Variant Perception

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

Variant Perception — Where We Disagree With the Market

At $41.83 the market has absorbed roughly three-quarters of the bull sum-of-the-parts target while the only published sell-side mark sits at $29.62 — a +150% twelve-month rally that the analyst community has not refreshed. That gap is the consensus signal: the price is doing the underwriting, and it is implicitly trusting the FY2028 $725M / $120M mid-term plan at face value. Where we disagree is narrower and more decision-useful than the bull/bear arguments already in the deck. The rally embeds an unearned adjacency premium, the prime segment's borrowed regulatory ringfence is being priced as a delay when it is actually a structural ratchet, and the ordinary-income headline that Japanese retail anchors on is flattered by $6–10M of non-recurring items that the multiple does not reflect. Each is observable in segment disclosures within the next 3–9 months.

Variant Perception Scorecard

A 64 score reflects two things. First, the strongest of our three views (the adjacency-credibility discount) is materially different from the observed market price but only partly different from the report's bear case; the variant lives in pricing, not in raw analysis. Second, consensus clarity is moderate because only one named analyst publishes a live target (Minkabu/IFIS $29.62) and the market price has decoupled from it — the rally itself is the cleanest revealed consensus. Evidence strength is high: every claim below traces to a disclosed segment table, a transcript quote, or an externally documented policy event.

Consensus Map

What the market appears to believe, where we can see it, and the assumption embedded in each view.

The most important row is the first one. Coverage is structurally thin — Simply Wall St counts four analysts (Daiwa, Ichiyoshi, Okasan, plus one), only one of which publishes a live target — so the price has become the consensus signal. At $41.83 against a Bull SOTP of $59.85 and a Bear target of $26.46, the market has priced ~80% of the bull-case path. That is where the variant view earns its keep.

The Disagreement Ledger

Three ranked disagreements. Each names a specific market assumption, the evidence that breaks it, and the metric that resolves it.

Disagreement #1 — the adjacency premium is unearned. Consensus would say "management has guided FY28 $725M / $120M, the reclaim engine is broadening, treat the plan as base case." Our evidence is the guidance scorecard: the same management retired the Feb 2024 Upside Plan ($867M FY26 → silently replaced with $529M FY26), the LE System $159M FY26 target (FY25 actual under $6M), and the SGRS 300K/m by 2027 milestone — all without a formal acknowledgement that they had been withdrawn. The right base case applies a 30–40% credibility discount to the non-reclaim portion of FY28 operating income. If we are right, the market would have to concede that $32–44M of FY28 plan EBIT is uncertain and the appropriate consolidated multiple is 6.5–7.5x EV/EBITDA on the un-discounted plan rather than 9x. The cleanest disconfirming signal is a Feb 2027 plan refresh that reaffirms FY28 numbers AND shows LE System above $19M run-rate AND announces an external OEM design win for the RSPDH Jiangxi Shinetech camera pivot.

Disagreement #2 — SGRS is a ratchet, not a delay. Consensus reads "150K → 50K wafers/month is a timing slip; the option value is preserved." Our reading is that this is the first observable signal that mature-node BIS controls will eventually capture wafer flows — every prior tightening cycle (Oct 2022 / Oct 2023 / Dec 2024) shows the same one-directional creep, and management's "Japan-China relations" framing is the gloss on a policy wedge that has only ever closed. The Inner Mongolia plant disclosure (19 Mar 2026) doubles down on Chinese geographic concentration at the precise moment the policy backdrop deteriorates. If we are right, the prime segment terminal value collapses from the Bull frame's ~$170M toward ~$63–95M, and the remaining $139–277M of GRITEK capex compounds the destruction. The cleanest refuting signal is GRITEK STAR Market (688521.SH) showing 8-inch ASP YoY narrowing below −5% and subsidy income normalizing around $9.5M annualized.

