The Four Liquids

Tokyo Ohka Kogyo keeps, in the appendix of a slide deck written for retail investors and dated 6 September 2025, a table of wavelengths. The i-line, 365 nanometres, covers linewidths of roughly 350 down to 250. KrF at 248 nanometres covers 250 down to 130. ArF at 193 covers 130 down to 10. EUV, at 13.5 nanometres, covers everything below 10. A footnote says only the main ranges are shown. The table is normally read as a history of light sources. It is also a list of the four liquids the company sells.

The company's own process diagram runs: coat, expose, develop, etch, ash, then the semiconductor regions, the insulating films and wiring, the integrated circuit. Three of those steps are marked as steps where its products are used — coating, developing, ashing. Exposure is not one of them. Exposure, in the company's description, transfers the mask pattern into the photoresist; development forms the pattern; etching removes the areas where no resist is standing. The machine everybody photographs projects an image. The image has to be recorded in something, and the something is organic chemistry poured in a thin film.

How concentrated is the supply of that film? The company's own answer, from the same deck: a 24.7% global share of semiconductor photoresists, presented as global No. 1. The footnote does the real work. The figure is for 2024, it is a forecast rather than a settled actual, it is measured in shipment volume rather than value, it combines EUV, ArF, KrF and g/i-line into a single number, and it is TOK's own calculation based on a Fuji Chimera Research Institute market study. No breakdown by generation is published.

The business is having a good year. Sales for the twelve months to 31 December 2025 were ¥237,029 million (about US$1.45 billion), up 17.9%; operating income ¥47,386 million (about US$289 million), up 43.2%; net income attributable to owners of the parent ¥33,345 million (about US$204 million), up 47.0%. The company itself flags two non-recurring items inside that: a one-off gain from recognising development-related inventory, and ¥1,422 million (about US$8.7 million) of contingent consideration from an equipment business sold in an earlier period. Reporting is a single segment, materials.

Within that, electronic functional materials were ¥124,700 million (about US$761 million), up 16.0%, and high-purity chemicals ¥109,400 million (about US$668 million), up 19.6%. The materials line splits 70% front-end semiconductor photoresist, 22% back-end packaging-related materials, 8% display and other — the identical split as the prior year. Growth is disclosed in five-point increments: front-end resist +15%, within which advanced materials +20%. Advanced means ArF and EUV reported together. For 2026 the company guides the back-end share up to 25% and the display line down.

Whatever the headline growth rate is in a given year, the filing will not take it apart. Production and sales for the single materials segment are published as a yen amount and a rate of change, and nothing else: for the year to 31 December 2024, production ¥135,055 million (about US$825 million), up 11.1%, and sales ¥200,966 million (about US$1.23 billion), up 23.8%, the sales line identical to consolidated revenue. The table has columns for segment name, amount and rate of change. It has no column for quantity. A full-text search of that report returns no kilolitres, no kilograms, no tonnes, no occurrence of the word volume and none of production capacity. Orders are not there either: the company produces to stock, so the backlog section says the item is not applicable, and the revenue note omits remaining performance obligations because it expects no material contract to run beyond a year and applies the practical expedient. A price per unit requires a unit. There is none in the filing, so volume, price and mix cannot be separated — not approximately, not at all.

What the company supplies instead is a verb. Twice, four years apart, the group-level sentence is the same: the semiconductor market was below the previous year. Underneath it, 2019 is small and medium LCD panel demand and lower Taiwan-bound ancillary chemicals; 2023 is smartphone and PC demand with inventory adjustment at chipmakers; 2024 is advanced and China-bound product expanding, with a weaker yen entered as an effect on profit and no amount attached; the February 2026 revision is generative-AI demand above plan, and a weaker yen. Price appears only in the future tense: a risk of declining selling prices, and, in the integrated report, price-adjustment measures begun in 2025. The one accidental volume signal is 2023, when production fell 21.4%, to ¥121,529 million (about US$742 million), while sales fell 7.5%, to ¥162,270 million (about US$991 million), on two bases the filing never says are the same.

The EUV line is where the disclosure thins. TOK does not break out EUV resist revenue at all; it is folded into advanced materials with ArF, and no shipment volume is given. What is shown are three chemistries — chemically amplified, small molecule, metal oxide — with the company's own estimate that the first is already in practical use as of 2025 and the other two land somewhere between 2025 and 2030, plus patterning examples at 14 nanometre lines and 18 nanometre contact holes. It projects EUV resist revenue growing 150% from fiscal 2024 to 2027, packaging materials 70%, and one packaging-adjacent line 700% from a base the chart draws as almost nothing. It does not disclose the base. A percentage without a denominator is a mood.

