The Percentage Column

Tokyo Electron's Japanese annual securities report for the year ended 31 March 2026 runs to 156 pages, and one small table on page 23 carries more weight than most of the rest. The sub-section is headed status of production, orders and sales; the table under it is introduced as sales by principal counterparty and the ratio of those sales to total sales, and it is two tables of two rows each, prior year and current. The current one: Samsung Electronics Co., Ltd., ¥368,079 million ($2,247 million), 15.1 percent. Taiwan Semiconductor Manufacturing Company Ltd., ¥315,813 million ($1,928 million), 12.9 percent. The prior-year table above it prints 11.8 and 11.5. A note says the amounts include sales to customers belonging to the same corporate group as the named customer. Neither table carries a third row.

The received wisdom is that this table is the reason to read the original. Tokyo Electron's English document is, by its own note on the second leaf, an English translation of part of the original Japanese annual securities report: the consolidated financial statements, 65 pages as published, opening at the balance sheet. Search it for orders received, backlog, status of production, principal customers, percentage of net sales — none of them appear, and neither does the word percentage. Its auditor's report, on its page 61, sends the reader to the Japanese original for fee-related information. The two-row table sits inside the management discussion, as the third item of a sub-section, and that sub-section is not in the English.

Except that the two names are. Inside the segment note, under Major customer information, the English document lists Samsung Electronics at ¥368,079 million and Taiwan Semiconductor at ¥315,813 million on its page 52, with a footnote of its own — the amount includes sales to the customer and its subsidiaries — and the prior year in a separate table on page 51 at ¥286,800 million ($1,751 million) and ¥280,618 million ($1,713 million). The Japanese note says a similar thing in different words: customers belonging to the same corporate group. The percentage column is the part that stays in Japanese. The English supplies the numerators, and it supplies the denominator too: consolidated revenue of ¥2,443,533 million ($14,918 million) stands on the English document's own income statement on page 3, and the same figure recurs in its regional and disaggregated revenue tables on pages 48 and 51. 368,079 ÷ 2,443,533 = 15.06%, which the Japanese column prints to one decimal as 15.1. 315,813 ÷ 2,443,533 = 12.92%, printed as 12.9. The Japanese-only content, on this point, is one division.

The rest of the sub-section is three sentences. Production volumes are omitted on the stated ground that production results follow a trend similar to sales results. Order results are not disclosed: short-term order trends fluctuate heavily with customers' investment behaviour and are therefore, the company writes, not necessarily an appropriate indicator for forecasting medium- to long-term performance. Sales are handled by cross-reference to the management discussion that already contains them. The word for orders appears five times in each of the three reports, and the five are the same five: the sub-section heading, twice inside the sentence that states the non-disclosure, once in the market volatility risk item, once in the safety risk item.

Run the table backwards and it looks like a story. FY2023: Intel Corporation ¥357,636 million ($2,183 million) at 16.2%, TSMC ¥320,427 million ($1,956 million) at 14.5%, Samsung ¥275,916 million ($1,684 million) at 12.5% — three names, and 16.2 + 14.5 + 12.5 = 43.2% between them. FY2024: Samsung alone, ¥237,441 million ($1,450 million), 13.0%. FY2025: two names at 11.8 and 11.5. FY2026: two names at 15.1 and 12.9. Read as a concentration series, that is a collapse followed by a recovery and a climb.

It is not a concentration series. It is a disclosure series. A name leaves the table when it falls below the reporting threshold — conventionally ten percent of sales, though the filing does not print the number — and below the threshold is not zero. The series is truncated by the threshold itself: below it, the value is an unknown between zero and ten, and that is a property of the reporting convention rather than of any company's choice about what to print. Intel does not appear in the FY2024 table or after. Falling off the table fixes the value as an unknown between zero and ten percent; it is not a statement that sales to that customer declined, and the amounts do not establish that they did. FY2024 shows the mechanism without ambiguity: Samsung's amount fell from ¥275,916 million to ¥237,441 million while its ratio rose from 12.5% to 13.0%, because revenue fell 17.1% to ¥1,830,527 million ($11,175 million). The numerator went down and the share went up.

FY2026 runs the opposite arithmetic. Revenue rose 0.5% to ¥2,443,533 million ($14,918 million), so the pair's combined share — 23.34% to 27.99%, sums neither table prints, computed here from the amounts — rose 4.65 points on the numerators. Samsung added ¥81,279 million ($496 million), up 28.3%. TSMC added ¥35,195 million ($215 million), up 12.5%. The layer beneath that is thinner. The annual report's revenue disaggregation note gives a two-line split — new equipment ¥1,817,250 million ($11,094 million) against ¥1,893,080 million ($11,557 million) the year before, field solution services and others ¥626,282 million ($3,823 million) against ¥538,488 million ($3,287 million) — which is minus 4.0% and plus 16.3% on that arithmetic. Those two lines add to ¥2,443,532 million against the ¥2,443,533 million printed beneath them; the prior year's two add to ¥2,431,568 million, which is the printed total. The seven regional lines in the same note add to ¥2,443,531 million and ¥2,431,565 million against printed totals two and three million higher. Beneath the two product lines there are no unit counts, no average selling prices and no quantitative bridge, so volume, price and mix cannot be separated from what the report discloses. The management discussion describes strong sales of high-value-added products and a steady increase in modifications, parts and service, and attaches no unit or price figure to either line.

