The Company That Counts Everything Except Its Own Parts

Somewhere in the phone in your pocket there are, by the manufacturer's own count, roughly 800 to 1,000 multilayer ceramic capacitors — Murata's figure for smartphones generally, with no split by price tier. Separately, in the same article from March 2021, the company says a high-end, state-of-the-art handset now holds more than 1,000. The two counts cover different populations and are not points on one scale. The company hedged the obvious follow-up question in the same breath: the interviewer, on Murata's own site, offered that the MLCC "may be" the most widely used component in smartphones, and the engineer answering narrowed the claim to a subset, saying only that most of the capacitors in a smartphone are MLCCs. You own about a thousand of a thing whose maker will not quite say it is the most common thing you own.

The arithmetic of very small things is the whole business. In the year to 31 March 2026, the product category called Capacitors produced revenue of ¥936.4 billion (US$5.717 billion), up 12.6%, or 51.1% of consolidated revenue of ¥1.831 trillion (US$11.18 billion). The company does not treat this as a triumph in its risk disclosure. Under "dependencies on specific partners and products," the annual securities report states plainly that capacitors account for 51% of consolidated revenue, and lists that concentration as a risk with a medium frequency and medium impact rating on its own internal scale.

There is a third count, and it is the one the company plans with. A slide deck from November 2024 puts a smartphone at 1,000 to 1,400 capacitors, a server baseboard at 1,800 to 2,500, and an AI-server baseboard carrying eight accelerators at 10,000 to 20,000. A footnote does the work: these are estimates of what a unit requires, not of how many Murata parts are in one. Murata says it makes more than a trillion pieces a year across some 50,000 part numbers — again a slide, not a financial statement — while the production and order tables in the annual securities report are denominated in yen at selling price rather than in pieces, with a note saying so. Volume and price travel through them fused.

Capacitance comes from area and thinness, so the part is built by stacking. The dielectric layers are ceramic sheets the company describes as about one-hundredth the thickness of a hair — the parenthetical in the article gives about 80 micrometres, and the English leaves ambiguous which figure it modifies — and several hundred of those layers go into a single component. What lets the green sheet come out dense and homogeneous is powder: barium titanate particles refined and made uniform in size. The company's technology pages say the grain control runs to the nanometre level, without publishing a number.

Murata's own site draws the obvious conclusion from all this, and draws it in its own favour: unless a firm develops both the materials and the manufacturing processes in house, it cannot lead in miniaturisation and capacitance. The filings put it more carefully, and the gap is worth reading. The English translation of the annual securities report describes a "vertically integrated production system ranging from materials to finished products." The product-marketing page says something adjacent but not identical: a quality management policy that manages every stage from design through material selection, procurement, production, sales and service. Selection is not manufacture.

The count in your phone is not a one-way ratchet, which is the part the headline number obscures. Murata's own explanation runs both ways: as the number of functions in a device increases, the number of installed MLCCs increases proportionally, and in certain electronic circuits the number of MLCCs can be reduced by using higher-capacitance products. Multi-band radios push the count one way — filtering out unused frequency bands requires many small-capacitance parts with low capacitance variation — while somewhere else in the same handset one larger part can do the work of several. Which circuits, the company does not say. Neither the proportionality constant nor the reduction is quantified.

Which is why the growth is no longer coming from handsets. For the year to March 2027 the company assumes smartphones fall to 1.20 billion units from 1.24 billion, a 3% decline, counted on a component-intake basis rather than any third-party shipment statistic; PCs fall 10% on the same basis; autos, counted as production, rise 1%. The stated cause of the phone decline is memory-semiconductor scarcity and pricing hitting mid- and low-end models, with limited effect at the high end. Against that, capacitor revenue rose 12.6% in the year just ended and the company plans roughly 13% growth in the next one.

The destination is data centres, and Murata began disclosing it as a line inside the Computers application only this year. Data-centre-related revenue was ¥101.6 billion (US$620.3 million) in the year to March 2025, ¥176.7 billion (US$1.079 billion) in the year to March 2026, up 73.9%, and is forecast at ¥325.0 billion (US$1.984 billion), up 83.9%. These are company estimates, not audited segment figures, and the definition of what counts as data-centre-related is not published. The capacitor content sits in power delivery: high-voltage stages on the primary and secondary side, and small, high-capacitance, low-voltage parts crowded near the GPU or TPU.

