TSUZUKI DENKI Co., Ltd. (8157): A Good Business at a Fair Price, Owned Tightly

Verdicts

Lens	Verdict	Buy below	Most load-bearing items
Buffett	watch	¥2,650	B23/B25/B101 — is the ~42% Fujitsu-box leg D6 plus service switching-costs a moat independent of Fujitsu E37E68F483? not clearly, so 1.5× book D3 pays ¥37bn of goodwill for a franchise part-owned by the supplier; B42/B60 ~7.7% owner yield on 14%-ROE net-cash capital F51D1 — good, but fair not cheap; B89/B93 no margin of safety near the price
Munger	pass	null	M37 great-at-fair-not-fair-at-great — hardware 42% D6 is intermediary economics, priced correctly as a good business at 1.51× book D3 / ~11.2× F40; M18 no nameable moat mechanism, 24.1% gross margin is integrator-thin F360F356; M44 pari-mutuel — the capital-return upgrade E87 and backlog F212 are already priced after the run from ¥1,215 F174
Pabrai	watch	¥2,400	P53 fifty-cent dollar fails — ¥3,995 is ~78–85% of conservative IV ¥4,700–5,200/sh D1F10E30; P1 max loss — ¥35,072M net-cash fortress D1 makes the downside a de-rating not a wipe-out (why watch, not pass); P38/P48/P12 — the ¥40bn M&A war-chest F236E28 spends the net-cash floor, Fujitsu is supplier-and-shareholder F483E139
Li Lu	too-hard	null	L1 ten-year knowledge bar — two decisive variables (Fujitsu's channel strategy E37E139F483; a ¥40bn M&A program not yet begun F236E28) are unknowable to a minority; L5/L18 no cushion — 1.51× book / ~11.2× D3F40 has no discount to absorb them; L26/L30 Aso covenant + Fujitsu triple-role entrench control E69E139
Claude	watch	¥3,526 (implied ¥3,799)	C34/C35 reverse-DCF prices −1.6% perpetual growth → cheap-ish, not "already priced" F10F76; C33/C89 bear case implies a buy-below below the stamp — no margin of safety at ¥3,995; C84/C40/C87 the record year is a hardware-cyclical peak with a stalling service leg F206F203; C102/C98 the net-cash "floor" is an announced M&A war-chest E28

No lens buys at the ¥3,995 stamp, but three name real buy-belows below it — the first study where the buy-side is genuinely engaged rather than declining a bad business.

The business

TSUZUKI DENKI (都築電気) sells Japanese companies the plumbing of their computer networks. Founded in 1932 to sell, install, and maintain telephone-exchange equipment E3, it grew into an electronics-trading house and then, over decades, into what it is today: an ICT systems-integrator — an "information network solution service" firm that proposes, builds, operates, and maintains the networks that run a corporate customer's business E10. The electronics-trading roots are still visible in the shape of the company, but the parts business itself is gone: the electronic-devices business was spun off to Restar (レスター) by company split in 2021 E7, and the residual Restar-related shares were transferred out in early 2024 E9. What is left is a focused integrator, ~90% of whose revenue is earned by the Japanese parent F86.

The group reports as a single reportable segment — 情報ネットワークソリューションサービス, with segment disclosure omitted because there is only one E48E143F444 — but it discloses the mix under revenue recognition as three business models E11:

• Hardware (機器, ~42% D6) — reselling information and communication equipment, mostly other vendors' servers and storage; revenue ¥43,378M F438, recognized at delivery or customer acceptance E140.
• Dev-Build (開発・構築, ~17% D7) — consulting, design, development, and construction; the technical, project-based leg; revenue ¥17,391M F440, recognized over time on a percentage-of-completion basis E141.
• Service (サービス, ~41% D8) — operation and maintenance of equipment and software, plus monthly cloud and managed services; revenue ¥42,957M F442, the recurring, "stock-type" leg recognized over the service period as the calendar passes E142E51.

