The Third Bucket

A silicon wafer is a disk sliced to less than a millimetre thick from a single crystal, pulled from polysilicon that Sumco's own English materials describe as eleven-nine pure: 99.999999999 percent. The company runs the sequence end to end — polycrystalline silicon process, monocrystalline pulling, wafer forming, specialized processing — and the process chart names four things coming off the end of it: polished wafers, annealed wafers, epitaxial wafers, and silicon-on-insulator. That is where the chart stops. Nothing is printed on the disk yet. Its defining property is that it has no features, which turns out to matter for how it gets paid for.

In 2025 the world shipped 12,973 million square inches of silicon wafer for semiconductor use, up 5.8 percent on 2024, according to SEMI's Silicon Manufacturers Group. Revenue on those wafers was US$11,402 million, down 1.2 percent, a third consecutive annual decline. Area up, money down, the two lines apart by roughly seven percentage points in a single year. Divide the second by the first and the blended figure is about US$0.879 per square inch, the lowest of the six years SEMI publishes. That division is mine rather than SEMI's, and it absorbs every change in diameter, product and customer mix at once.

Sumco's own year lands in that gap. Consolidated revenue for the twelve months to December 2025 was ¥409,670 million (US$2,501 million), up 3.3 percent. Operating profit was ¥1,342 million (US$8.19 million), down 96.4 percent, an operating margin of 0.3 percent. Below that: an ordinary loss of ¥3,886 million (US$23.7 million) and a net loss attributable to owners of ¥11,751 million (US$71.7 million). The company reports a single segment, high-purity silicon, so there is no 300mm-versus-200mm split, no product mix, nothing between the top line and the loss.

The first quarter of 2026 was reported on 12 May and did not settle it. Revenue was ¥101,402 million (US$619.1 million), down 1.0 percent on the same quarter of 2025 and ¥1,402 million (US$8.56 million) above the company's own forecast. Cost of sales rose 11.4 percent to ¥92,944 million (US$567.4 million), which left gross profit of ¥8,457 million (US$51.63 million) against ¥19,010 million (US$116.1 million). Operating profit of ¥5,990 million (US$36.57 million) became an operating loss of ¥5,273 million (US$32.19 million), and the net loss was ¥8,469 million (US$51.70 million). Revenue flat, the loss arriving on the cost line. The full year of 2025 had the same shape underneath it: ¥1,342 million of operating profit for the year against ¥7,457 million (US$45.52 million) booked in the first half, which leaves ¥6,115 million (US$37.33 million) of operating loss in the second half, by subtraction.

The mechanism is visible in two rows that crossed. Capital expenditure was ¥79,957 million (US$488.1 million) in 2025, against ¥315,415 million (US$1,926 million) in 2023 and ¥214,927 million (US$1,312 million) in 2024. Depreciation went the other way: ¥115,692 million (US$706.3 million) against ¥78,986 million (US$482.2 million) a year earlier, which the company attributes to the leading-edge 300mm capacity it built. Operating cash flow was ¥100,040 million (US$610.7 million) and EBITDA ¥112,442 million (US$686.5 million), a 27.4 percent margin. The dividend was ¥20.00 (US$0.12) a share, ¥7.0 billion (US$42.7 million) in total, from ¥21.00 (US$0.13) the year before.