Once, in December 2025, the company did separate quantity from content, for the AI server alone. Against its assumptions of a year earlier it left AI-server unit forecasts unchanged and raised the capacitors per baseboard from 10,000–20,000 to 15,000–25,000, attributing that to more accelerators per board and to capacitors inside the accelerator itself; if board area caps the count, the slide adds, the requirement moves to smaller parts of higher capacitance. The demand curve now exists in two versions that cannot be spliced: 18% a year from a 2023 base, five times by 2030, in the earlier deck; about 30% a year from a 2025 base, 3.3 times by 2030, in the later one. The caveats do not match either. The later figure is stated in units of Murata's own capacitors and labelled a company estimate, and it cannot be turned into revenue, since price and mix are not given; the earlier one is a unit-based forecast of demand, with the estimate label attached to market share and to the per-unit counts rather than to the growth rate itself.

Orders ran ahead of shipments in the January-to-March quarter — a book-to-bill of 1.24 overall and 1.36 for MLCCs — and the capacitor order backlog closed the year at ¥269.2 billion (US$1.643 billion), up 89.5%. The company itself supplies two deflators. Backlog is translated at quarter-end rates, and the yen moved from ¥148.89 to ¥159.93 against the dollar over six months. And the company acknowledged that some fourth-quarter orders reflected a market psychology of procuring early because capacitor price increases were being talked about. On which: no near-term increase is planned, and the drag from price erosion is budgeted to narrow from ¥105.0 billion (US$641.0 million) to ¥73.0 billion (US$445.7 million).

Volume and price come apart in the only document that quantifies the split, every line of which the company labels an estimate. Operating profit for the nine months to December 2025 was ¥203.0 billion (US$1.239 billion) against ¥234.2 billion (US$1.430 billion) a year earlier, and Murata's own bridge assigns the change: rationalisation ¥31.0 billion (US$189 million); higher production volume ¥107.0 billion (US$653.2 million); price reductions negative ¥80.0 billion (US$488 million); currency negative ¥13.0 billion (US$79.4 million); depreciation effectively flat, at negative ¥0.1 billion (US$0.6 million); semi-variable and fixed costs negative ¥25.0 billion (US$153 million); and a residual called other, product mix and so on, negative ¥51.1 billion (US$312 million). Counting that ¥0.1 billion, the lines close to the reported change exactly. On the company's arithmetic, it produced materially more, and price took back three-quarters of what the extra volume earned, 80.0 against 107.0.

Mix, where a move into data-centre parts would show up, sits in that residual, and the residual has other work to do. In a different bridge — from the September quarter to the December quarter, not the nine-month year-on-year comparison above — the same line is negative ¥46.9 billion (US$286 million), while the December quarter carried ¥49.8 billion (US$304 million) of one-time charges: goodwill and fixed-asset impairments, environmental costs, none of which the bridge gives a line of its own. The residual is smaller than those charges, and the deck does not break the line out. Where the charges sit is not disclosed; the mix effect inside the residual cannot be sized. Then the March quarter took the year back: operating profit ended at ¥281.8 billion (US$1.721 billion), up 0.8% from ¥279.7 billion (US$1.708 billion), so a company ¥31.2 billion (US$190 million) behind through December finished ¥2.1 billion (US$13 million) ahead.

The other half of the company had a different year. Devices and Modules revenue fell 5.9% to ¥656.0 billion (US$4.005 billion) and the segment swung from an operating profit of ¥10.0 billion (US$61.1 million) to an operating loss of ¥26.5 billion (US$162 million). Impairment losses of ¥48.9 billion (US$299 million) sit inside that, of which ¥43.8 billion (US$267 million) was goodwill written down in the surface-acoustic-wave filter business; the stated reason is that the shift to higher frequencies in communications markets has been slower than originally assumed. Pre-tax segment ROIC, on the company's own definition, was 22.4% for Components and negative 3.5% for Devices and Modules.

Research spending is allocated in the opposite direction from returns, which is either a portfolio strategy or a description of one. Of ¥158.9 billion (US$970.1 million) in consolidated R&D for the year, ¥93.0 billion (US$568 million) went to Devices and Modules and ¥57.1 billion (US$349 million) to Components. The report's risk section puts total R&D at 8 to 9% of revenue. The segment that earns a 26.8% operating margin, on ¥1,175.2 billion (US$7.175 billion) of segment revenue and on products the report exemplifies as multilayer ceramic capacitors and ferrite beads, receives rather less than the loss-making one on ¥656.0 billion (US$4.005 billion).