Two relationships govern the shape of the whole. Fujitsu is the dominant supplier: the group buys Fujitsu Group products (equipment, program-products, maintenance, service, consulting) under a sales-partner agreement that auto-renews one year at a time E68, and it is also the second-largest shareholder and a related party (below). Aso Corporation (麻生) is the largest shareholder and a capital-and-business-alliance partner, with a medical/nursing-care vertical the two companies pursue together — the "AI discharge-date prediction service" is one product of it E15E70. The customer base spans manufacturing, government/public sector, finance, services, transport, healthcare and logistics E49E50E51; the tailwind, on the company's own telling, is a medium-to-long-term rise in corporate IT-investment demand driven by labor shortages, cybersecurity risk, and generative AI E18.

The numbers

FY2026/3 was the fourth straight record on the operating and ordinary lines E47. Revenue was ¥103,728M (+5.6% F191F5); operating profit ¥8,178M, up +26.2% F364F192; ordinary profit ¥8,320M (+26.1%, a record) F10; net income attributable to owners ¥6,472M, up +35.9% (a record) F386F194. ROE was 14.0% F51, up from 11.3% and averaging into the teens over five years [F47–F51]; the equity ratio was 55.1% F46. Orders grew +9.7% to ¥110,384M F204, and the order backlog grew +32.9% to ¥26,902M F212F213 — but the growth is concentrated in the cyclical leg: Hardware backlog rose +62.7% F206F207 on large public-sector and financial server/storage wins E49, which flags the hardware-cyclical character of the record year rather than a broad-based acceleration.

The balance sheet is a fortress. Against ¥43,467M of cash and deposits F244 sits only ¥8,395M of interest-bearing debt F481, leaving net cash of ≈¥35,072M D1 — about 48% of the ¥72,763M market cap D4. BPS is ¥2,652.76 F35, so the ¥3,995 stamp is 1.51× book D3; PER is ~11.2× (¥3,995 ÷ EPS ¥355.94 F40). Owner earnings are backed by real cash: operating cash flow was ¥6,322M F414 against a capital-light asset base (tangible fixed assets just ¥1,814M F268) and modest capex of ¥711M including intangibles F429. Two one-offs sit inside the record: a ¥2,419M gain on the sale of investment securities F372F476 and ¥1,349M of extraordinary charges — a ¥956M ERP core-system rebuild loss F375 and a ¥367M software impairment F376F434 — so management publishes a more conservative "real ROE" of 12.3% ex-extraordinary E30F242.

Capital allocation is turning genuinely shareholder-friendly. The company cancelled 1,200k treasury shares in Feb 2025 (cutting issued shares 20,177k → 18,977,894 F494F102); it sold every one of its three megabank cross-holdings this year — MUFG ¥1,916M, SMFG ¥263M, Mizuho ¥215M, all fully out E134E135E136 — leaving residual listed cross-holdings of just ¥1,845M / 9 issues E131; and it upgraded its dividend policy from a 40% payout / 3.5% DOE floor to 60% payout / 6.0% DOE floor from FY2027/3, on a business-profit (ex-extraordinary) base E87. The current dividend is DPS ¥126 at a 37.1% payout F122F152. The mid-term plan "Trust & Challenge 2029" targets ¥7bn of growth investment plus ¥40bn of strategic/M&A investment E27E28F236, and an ROE of 14.5% E30 — with management candid that the prior plan improved profitability but "could not invest sufficiently in growth" E22.

Governance is the wrinkle. The two largest holders together are ~49% allegiant. Aso holds 23.97% E80 under a voting-maintenance covenant — the company agrees that if it takes any action that could lower Aso's voting ratio, it will maintain that ratio — plus one Aso director E69. Fujitsu holds 12.80% E81 and is simultaneously the dominant supplier: ¥18,965M of equipment procurement this year (accounts payable ¥4,502M), which is 61% of the group's ¥31,094M procurement F483F484F216, transacted "per the Fujitsu Group Sales-Partner Agreement" E139 — a related party with a board seat E109E112. Below the two blocks: Fuso Dentsu 4.08% F507, an employee shareholding association 3.02%, and BIP/ESOP trusts ~3.0% E82E83; foreign holders are just 6.7% E79. The board is 8 of 10 outside E92 (moving to 7 outside post-AGM E93), with a voluntary nomination-&-compensation committee E94; performance pay is tied to operating profit E125. The auditor is Grant Thornton Taiyo (太陽), a clean opinion over a 63-year continuous tenure E119E120, with a single Key Audit Matter — the reasonableness of the dev-build total-cost (POC) estimate E148.