Two of those numbers come in two definitions each, and neither pair is reconciled for the reader. The ¥115,692 million (US$706.3 million) is the cash flow statement's depreciation, and ¥4,592 million (US$28.03 million) of it was charged below the operating line; take that out and ¥111,100 million (US$678.3 million) remains, which added to the ¥1,342 million (US$8.19 million) of operating profit gives ¥112,442 million (US$686.5 million). That last step is arithmetic rather than disclosure. In its first-quarter 2026 materials the company defines EBITDA as operating profit plus depreciation charged within operating profit, and publishes the figure for that quarter; it does not publish one for the year. The check on the arithmetic is that ¥112,442 million over revenue of ¥409,670 million is 27.4 percent, which is exactly the EBITDA margin the governance report prints as 2025's actual in the pay formula. In the first quarter of 2026 the same split runs ¥30,812 million (US$188.1 million) with ¥2,108 million (US$12.87 million) below the line, against ¥87 million (US$0.53 million) a year earlier, and no reason given. Capital expenditure has its own gap: ¥79,957 million (US$488.1 million) in the business report against ¥111,033 million (US$677.9 million) of cash paid for tangible and intangible assets in the cash flow statement. The quarterly presentation labels its own capital expenditure as recognised on acceptance; the business report attaches no basis at all to the annual total. The ¥31,076 million (US$189.7 million) between the two is within ¥33 million (US$0.20 million) of the ¥31,109 million (US$189.9 million) fall in equipment-related payables reported on the same statement.

A decomposition does exist, and it is four buckets wide. The results deck bridges operating profit under four labels: cost, depreciation, sales and production related and other, and foreign exchange including the Taiwan dollar. Between the first quarter of 2025 and the first quarter of 2026 the move is cost minus ¥1.8 billion (US$11 million), depreciation minus ¥6.1 billion (US$37 million), sales and production related and other minus ¥3.5 billion (US$21 million), and foreign exchange plus ¥0.3 billion (US$2 million), totalling ¥11.1 billion (US$67.8 million); the deck rounds its endpoints to hundred-million-yen units, so that total is the ¥11,263 million (US$68.76 million) the accounts show. Against the previous quarter the depreciation line turns positive, plus ¥4.7 billion (US$29 million), which is the fourth quarter's own charge falling away rather than a trend. The values are printed on the chart rather than in the text, and they add to the reported figure.

The half-year guidance states it in one line. Against ¥7,457 million (US$45.52 million) of operating profit in the first half of 2025, the company forecasts a ¥7,700 million (US$47.01 million) operating loss on revenue of ¥213,400 million (US$1,303 million), up ¥8,028 million (US$49.01 million). The deck splits the ¥15.1 billion (US$92.2 million) swing as cost minus ¥4.0 billion (US$24 million), depreciation minus ¥11.6 billion (US$70.8 million), sales and production related and other minus ¥4.6 billion (US$28 million), foreign exchange plus ¥5.1 billion (US$31 million). Depreciation is ¥11.6 billion of ¥15.1 billion, about 77 percent. The stated currency sensitivity is roughly ¥1.2 billion (US$7.3 million) of annual operating profit per yen against the dollar, which is not the same measure as the bridge's exchange line, because that one also carries the Taiwan dollar.

The third bucket is where the question is kept. Volume, unit price and product mix sit inside it together, and nothing opens it. In 2024 that bucket was minus ¥56.8 billion (US$347 million) against an operating decline of ¥36.1 billion (US$220 million) — larger than the decline itself, offset by ¥14.7 billion (US$89.7 million) of currency and ¥11.7 billion (US$71.4 million) of cost — while depreciation was minus ¥5.7 billion (US$35 million). In the forecast the company published on 12 May for the first half of 2026, the weights invert: of the year-on-year deterioration, depreciation contributes minus ¥11.6 billion (US$70.8 million) and the bucket minus ¥4.6 billion (US$28 million). Those are contributions to a change rather than levels, and one of the two periods being compared has not happened yet. The direction of each half is given in words only. Shipments fell as customers corrected wafer inventory; long-term contract prices were kept, then maintained, quarter after quarter. The securities report explains 2024 by volumes down sharply, materials cost, heavier depreciation and a helpful currency, and the words price and mix do not appear anywhere in that analysis section. The earnings release carries no bridge. Volume and price cannot be separated from what is disclosed.