Capacity is the constraint being bought out of. Capital expenditure was ¥247.8 billion (US$1.513 billion) in the year to March 2026 against a plan of ¥250.0 billion (US$1.526 billion) for the year ahead, and on top of the medium-term envelope the company placed an additional order budget of about ¥80.0 billion (US$488 million) for server-bound MLCC capacity, roughly ¥40.0 billion (US$244 million) of it landing in the current year, stating it would exceed the ¥680.0 billion (US$4.151 billion) three-year plan if required. It does not name the sites for that additional budget. Separately, the annual securities report lists next-year facility plans of ¥58.0 billion (US$354 million) at Izumo, ¥32.0 billion (US$195 million) at Fukui and ¥20.0 billion (US$122 million) in Thailand. Capacitor utilisation in the most recent quarter is described as close to 95%, against a 90 to 95% assumption for the year ahead, and the deliberate inventory build of about ¥25.0 billion (US$153 million) is company-wide rather than capacitors alone.

The three-year envelope is arithmetically closed: operating cash flow of ¥1.300 trillion (US$7.937 billion) in; capital expenditure of ¥680.0 billion (US$4.151 billion), strategic investment ¥220.0 billion (US$1.343 billion) and shareholder returns ¥400.0 billion (US$2.442 billion) out, summing to the inflow exactly. For the year to March 2026, dividends plus buybacks came to 93.4% on the formula the convocation notice prints — total dividends plus total treasury purchases, over net income — a ratio the notice presents as conditional on the dividend resolution passing, and whose denominator it does not tie to a named line of the accounts. On 30 April 2026 the board authorised buying up to 75,000,000 shares, 4.12% of those outstanding excluding treasury, for up to ¥150.0 billion (US$915.8 million), and resolved to cancel them, with cancellation scheduled for 26 February 2027. The 142,709,692 treasury shares already held are not part of that: they stay through the plan period, partly for strategic investment. The reason given is one line: to improve capital efficiency.

Where the revenue is booked and where the company is built are not the same place. Greater China accounted for ¥865.0 billion (US$5.281 billion), or 47.2% of revenue, Japan ¥132.8 billion (US$810.7 million), or 7.3%, with 92.7% booked outside Japan — all on a customer-location basis, meaning the country where the part is delivered for assembly rather than where the finished device is eventually used. Non-current assets run the other way: ¥901.3 billion (US$5.503 billion) of ¥1,506.5 billion (US$9.197 billion) sits in Japan. Group headcount is 74,302, of which 42,813 is Components.

Set against its own targets, the year reads differently. The plan announced in November 2024 asks for ¥2 trillion (US$12 billion) of revenue, a margin of at least 18% and a post-tax ROIC of at least 12%, by March 2028. The year just reported delivered ¥1.831 trillion (US$11.18 billion), 15.4% and 9.7%, against 16.0% and 10.0% the year before: revenue up, both ratios down. The same ¥2 trillion was the previous plan's target for the year to March 2025, which came in at ¥1.743 trillion (US$10.64 billion) against a 20% margin goal. The number did not move. The year attached to it did. ROIC changed definition at the same time, pre-tax to post-tax, so that for the year to March 2025 both are printed, 13.0% and 10.0%; the eleven-year summary table carries only the pre-tax series. The post-tax figure is half the weight in the executives' performance-share award.

The clocks do not match, either. Murata puts the MLCC generation cycle at seven to ten years, against two to three for semiconductors, with three to five years needed to develop a technology. An executive, Omori, told analysts in April that 2026 and 2027 had come into view for AI data-centre demand and that 2028 and beyond remained unclear. The part is designed on a horizon of a decade; the demand for it is legible for about two years; and the count of them in your pocket, the number the whole thing rests on, comes from a 2021 article the company itself flags may differ from the latest information.

On currency: yen figures here are converted at ¥163.8 = US$1, the rate on 24 July 2026, applied uniformly to every period mentioned, including figures reported in years when the yen traded somewhere else entirely. Conversions are computed from the reported figures in millions of yen and carry the significant digits of the yen figure, four at most. Murata translates on its own basis and for its own purposes — an average of ¥150.78 to the dollar for the year to March 2026 — and that rate appears here as one of the company's disclosures, not as the conversion method. Percentages, ratios, multiples, share counts and headcounts are not converted.