The five lenses

Buffett — watch, buy below ¥2,650

I start where I always start — with the store, not the stock. Tsuzuki sells Japanese companies the plumbing of their computer networks: it resells hardware F438, designs and builds the systems F440, and keeps them running afterward on monthly service contracts F442E142. Roughly nine dollars in ten come from the parent in Japan F86. A shopkeeper would follow that. So we clear the first gate — this is inside the circle, a value-added distributor and jobbing shop for corporate IT E10E11.

Now, is it a franchise — a toll bridge — or a good hardware store on a busy corner? That's the whole question, and I'll be honest it's a close call. The switching costs are real on the service side: once you run a customer's network and hold their maintenance contract, you don't get thrown out lightly, and management has been pushing exactly that recurring, stock-type revenue E51E25. But two-fifths of the business is moving boxes, and here's the fact that hangs over everything — the boxes are Fujitsu's. Tsuzuki bought ¥18,965M from Fujitsu this year F483, a quarter of all cost of sales F358, under a sales-partner contract that renews one year at a time E68. And Fujitsu owns 12.8% and sits on the board F485E109. When your largest supplier is also your second-largest owner and a competitor's partner, you have to ask whose franchise this is. Management lists the dependence as a named risk E37 and is trying to build proprietary services to reduce it E38 — candor I respect. But a moat you're actively digging because the one you have belongs partly to your supplier is not one I'd bet the ranch on [B23][B25].

The returns, though, are genuinely good and don't lean on borrowed money — the test that matters most [B60]. Return on equity was 14.0% F51, earned with net cash of ¥35bn D1 and a 55% equity ratio F46. Operating profit rose 26% to ¥8,178M F364, the fourth straight record E47. But I never pay the sticker price for good news, so let me compute what an owner gets. Strip the ¥2,419M securities gain F372 and the ¥1,349M of charges F377; management's own honest figure is a "real ROE" of 12.3% E30. Tax the operating profit fully at 30.6% F465, add back the ¥1,251M of depreciation F401, subtract the almost-nothing this capital-light business needs to stand still (tangible assets ¥1,814M on ¥103,728M of revenue F268F356) — owner earnings land around ¥5.5–5.7bn. Against the ¥72.76bn cap D4 that's an owner-earnings yield of about 7.7% [B42], with ¥35bn of net cash inside the price. Against a Japanese long bond that's fair — the honest word. This is a wonderful-ish business at a fair price, not a fair business at a wonderful price [B102], and no Graham bargain: at ~2× net current assets [F260−F328] the cheapness does not hit me over the head with a baseball bat [B89][B98].

Two things earn it a hard look rather than a pass: the capital-return inflection is the genuine article — a share cancellation E76, all three megabank cross-holdings sold [E134–E136], and a step to 60% payout from FY2027 E87 — and the ¥35bn cash pile starting to come back changes the arithmetic in the owner's favor. But at 1.5× book I'm paying ~¥37bn of economic goodwill D4 for a franchise I can't fully attribute to the company. So: an understandable, decent, well-financed business, honestly run, at a fair-not-cheap price, with a moat I can't fully vouch for. That's a watch, not a buy at ¥3,995. I'd want it around book — roughly ¥2,650 F35 — where the owner-earnings yield pushes toward double digits and the Fujitsu question becomes an option rather than a wager.

