Business

Figures converted from JPY at historical FX rates — see data/company.json.fx_rates. Ratios, margins, and multiples are unitless and unchanged.

Know the Business

Disco is a niche monopolist that owns three physical steps of semiconductor manufacturing — cut, grind, polish — with ~70% global share in dicing and back-grinding, and a razor-blade consumables stream attached to a growing installed base. FY2026 revenue was $2.74B at a 42.3% operating margin and 25.1% ROE; the four-year cumulative ordinary margin of 41.4% is the tenth consecutive year above management's 20% floor, which is the cleanest single proof of franchise durability. The market is paying ~51x earnings — comparable to ASML and KLAC — so the upside is no longer about discovery; it is about whether the AI/HBM grinding cycle and the consumables base can keep ordinary margins above 30% through the next downturn.

1. How This Business Actually Works

Disco sells one of the most valuable things in modern manufacturing: a physical step inside a chip fab that almost no one else can do as well. Every silicon wafer that becomes an AI accelerator, an HBM stack, a power-control chip, or a smartphone SoC must be thinned (back-grinding, from ~700µm to as little as 30µm), separated into individual die (dicing), and often polished to a mirror finish. Disco is the global leader in each of these three steps, holding an estimated 70–80% share on advanced nodes, and selling under a brand philosophy it calls Kiru, Kezuru, Migaku — Cut, Grind, Polish.

The economic engine sits on three legs. The tools themselves — dicing saws (32% of FY2026 sales), grinders and polishers (27%), plus accessory equipment (4%) — are sold as $250K–$950K boxes to roughly thirty fabs and OSATs globally; this is the lumpy, cyclical line. The recurring leg is consumables (22% of sales) — diamond-tipped blades and grinding wheels that wear out cut-by-cut on a customer's wafer volume, not capex. Service and parts (10%) sits on the installed base. The consumables stream is the analog of a razor-blade business: each new tool placed in a fab dedicates a recurring blade order for as long as the tool runs. That mix is why corporate gross margin sits at 70%+ and why operating profitability has stayed positive in every cycle since FY2002.

The moat is application-engineering depth, not patents. A leading-edge dicing process is co-developed with the customer over months in Disco's Tokyo demonstration center — by the time a tool is qualified for an HBM thinning line, ripping it out for an Accretech alternative would cost the customer quarters of re-qualification on a part of the wafer where a 50-basis-point yield slip dwarfs the equipment savings. That switching cost is what lets Disco take 70%+ share in a market where the second-place player exists, is competent, and would happily take the business.

2. The Playing Field

Disco is a small fish in the front-end semicap aquarium and the biggest fish in the back-end pond. The relevant comparison is not "biggest semi-cap" — it is "highest-quality, narrowest moat, most capital-efficient." On the metrics that decide franchise quality — gross margin, operating margin, ROE on a clean balance sheet — Disco is in the top quartile of global semicap, even before any AI-cycle adjustment.

Latest annual data; market caps as of 2026-05-18. Non-USD peers converted at spot for like-for-like scale. Accretech and TEL figures are approximate from IR-site disclosure; KLAC ROE is inflated by ongoing buybacks shrinking the equity base.

Three things the peer set reveals. First, Disco prints higher gross margin than every front-end peer including ASML, despite competing in a smaller TAM — that is the consumables tax compounding inside the mix. Second, Disco's ROE is mid-pack at 25% only because its balance sheet is 79% equity-funded with $1.78B of cash and no debt; ROIC on operating capital is far higher, and KLAC/LRCX's ROEs are inflated by buyback-driven equity shrinkage. Third, the only true direct competitor is Accretech (7729.T), and the gap is brutal — same end-customers, same Japan cost base, ~38% gross margin vs. Disco's 70%, ~22% SPE-segment operating margin (~20% full-company) vs. 42%. That ~20-point operating-margin gap on identical inputs is the cleanest empirical proof of Disco's pricing power.