A finer mix exists in the results presentation, on a denominator the annual report never uses. New semiconductor production equipment — SPE, in the presentation's own abbreviation — is ¥1,775,400 million ($10,839 million), and splits 59% non-memory, 31% DRAM, 10% non-volatile memory. Field solutions — parts and service, plus used equipment and modifications — is a separate ¥626,000 million ($3,822 million). The presentation states its figures in hundreds of millions of yen and states, on its disclaimer slide, that amounts are truncated below the unit shown; truncation accounts for less than ¥100 million a line. Neither line matches its annual-report counterpart. Its equipment line runs ¥41,850 million ($255 million) below the annual report's new equipment. Its field solutions line runs ¥282 million ($1.7 million) below the annual report's field solution services and others, and ¥188 million ($1.1 million) below it in the prior year — both gaps beyond what the stated truncation accounts for. The two documents label the lines differently: one names semiconductor production equipment and the other new equipment, one carries an and others and the other does not, and each states its own figure. The application percentages and the customer percentages cannot be laid on top of each other, because one denominator excludes field solutions and the other includes it. The presentation states the exclusion in a footnote on the same slide — the graph shows the composition of new equipment sales; field solution sales are not included — and carries it on four further slides. The annual report gives no cross-reference between the two families of percentages.

The other heavily loaded number is the regional table, and its definition is the load-bearing part. Sales are classified by countries or regions based on location of customers — not shipment destination, not where the tool is installed. Seven regions: Japan, North America, Europe, South Korea, Taiwan, China, Others. There is no Southeast Asia line; it is inside Others, which the results presentation labels Southeast Asia and others on its annual regional slide. FY2026: China ¥832,555 million ($5,083 million), South Korea ¥543,858 million ($3,320 million), Taiwan ¥499,853 million ($3,052 million), Japan ¥239,427 million ($1,462 million). The percentages are not printed. The components are.

So two customer-level cuts of the same year now exist, and they do not nest. Samsung's ¥368,079 million is 67.7% of the South Korea line; TSMC's ¥315,813 million is 63.2% of the Taiwan line. Both ratios are our own cross-table arithmetic rather than disclosed figures, and on the definitions the two tables state, neither is one region's sales to one customer: the numerator aggregates the named company with its group companies wherever they sit, the denominator sorts by where the customer is. A named customer's overseas sites would be counted in the customer row and in some other region's row at once, though the filing states neither the split nor whether it occurs. The two tables are cut on different axes and share only the consolidated revenue denominator — the customer table's own two-name total is outside what either table prints.

China is both the largest region, at 34.07% by that arithmetic, and the largest decline: ¥832,555 million ($5,083 million) against ¥1,015,060 million ($6,197 million), down ¥182,505 million ($1,114 million), or 18.0%. A second series exists in the results presentation, which prints the annual share as 34.1 on slide 27 and puts the January–March quarter at 26.8% on slide 5. A third appears in the earnings call, where management put China's share of the wafer fab equipment market at the high thirties in calendar 2025, falling to the mid thirties in 2026. That last one is the market's shape, not the company's.

None of the three carries a cause for the movement. China appears nine times in the 156-page report, on six pages: twice in the corporate chronology on page 4, once in the subsidiary table on page 6, twice in the property table on page 28, once in the revenue-recognition table on page 123, once in each of the two regional tables on page 126, and once in the management discussion on page 21. That last one is a market-environment sentence — capital investment in China showed a pause while investment in semiconductors for generative AI grew markedly — stated about the equipment market rather than about the company's own China line, which appears only in the notes. The geopolitics risk item names no country. The legal item names fields rather than individual regulations: imports and exports, environment, competition, trade secret protection, labour, prevention of corruption and bribery, economic security.

The risk section runs to sixteen items: market volatility, R&D, geopolitics, procurement and production and supply, safety, quality, environmental response, legal, intellectual property, information security, human capital, infectious diseases and natural disasters, finance, M&A, IT and operations, site expansion. Customer concentration is not one of the sixteen headings; the disclosure form does not enumerate risk headings, and the company chooses its own. The nearest sentence, inside market volatility, states that the group's sales tend to be affected by the investment trends of leading advanced semiconductor manufacturers. The two named customers were 27.99% of revenue that year on the arithmetic above, a sum the tables do not carry. The concentration is quantified customer by customer in the sales sub-section; the risk section addresses the same exposure in qualitative terms.