What a student should take from this: when a company's largest supplier is also a major shareholder and a competitor's ally, its returns may not be its own — before you call recurring revenue a "moat," ask who would still be standing if that one relationship changed. And notice the trap of the record year: a fourth straight record and a 14% ROE F51 make it easy to pay 1.5× book D3, but the discipline is to compute the owner-earnings yield after stripping one-time gains and to let the purchase price — not the story — decide.

Munger — pass

Invert first. How does this become a disaster? The cleanest obituary: Fujitsu restructures its channel, absorbs distribution in-house, or reweights toward a rival integrator. Tsuzuki, whose procurement from Fujitsu runs ¥18,965M — roughly 61% of its ¥31,094M total F483F216 — gets squeezed on margin and volume at once. Fujitsu also holds 12.80% F505 and sits on the board E109, and the sales-partner agreement auto-renews yearly E68, so neither party has made a long-term commitment on paper. That's not speculation — the filing names it, medium impact, medium likelihood E37, and the remediation reads vaguely (expand proprietary AI/IoT/cloud services E38). Admirable aspiration; not yet a moat. The second kill path is commoditization: the company itself admits generative AI and in-sourcing are depressing the value of "simple dev-build" E19, and R&D of ¥173M F427 on a ¥103bn base is a rounding error. The third is M&A self-destruction: a ¥40bn "strategic investment" program F236E28 against ¥35bn of net cash D1, funded by cash plus new debt E62, run by management that self-assessed as having invested "insufficiently" in growth E22. At 1.5× book they will pay more than book for whatever they buy; two or three deals gone sideways and the cushion dissolves.

Now, what is the moat mechanism? I keep looking. Service revenue at ¥42,957M — 41.4% of the total F199D8 — has the best moat-candidate qualities: recurring, cloud-and-maintenance-heavy E51, 17.9-year average employee tenure E146, a 25.9% backlog-to-revenue ratio D11. But switching costs in enterprise ICT maintenance are modest once contracts roll; the industries served re-tender periodically; and the ledger discloses no customer concentration, no churn, no retention statistics. What I can see: operating margin 7.9% D2 — decent for a Japanese ICT-SI, not the 20%+ that signals real pricing power — and a gross margin of 24.1% (¥24,965M / ¥103,728M F360F356), the margin of an intermediary, not a franchise. The mechanism — "long relationships + multi-vendor integration + engineering talent in verticals" — is real but weak, and competitors can replicate it [M18].

ROE of 14.0% F51 is the headline, but management is transparent it owes something to the megabank cross-holding sales booked as ¥2,419M of extraordinary gains F372; their own "real ROE" is 12.3% E30. A reasonable number for a well-run integrator; not one that earns a franchise premium. On the napkin: recurring after-tax operating earnings ≈¥5.7bn F364, market cap ¥72.76bn D4, back out ¥35bn net cash D1 — EV ≈¥37.7bn, ~6.6× EV/EBIT. Not expensive. But the pari-mutuel question is whether this bet is mispriced, and it isn't: the stock has run from ¥1,215 to ¥3,995 F174F172, TSR 234.7% vs TOPIX 202.2% F162F167, and the capital-return upgrade to 60%/6.0% DOE E87 is announced and therefore priced [M44]. The opportunity-cost test decides it: a Japanese franchise with a genuine proprietary-platform moat trades at a similar multiple, and beats this on quality. The business is good, the balance sheet fortress-clean, management moving the right way — but "good business, moving the right way" earns a pass and a revisit, not a buy. Great-at-a-fair-price, not fair-at-a-great-price [M37] — my least-favorite quadrant.

What a student should take from this: name the moat mechanism or admit it doesn't exist — "good relationships in a competitive industry" is not a moat, it is a description of average. Where your largest supplier is also your second-largest shareholder, you have a structural conflict that board composition alone cannot resolve: Fujitsu's incentive is to maximize Fujitsu's margin, not Tsuzuki's. And a ¥35bn net-cash pile D1 is a gift if allocated brilliantly and a slow-motion wound if they pay 2× book for mediocre acquisitions — "transformational M&A" from management that self-assessed as "insufficient" on prior growth investment deserves heavy skepticism.