The geographic breakdown, by contrast, is printed in full — though the filing carries no footnote on the classification basis, so whether it is by customer location is not stated, and Europe is not broken out. For 2025: Taiwan ¥85,604 million (about US$523 million), China ¥48,611 million (about US$297 million), Japan ¥36,638 million (about US$224 million), Korea ¥30,447 million (about US$186 million), the United States ¥24,019 million (about US$147 million). Non-Japan is 1 − 36,638 ÷ 237,029, or roughly 84.5%, on our arithmetic; the company prints the components and not the ratio. One customer, Taiwan Semiconductor Manufacturing Company, accounted for ¥79,631 million (about US$486 million), 33.6% of consolidated sales, against 30.4% the year before — as disclosed; the filing does not say whether that is the group or the single legal entity.

The securities report lists thirteen risks: industry cycles, currency, research and development, intellectual property, raw material procurement, product liability, disaster and accident, environment, legal, overseas operations, information leakage, cybersecurity, climate. Customer concentration is not listed as a risk item of its own. Geopolitics appears inside the raw-material procurement risk, concerning the company's own suppliers. Export control appears as government regulation over import and export restrictions, under legal risk, with no country named. A company that puts its own share of semiconductor photoresists at 24.7% — a 2024 forecast, volume basis, its own calculation from a Fuji Chimera study — files no risk factor for being the place everyone buys from.

The share it files no risk factor for has a history. The named customer was 23.4% of total sales in 2018, 23.6% in 2019, 29.1% in 2022, 28.0% in 2023, 30.4% in 2024 and 33.6% in 2025; the reports read here leave 2020 and 2021 blank. The 2023 dip is not a retreat — purchases from that customer fell 11.0%, from ¥51,029 million (about US$312 million) to ¥45,419 million (about US$277 million), in a year when total sales fell 7.5%. Nor is 2025 the rest stalling: the ¥79,631 million (about US$486 million) in the report for the year to 31 December 2025, against ¥61,135 million (about US$373 million) in the report for 2024, is 79,631 ÷ 61,135 − 1, a rise of 30.3%, against 17.9% for the group. A ratio rises when the numerator moves faster, which is the whole published explanation. Divide the customer line by the Taiwan line in each of those two reports — 79,631 ÷ 85,604 and 61,135 ÷ 68,623 — and one customer is 93.0% of Taiwan against 89.1%, if those two tables share a basis; the filing does not say. As for the risk list: the reports for 2019, 2023 and 2024 each carry eleven items, and they are the same eleven. The thirteen above are those eleven plus cybersecurity and climate. Which year each of the two arrived is not something the reports read here establish.

Capital spending was ¥28,723 million (about US$175 million) in 2025 and is guided to ¥35,800 million (about US$219 million) for 2026, against depreciation of ¥8,807 million (about US$53.8 million) rising to a planned ¥11,300 million (about US$69.0 million). The largest item is a building at Koriyama, Fukushima, which the securities report calls an electronic functional materials manufacturing building and the February 2026 deck calls the world's largest photoresist manufacturing building, at around ¥20.0 billion (about US$122 million) or more. The filing's figures: ¥20,000 million (about US$122 million) budgeted, ¥12,573 million (about US$76.8 million) already paid as of 31 December 2025, started July 2024, completion during 2026. The company gives a completion date rather than a production date and says the resulting capacity increase is difficult to estimate. The two decks differ: September 2025 puts the site's earnings contribution in the second half of 2026, February 2026 marks it at the first half of 2027, on an EBITDA-trajectory chart that never labels the point a start-up date. Neither deck says anything was changed.

In Korea, phase one at Pyeongtaek: approximately ¥12,000 million (about US$73 million), about 6,300 square metres of building area, a high-purity chemical manufacturing building plus warehouses and ancillary facilities, operations scheduled for the second half of 2027, at the group's second Korean site after Incheon. The English release explains it under a customer-oriented strategy and the aim of establishing stable supply of high-purity chemicals in South Korea. It names no customer, gives no capacity, and never uses the word proximity. The demand behind it is described as further growth in demand, with no figure attached.

On 9 February 2026 the board revised the 2027 targets: sales from ¥270.0 billion to ¥295.0 billion (about US$1.65 billion to US$1.80 billion), operating income from ¥48.0 billion to ¥58.0 billion (about US$293 million to US$354 million). Two reasons are given and only two, generative-AI demand above plan and a yen weaker than assumed. The assumed rate moves from 135.0 to 150.0 yen per dollar as a three-year average, with no sensitivity and no other currency disclosed, so the ¥25.0 billion (about US$153 million) cannot be divided between the two causes. The same release prints the 2025 result alongside the old target: operating income ¥47.3 billion (about US$289 million) against a 2027 target of ¥48.0 billion, which is 47.3 ÷ 48.0 = 98.5% of it, and a 2025 return on equity of 15.6% against a 2027 target now raised to 14.0%. The release mentions neither comparison. On EBITDA the documents disagree about the old target: the mid-term plan deck revised that day prints ¥60.0 billion, while the release announcing the revision and the 2024 integrated report both print ¥61.0 billion (about US$366 million and US$372 million). The new figure, ¥72.0 billion (about US$440 million), is the same in the release and in the revised deck; the integrated report, written before the revision, carries only the old one. The raise is either ¥12.0 or ¥11.0 billion (about US$73.3 million or US$67.2 million).