Disagreement #3 — the ordinary-income anchor is partially synthetic. The Japanese investor base anchors on ordinary income (経常利益), framed by management as "record high for five consecutive years." That headline at FY25 $105.9M embeds $13.4M of Chinese subsidies (which the CEO labelled partly a past-period catch-up) and a residual $2.6M negative-goodwill flatter on net income. Normalize, and the recurring run-rate is ~$96–99M — a 4–8% haircut. On its own this disagreement does not move the equity by more than ~$3/share, but it tightens the asymmetry created by disagreement #1: the FY28 $120M op-income target sits on top of an FY25 base that is itself slightly flattered.

Evidence That Changes the Odds

The seven items below are not generic facts — each one moves the probability of the variant view in a direction the consensus market price has not yet absorbed.

How This Gets Resolved

Six observable signals. Each has a current state, a print that validates the variant view, a print that refutes it, and a timing window. Resolution is a multi-quarter trajectory, not a single earnings event.

The signals are weighted unevenly. Signal #1 (Feb 2027 plan refresh) carries the bulk of the resolution because it directly tests management's adjacency-delivery track record over a fresh cycle. Signal #2 (reclaim margin trajectory) is the closest thing to a binary near-term test but only updates the reclaim leg, not the consolidated underwriting. Signals #3 and #6 (SGRS + BIS) move slowly but in one direction; #4 (buyback) tests the capital-allocation half of the alignment story; #5 (coverage refresh) is the mechanism by which a variant view becomes consensus.

What Would Make Us Wrong

The most honest red team is to take each of the three disagreements and name the evidence that would force us to retract it before the market does.

On the adjacency premium (#1). We are wrong if the Feb 2027 plan refresh is the first clean reaffirmation in three years — FY28 $725M / $120M held at face value, LE System cleanly above a $19M annual run-rate (it has spent FY25 at under $6M), and RSPDH camera modules either showing an external OEM design win or pivoting cleanly toward a recurring automotive revenue base. The Feb 2026 Sumitomo Electric LE System selection is one data point in the right direction; one more credible external commercial validation would weaken the bearish overlay materially. We are also wrong if the second mid-term plan in the 5-to-10-year window comes in matching the Feb 2026 plan within ±5% on operating income — that would convert "three retired plans" into "three plans then a stable plan," and the discount we are applying becomes recency bias.

On the SGRS ratchet (#2). We are wrong if BIS does not issue a new mature-node wafer rule within the next 12 months and GRITEK's STAR Market filings show 8-inch ASP YoY narrowing below −5%, and a first non-Chinese fab qualification for SGRS materializes (e.g., a Japan-headquartered foundry / IDM technology-transfer disclosure). The structural-ratchet read fails fastest if the policy backdrop holds and the operational read recovers — both have to break the same way for it to be merely a delay rather than a permanent reset. We are also wrong if SGRS proves capable of serving the Chinese mature-node demand pool at scale without non-Chinese qualifications — i.e., the addressable market is large enough that the "borrowed ringfence" is itself a durable franchise even with policy uncertainty.

On the ordinary-income flatter (#3). We are wrong if FY26 Gritek subsidy income comes in at or above $12.6M — that would invalidate the CEO's own "past-period catch-up" framing and mean the recurring subsidy stream is structurally larger than disclosed, which actually raises normalized ordinary income rather than lowering it. We are also wrong if the next M&A footnote shows no bargain-purchase gain — that would break the pattern we are pricing in. This is the lowest-conviction of the three views and the easiest to retract; it survives only as a tightening overlay on disagreement #1, not on its own.

The first thing to watch is the Q2 FY12/26 segment table on 7 August 2026 — not because it resolves the variant view (it does not), but because a 37%+ reclaim margin print under live Sanbongi 7 ramp dilution confirms the asymmetric trust split (reclaim delivers, adjacencies do not) that underpins disagreement #1 and tightens the bear-case asymmetry on the rest of the equity.