The right peer is KLAC. Both companies are narrow-niche process specialists with 60%+ gross margins and 35%+ operating margins, both trade at ~50–60x earnings, both compound at 20%+ ROE without leverage. The wrong peers are AMAT and TEL — those are broad-portfolio capital allocators with mid-40% gross margins; treating Disco as "another semi-cap" misses the entire reason it earns a premium.

3. Is This Business Cyclical?

Yes, but the cycle hits the equipment line, not the margin floor. The historical record shows two violent revenue drawdowns — FY2002 (-59% from the dot-com bust) and FY2009 (-42% in the GFC) — and several milder ones (FY2012, FY2019, FY2020). In every one of them except FY2002, operating margin stayed positive, because consumables and parts kept running while the equipment line collapsed. The chart below is the same revenue history overlaid with operating margin — it is the picture of a cyclical business with a structural margin floor.

The structural read is straightforward: consumables have grown from 13% of revenue in FY2001 to 22% in FY2026, and parts/service from 8% to 10%. A third of the business is now non-capex-linked. That is why FY2019 had a 12% revenue drop and 26% operating margin — barely a scratch — and why a future downturn is unlikely to drag operating margin below ~25% unless the equipment line falls 40%+ and consumables decline in unison (i.e., a multi-quarter wafer-volume contraction, not just a capex pause). The risk that is worth pricing is the next time both legs fall together — a deep, prolonged wafer-volume recession of the kind only FY2002 produced. The market currently prices nothing close to that.

4. The Metrics That Actually Matter

Five numbers explain most of the value Disco creates or destroys. The first three are leading indicators; the last two are the franchise verdicts.

5. What Is This Business Worth?

Disco is a single economic engine — not a sum of parts. There is one business model (Cut, Grind, Polish + consumables), one customer set (~30 fabs and OSATs), one production base (three Japan plants), and no material listed subsidiaries or non-core holdings. The right valuation lens is normalized-cycle earnings power applied to a franchise that compounds capital at ~25% ROE without leverage, anchored to where comparable narrow-niche specialists (ASML, KLAC) trade through a cycle. Forced SOTP would mislead — the consumables base, equipment business, and service stream all reinforce each other and would be worth less apart than together.

The reader should leave knowing that the cheap-or-expensive question reduces to two judgments. First, does the AI/HBM wafer-volume thesis still have runway, or is FY2026 the peak quarter? If shipments grow ~15-20% from here into FY2027–28, current multiples are reasonable; if they roll over, ~51x earnings is rich. Second, is the consumables base big enough to keep operating margin above 25% if the equipment line halves? The FY2019 experience says yes. FY2002 says no, but the consumables share then was less than half of today's.

6. What I'd Tell a Young Analyst

Three things. First, follow the shipment line and the four-year cumulative ordinary margin, not the quarterly revenue print. Shipments lead revenue by one to two quarters, and the four-year cumulative margin is the only number management actually targets — it filters cycle noise that everyone else gets confused by. Q1 FY27 shipment guide of $833M (+18.8% YoY) and 41.4% trailing four-year ordinary margin are the two cleanest signals in the deck.

Second, the moat is not patents, it is application engineering and switching costs. When a buy-side analyst asks "why can't a Chinese competitor or AMAT copy this," the answer is not "patents protect Disco" — it is that every leading-edge process recipe is co-developed between Disco engineers and a specific fab over months in Tokyo, and replacing the tool means re-qualifying a yield-sensitive step. Lose this thread and the stock looks overpriced.

Third, the thing that would change the thesis is not a bad quarter — it is a structural break in consumables. The cycle floor depends on consumables and parts being a third of the business and tied to wafer volumes, not capex. The risks worth tracking are (a) a credible Chinese consumables substitute on trailing nodes, (b) US export controls widening from front-end to back-end equipment or to consumables in China, and (c) hybrid-bonding architectures that change how much wafer thinning happens per package. Headline risks like the SiC/EV slump or a quarter of weak smartphone capex are noise; the things on this list are the things that would actually break the model.