The regulatory backdrop is written at the same altitude. Japan's Ministry of Economy, Trade and Industry added 23 semiconductor manufacturing items to export control by ministerial ordinance No. 25 of 2023, published 23 May 2023 and effective 23 July 2023: three cleaning, eleven deposition, one annealing, four lithography, three etching, one inspection. The stated rationale is preventing military diversion as the international security environment grows more severe, complementing the Wassenaar Arrangement while comprehensively considering the latest export control trends of relevant countries in respect of semiconductor manufacturing equipment. The rationale names no country, and it names no company. What it names is equipment categories and two dates.

The word for regulation appears once in the published Q&A summary of the results call, in a single answer. An analyst put a roughly five-point decline in calendar 2025 etch share into his question; the answer names three causes for the decline — the timing of deliveries between calendar 2024 and 2025, customer mix, and customers' delivery adjustments due to regulation — and does not restate the five points. It names no particular regulation, no deliveries and no amount. Across the fifteen questions in the summary, no figure is attached to export control, and no question asks about the company's own China ratio or the basis on which it is computed.

The order book appears again in the same document, and again without a figure. Asked how the second half of FY2027 could exceed the first, management said demand is quite strong, and that while orders have not been taken, on an inquiry basis the production slots for the second half are on pace to fill. The word for orders appears seven times in the Q&A and never with a number attached. The measure offered for the second half is inquiries rather than orders, and in the annual report order results are the item stated as not disclosed.

Guidance was shortened at the same moment. The company forecasts the coming half year only: FY2027 first-half revenue of ¥1,570,000 million ($9,585 million), up 33.1%, and operating income of ¥431,000 million ($2,631 million), up 42.2%, both against the prior-year first half. In the call, the reasons given are customer yields, cleanroom availability, geopolitical factors and energy, and the stated aim is to avoid causing confusion in the capital markets. The forecast covers the first half only, so the release does not carry a full-year comparison of the kind the prior years' releases carried. The release states that the full-year figure will be disclosed with the interim results, and that the year-end dividend, undetermined as of the release, will be disclosed at the same time.

The year underneath all this is a particular shape. Revenue up 0.5% to ¥2,443,533 million ($14,918 million), which the report calls a second consecutive record. Operating income down 10.4% to ¥624,936 million ($3,815 million). Net income up 5.6% to ¥574,454 million ($3,507 million), also a record, helped by a pre-tax gain of ¥115,494 million ($705 million) on sales of investment securities, which the report attributes to a partial sale of strategic shareholdings. Gross margin fell 1.8 points to 45.3%, which the management discussion attributes to higher raw material prices and increased personnel costs; operating margin fell 3.1 points to 25.6%, attributed to active research and development. R&D rose 11.1% to ¥277,866 million ($1,696 million), or 11.4% of sales. The report names the factors and does not size them.

Foreign ownership, the number that brings foreign readers to a Japanese filing in the first place, is 47.61% — computed from the unit counts in the shareholder table on page 37, since the printed percentages sum to 47.60. The denominator excludes fractional shares and still includes 15,474,275 treasury shares; remove those and the figure is 49.22%. The same table puts financial institutions at 36.37% on that unit basis. The major shareholders table on the next page lists the two trust banks that hold on behalf of beneficial owners at 24.43% and 10.13% of shares outstanding excluding treasury — a different denominator, named in that table's column heading, and not a breakdown of the 36.37%. The corporate governance report reproduces the same two figures under a column headed only percentage. Neither table classifies holdings registered through nominee trust accounts by the nationality of the beneficial owner, so the register-basis 47.61% and a beneficial-ownership figure are different measures. Neither table permits the second to be computed, and no direction of difference is asserted here.

So the thing the Japanese filing uniquely contains, on the customer question, is a percentage column, and that column is a division the English reader can do unaided. The customer-location basis is stated in English too, in the same words. The sub-section's prose — heading and three sentences, 263 characters once whitespace is normalised — is identical in the FY2024, FY2025 and FY2026 Japanese reports; whether any English document carries it is a different question, and not one these pages answer. The customer table names two customers and does not name anyone below ten percent. The regional table gives seven regions and defines a region as where the customer is. Both tables carry what their definitions admit. The definitions are two sentences long, and they are printed in both languages.

Yen amounts are converted at ¥163.8 = US$1, the rate on 24 July 2026, applied uniformly to every period shown; dollar equivalents are therefore not the rates prevailing in the years concerned, and significant figures follow the yen originals. The rate is a single external market quotation used for reference — not a company disclosure, not an accounting rate, and not a rate any of the filings cited here uses. Percentages, ratios, point changes and counts are left unconverted, as are the one-, two- and three-million-yen differences between component sums and printed totals. Where a percentage or a change is not stated in the source, it is computed from disclosed yen amounts, and where the division is short enough to print, it is printed.