In July 2019 the materials layer became foreign policy. On 1 July, effective 4 July, Japan removed three items — fluorinated polyimide, resist, and hydrogen fluoride — together with their related manufacturing technology, from general bulk export licensing for shipments to South Korea, requiring individual applications and case-by-case review. The items are defined by schedule entries, resist being item 7(19) of Appended Table 1 of the Export Trade Control Order. The stated grounds are two: that trust between Japan and South Korea had been seriously damaged, and that inappropriate incidents had occurred concerning export control related to South Korea. The incidents are not described. No company and no product is named anywhere in the notice. It was not an embargo; it was a change of licence category.

The filings show what the licence change did to the numbers, which is not much anyone can attribute. Korean sales went from ¥12,286 million (about US$75.0 million) in 2018 to ¥11,659 million (about US$71.2 million) in 2019, down 5.1%, and from 11.7% of sales to 11.3%. The company assigns none of that to the measure; its stated cause for the year is the semiconductor and display market. In 117 pages the tightening is mentioned once, in the past tense, inside a list of management measures that also contains personnel-system reform and training. No risk factor names Korea. The Korean capacity project in that filing — capacity doubled, completion April 2020 — was started in December 2018, seven months before the measure, and Korean fixed assets rose 10.5% that year.

The consolidated filing gives sales by six regions and fixed assets by five, China being in the first list and absent from the second, and profit by no region at all. An audited profit figure for one overseas arm — the only one this reporting could locate — therefore sits outside the group accounts entirely: filed with Korea's Financial Supervisory Service, in Korean, under Korean GAAP, by TOK Advanced Materials, registered in Incheon, 90% TOK and 10% Samsung C&T, which is not Samsung Electronics. Its sales were ₩314.5 billion in 2025, up 20.6%. Operating profit of ₩40.6 billion on that is 40.6 ÷ 314.5 = 12.9%, against ₩27.8 billion on ₩260.7 billion, or 10.7%, the year before. Net profit was ₩36.9 billion, on an effective tax rate the notes put at 9.23% against 19.57%, from credits whose statutory basis is not stated. It pays no dividend.

That filing is also the only place in these documents where a growth number comes apart at all, and it comes apart by line item rather than by unit. Of the ₩53.8 billion increase in sales, ₩47.9 billion — 47.9 ÷ 53.8, or 89.2% — is manufactured product, up 25.8%; goods bought in for resale rose 2.7%; a services line rose 24.7%. Cost of sales on the manufactured line rose 23.6%. Headcount on the regulator's company profile went 136 to 170 to 177, and payroll rose 18.6%. In 2025 the entity moved ₩100.7 billion out of construction in progress and into working assets, against ₩0.6 billion the year before, with ₩41.4 billion still unfinished, and borrowed ₩35.0 billion from its parent at 4.03% for facilities. From the 2024 audit onwards the stocktake locations include a Pyeongtaek warehouse. What none of it gives is a unit either: made versus bought is a split of what was sold, not of how much.

What that entity cannot settle is the thing its margin sits on. Purchases from related parties were 72.1% of its cost of sales in 2025, down from 80.8% a year earlier on the same related-party basis. Of those purchases, ₩170.3 billion of ₩172.1 billion, or 99.0%, came from the parent; set that against a cost of sales of about ₩238.7 billion — 314.5 × (1 − 24.1%), using the gross margin the audit report states — and the parent alone is roughly 71.4% for 2025, while the 2024 note gives no parent-only split. A technology fee of 3.5% to 8% of net sales, running at ₩14.5 billion on ₩314.5 billion, or 4.6%, goes back to Tokyo. Only 4.0% of the entity's sales are to the group, ₩12.6 billion of ₩314.5 billion; the rest goes to customers it does not name. The transfer-pricing method is not disclosed, so whether the margin moved on operations, mix, the yen or the price at which the parent sells to itself cannot be determined from either filing. Its royalty note still describes a contract signed on 22 October 2012 to run for ten years.

On conversion: yen figures are converted at ¥163.8 = US$1, the yen rate quoted by Trading Economics on 24 July 2026, and that single current rate is applied uniformly to every period mentioned above. It is not the rate prevailing in each period, so the dollar equivalents for 2018, 2019 and 2023 are translations at today's rate rather than the amounts anyone transacted at the time. Korean won figures are left in won, because the single rate used here is a yen rate and no won rate is applied to them. Dollar equivalents are rounded and marked as approximate; percentages, ratios, multiples and counts are left unconverted.