For the heavier depreciation the company gives its own explanation, and it points backwards: the capital expenditure it carried out to expand leading-edge 300mm capacity, named as the reason in the results release and again in the business report. That is the spending line above, falling in each of the two years since 2023. What the filings will not say is when the next build begins. The quarterly presentation gives the quarter's capital expenditure and the next quarter's depreciation, and no annual figure for either — no full-year capital expenditure plan, no full-year depreciation outlook. For 2026 the company published a first-quarter forecast and nothing else: revenue of ¥100.0 billion (US$610.5 million), an operating loss of ¥6.0 billion (US$37 million), an assumed rate of ¥155.0 to the dollar. Three months later, with that quarter reported, there was still no full-year guidance, only a cumulative second-quarter forecast. There is nothing to disagree with.

The 200mm side, meanwhile, is being subtracted physically. Wafer production at the Miyazaki plant ends by the close of 2026, with 200mm moving to Nagasaki and Imari, 150mm to Indonesia, and 125mm and below stopping as unprofitable; crystal pulling stays. The charge was ¥5,814 million (US$35.49 million), taken entirely in the fourth quarter of 2024 — small against revenue, but it dates the decision. In 2025 there were no extraordinary items at all, which is why the ordinary loss and the pre-tax loss are the same ¥3,886 million (US$23.7 million).

The dividend was set in the same gap. For 2025 the company declared ¥20.00 (US$0.12) a share, ¥7,003 million (US$42.75 million) in total, with the payout ratio printed as a dash; the release gives no reason for the dash, and the year's ¥20.00 sits against a per-share loss of ¥33.60 (US$0.21). Cash dividends in the financing section were ¥5,602 million (US$34.20 million), plus ¥1,909 million (US$11.65 million) to non-controlling interests, which is a different basis for the same word. For 2026 both halves were undecided in February. On 12 May the interim was revised from undecided to ¥10.00 (US$0.061) a share — a forecast, in the company's own wording something it plans to do, not a declared dividend — and the year-end was left undecided, so no annual figure exists. The stated basis is six items: the half's expected profit level, the outlook, capital expenditure funding needs, free cash flow, EBITDA, and the state of distributable reserves. No threshold or ratio is attached to any of them.

There is also no multi-year number to miss. Plans are drawn up half-yearly and not disclosed, and the governance report answers two code principles by explaining rather than complying, on the stated ground that medium-term forecasts of sales and profit do not necessarily help stakeholders. The cost of capital appears as the cost of capital the company understands, without a figure; IRR is used case by case, without a hurdle. The one ROE in the document is in the pay formula: a 10.0 percent target against minus 2.1 percent, that ROE computed with ¥572 million (US$3.49 million) of after-tax subsidy income stripped out, beside a 40.0 percent EBITDA margin target against 27.4 percent, for which the report states no basis at all. Nothing was paid. The cash bonus runs off half-year net profit, plus ¥2,584 million (US$15.78 million) and then minus ¥14,908 million (US$91.01 million); the two sum to minus ¥12,324 million (US$75.24 million), which is the reported net loss of ¥11,751 million (US$71.7 million) plus the ¥496 million (US$3.03 million) and ¥75 million (US$0.46 million) of after-tax subsidy income the bonus formula removes from each half, leaving about ¥2 million (US$0.01 million) unaccounted for. Two formulas, two different subsidy adjustments, one word for both. The only dated multi-year target in the report is 10 percent women in management at the parent and 12 percent across the group by 2030.

On concentration, the company is willing to be specific about itself. It puts its share of wafers for leading-edge logic above 50 percent, and phrases it as more than one in two of the world's advanced electronic devices starting on its silicon. The source line under the chart reads: SUMCO estimates based on various data. Leading-edge is not defined, the basis is not stated, and the competing slice is labelled Other companies. In 2025 it collected supplier awards from TSMC for a twelfth consecutive year and a seventeenth time, from Samsung Electronics for a fifth year and nineteenth time, from Sony Semiconductor Solutions for an eleventh, and from KIOXIA for the first. Those are counts of ceremonies, which is a different class of evidence than a share.

