) for the second half, an arithmetic the company did not print and did not contest. Asked whether the quarters inside that had been running at ¥30–40 billion (US$183–244 million), the president said the company does not disclose that detail, and neither confirmed the number nor denied it. The width of the range, he says, is the number of ACTIS units. The two questions the company itself lists as most asked in investor meetings are the level of orders and demand for the new machine. It has an answer to both and a number for neither.

Research spending is where the operating line went, and the company said as much once before, in the other direction. In the record year to June 2023, research and development was ¥10,977 million (US$67.01 million), up 27.2%, and it was named as the main reason selling, general and administrative expense rose 42.0% to ¥21,726 million (US$132.64 million) — which, against sales rising faster still, left that expense at 14.2% of sales, down 2.7 points. The same line now runs the other way. Half-year research and development, disclosed as a component of the expense line, was ¥6,511 million (US$39.75 million) against ¥4,314 million (US$26.34 million), inside half-year expense of ¥13,496 million (US$82.39 million) against ¥10,723 million (US$65.46 million). Across nine months the expense line rose 21.2%, to ¥21,223 million (US$129.57 million). That is the whole of the operating decline: gross profit rose ¥2,615 million (US$15.96 million), the expense line rose ¥3,715 million (US$22.68 million), and the ¥1,100 million (US$6.72 million) difference is the 1.4% by which operating profit fell in a period when sales rose 0.4%. The gross margin producing it moved down inside the period — 60.6% in the first quarter, 59.0% in the second, 55.5% in the third — in the same three months that services reached 34.4% of sales. Price, machine mix and the rising service share cannot be separated in what is published, so the third-quarter margin has at least three available explanations and no disclosed weight for any of them.

Cash left faster than it arrived. Operating cash flow for the nine months was ¥36,823 million (US$224.80 million) against financing outflows of ¥43,169 million (US$263.55 million); investing took ¥1,210 million (US$7.39 million), so operating less investing was ¥35,613 million (US$217.42 million), and the financing line was larger than that. Dividends of ¥31,131 million (US$190.05 million) are exactly two payments: ¥19,300 million (US$117.83 million) on last year's year-end ¥214 (US$1.31) a share, and ¥11,831 million (US$72.23 million) on the interim ¥132 (US$0.81). The buyback took ¥12,003 million (US$73.28 million). It was authorised on 7 August 2025 for up to 1,000,000 shares — 1.1% of shares outstanding excluding treasury — and up to ¥12.0 billion (US$73.3 million), to be bought on the exchange between 8 August and 31 December, and it closed on 15 December with 560,600 shares bought: the money limit bound, the share limit did not. The stock is held rather than cancelled, earmarked for possible use in acquisitions, and shares issued are unchanged. The stated policy is a consolidated payout ratio of about 35% as a guideline, with working capital and growth investment ranked ahead of shareholder returns and further buybacks described as something to be considered opportunistically rather than scheduled. The outturn was between 35.1% and 35.7% in each of the seven years to June 2025, which is closer to a formula than a guideline. Held flat at ¥329 (US$2.01) a share against guided earnings of ¥801.89 (US$4.90) a share for the year to June 2026, it computes to 41.0%.

On the question everyone actually asks — the share — the filings are more careful than the discourse. The president writes that ACTIS has secured overwhelming superiority and that no competing products currently exist; blanks inspection is called an industry standard machine; the strategy is named global niche top. But across 58 pages of the integrated report there is not one share percentage and not one third-party research citation. In the English partial translation of the statutory report, monopoly and sole appear zero times, and no product is described as a world's first — the phrase surfaces once, in a line about the kind of person the company wants to hire. Market share appears six times, all qualitative: four on the risk of losing share or the effort to expand it, two on the goal of attaining and maintaining high shares. The near-total share figure that circulates in English cannot be sourced to either document.

The industry backdrop points the other way from the order book. Japanese-made semiconductor production equipment sales reached ¥5.1986 trillion (US$31.74 billion) in the year to March 2026, and the trade association forecasts ¥6.5502 trillion (US$39.99 billion) for the following year, a 14.3% compound rate through fiscal 2028. The monthly series for June 2026, on a three-month moving average, was up 26.9% year on year. That association stopped publishing bookings and its book-to-bill ratio in 2017, which leaves the order figures of individual filers as the visible ones.

The company's own plan for the year to June 2030 targets sales of ¥400–500 billion (US$2.44–3.05 billion) at an operating margin of 35% or more — below the 48.8% just recorded — with a service share of 20% or more and an average annual growth rate above 10%, described as ahead of the market's. The page does not say which year that rate compounds from, or how many years the plan runs; it says only that the year to June 2026 is the second of it. Compound the ¥251,477 million of the year to June 2025 at 10% for five years and the result is about ¥405 billion (US$2.47 billion), which clears the bottom of the range and not the top, on our arithmetic and our choice of base year. The intervening year is going the other way. Guidance for the year to June 2026, published on 30 January 2026 at ¥220,000 million (US$1,343 million), or 12.5% below the prior year, was left unchanged on 30 April, on the same day the order forecast went up and the fourth-quarter exchange rate assumption was revised to ¥155 to the dollar. Nine months in, sales were 77.1% of the guided figure, leaving ¥50,461 million (US$308.06 million) for the fourth quarter against about ¥41.2 billion (US$252 million) in the third. Lead time on ACTIS is about a year, and the units shipping from the third calendar quarter of 2026 were prepared in advance of the orders. The actual result was not disclosed as of 28 July 2026.

Yen amounts are converted at ¥163.8 = US$1, the rate on 24 July 2026 as reported by TradingEconomics, applied uniformly to every period regardless of when the amount was recorded; significant figures follow the original yen disclosure. Ratios, percentages, multiples and headcounts are not converted. A single spot rate is not a period average: the company's realised rate for the year to June 2025 was ¥148 to the dollar and its planning assumption for the following year ¥145, so nothing here will reconcile to amounts computed on those bases.