Long-term contracts are where the shape of the market becomes a pricing fact. For the fourth quarter of 2025 the company said long-term contract prices were held for both 300mm and 200mm; for the first quarter of 2026, that they are being maintained. It discloses no count, no tenor, no counterparties and no covered volume. The nearest quantity in the accounts is ¥13,161 million (US$80.35 million) of transaction price allocated to unsatisfied performance obligations, to be recognised over roughly six years, and that figure excludes contracts originally expected to run a year or less, so it is not the contract book. Meanwhile, on the same slide, customers are described as planning full inventory correction in legacy products, with purchase volumes expected to be adjusted.

Then there is the question of who buys any of this. The only counterparty disclosed at ten percent or more of consolidated revenue is Sumitomo Corporation, at ¥110,600 million (US$675.2 million), 27.0 percent, up from ¥98,193 million (US$599.5 million) and 24.8 percent. Sumitomo Corporation is a trading house. The number therefore describes a sales channel rather than an end customer, and the identity and concentration of the semiconductor makers behind it are not disclosed anywhere in the document. The risk factors say a substantial portion of sales comes from specific major customers, and attach no figure to it. The filing leaves open whether the two are measuring the same thing: the risk factor's major customers and the note's disclosed counterparty need not sit at the same level of granularity.

Geography splits the same way. Sales by customer location put Taiwan first at ¥153,232 million (US$935.5 million), ahead of Japan at ¥79,562 million (US$485.7 million), with China, Europe and elsewhere, South Korea and the United States behind. Property is the mirror image: tangible fixed assets of ¥440,321 million (US$2,688 million) in Japan against ¥209,487 million (US$1,279 million) in Taiwan. The Imari site alone carries ¥342,182 million (US$2,089 million) of book value and 3,296 employees. Overseas revenue is 80.6 percent of the total, from sixteen production sites in four countries. The plant is Japanese. The demand is not.

None of this was inherited. In July 1999 Sumitomo Metal Industries, Mitsubishi Materials and Mitsubishi Materials Silicon jointly established a company for the development and manufacture of 300mm wafers, named Silicon United Manufacturing. Production began in October 2001. In February 2002 the company took over Sumitomo Metal's silicon business and merged with Mitsubishi Materials Silicon, renaming itself Mitsubishi Sumitomo Silicon — that is the order in the Japanese filing; the company's own English history page reverses it. In August 2005 it became SUMCO, and in November listed on the Tokyo Stock Exchange. Komatsu Electronic Metals, now SUMCO TECHXIV, was taken by tender offer in October 2006 and made wholly owned in May 2008. Two industrial groups' wafer arms became one company, at the diameter transition, on purpose.

Which raises the arithmetic problem with calling this a two-company market: from public filings, it cannot be added up. Shin-Etsu Chemical reports an electronics materials segment with revenue of ¥1,015,765 million (US$6,201 million) and segment profit of ¥344,537 million (US$2,103 million) for the year to March 2026. That segment contains semiconductor silicon together with rare-earth magnets, encapsulants, LED packaging materials, photoresist, mask blanks and synthetic quartz, and the filing states that product-level information is omitted because the segment note covers it. Shin-Etsu Handotai, the wafer subsidiary, is wholly owned, unlisted and publishes no separate securities report. One participant is a wafer pure play; the other does not disaggregate. The comparison people make is not one the disclosures support.

The cover of Sumco's annual report carries the line: Silicon Wafers are the Fountainhead of Technology. In the securities report, the forward capital expenditure table reports no applicable items, the segment note reports one segment, and the only counterparty disclosed above ten percent is a trading company. In the February results deck, the AI demand curves carry no data labels; the axis has gridline values and the figures have to be read off the scale. The blank is documented as a blank.

Currency note: yen amounts are converted at ¥163.8 = US$1, the rate on 24 July 2026, applied uniformly to every period cited here regardless of the rate prevailing at the time; the company's own assumptions were ¥155.0 = US$1 for the first quarter of 2026 and ¥160.0 for the second, against a realised first-quarter rate of ¥155.4. Figures the company presents in hundred-million-yen units are converted on the same basis and carry that rounding. Converted figures otherwise carry the significant figures of the yen source. Ratios, percentages, multiples and counts are not converted.