Pabrai — watch, buy below ¥2,400

I start where I always start: with the downside, not the story. If I buy at ¥3,995, what's the realistic worst case, and how much of my money does it eat? Honestly, this isn't the setup I usually hunt. My territory is a low-risk, high-uncertainty business priced as if it were about to die — a burning theater everyone's running out of. Tsuzuki is the opposite: nobody is afraid. It's the fourth straight record year F5F364F386, the stock has more than doubled its total return in three years F162F167, and it trades near its all-time high of ¥4,200 F172. The drawdown that usually creates my bargain isn't here [P73][P74].

Now the floor, honestly built [P1]. This is a fortress. Net cash is ¥35,072M — cash ¥43,467M against ¥8,395M of debt D1F244F481 — ¥1,926 per share, roughly 48% of the price, under a business earning ~¥5.7bn after tax even stripping the one-off ¥2,419M securities gain F372. Equity ratio 55.1% F46, no going-concern note, auditor in place 63 years F539. The leverage question that kills most of my mistakes simply doesn't apply [P20][P26]. When I mark this down, "tails" doesn't mean ruin — it means a de-rating: normalize earnings down ~20% and compress the multiple to its own recent low near 7× F54, and I get roughly ¥36–40bn against today's ¥72.8bn D4, a −40% to −45% branch, cushioned but real. Survivable, not a permanent-loss trap.

But surviving isn't a bargain, and here the checklist bites. Run the fifty-cent dollar with a crayon [P52][P53]. Normalize first, because FY2026 is a peak — ordinary profit walked 4,227 → 8,320 over five years [F6–F10], and management's own "real ROE" is 12.3% E30. Call normalized after-tax operating earnings ~¥5bn; at 10–11× that's ~¥52–55bn for the operating business; add ~¥35bn net cash D1 and IV is ~¥85–95bn, or ¥4,700–5,200 a share. Price ¥3,995 is about 78–85% of that. A fair price for a good business — maybe a small discount — but emphatically not fifty cents on the dollar. If I have to squint to see the gap, there is no gap [P50].

Two things keep me from dismissing it, and two decide the verdict. In its favor: this is a cannibal learning capital discipline, and I'm genuinely paid to wait [P17][P39] — 1,200k shares cancelled F494, all three megabank cross-holdings sold [E134–E136], and a payout stepping to 60% / 6.0% DOE from FY2027 E87, ~5% and rising while I wait. Against it: first, Fujitsu is both the 12.80% holder E81 and the dominant supplier — ¥18,965M of procurement through one related party on the board F483E139 — a concentration I can't fully underwrite from outside and that caps position size [P12]. Second, the ¥40bn strategic/M&A war-chest E28F236, funded by "cash and debt": that is growth by acquisition of unproven units, funded partly by levering the pristine balance sheet — the fortress that gives me my floor is the very thing they intend to spend [P38][P48]. So: a low-risk business but not a low-price one, with a real path to being paid while waiting — the textbook definition of watch. Few bets, big bets [P66]: I wouldn't make this a 10% position at ¥3,995. Where I'd swing is down near ¥2,400 — roughly half of the low end of my IV band, my standing fifty-cent-dollar margin — where the wait-and-get-paid mechanism is intact and I'm finally buying quality at a real discount.

What a student should take from this: a pristine balance sheet and a rising ROE make the downside safe, but "safe" and "cheap" are different questions — and I require both. At 1.5× book and ~10× earnings for a good business you have a fair price, not a fifty-cent dollar; the discipline is to name that honestly and wait. And watch what management intends to do with the fortress: net cash is a floor only until it's spent on unproven M&A.

Li Lu — too-hard

I begin where I always begin: at the boundary of what I can actually know. The question is not whether Tsuzuki is a good company — on the record it plainly is — but whether I can honestly claim to understand this integrator's next ten years better than almost anyone who already owns it. When I name the two or three variables that decide the outcome, the most important ones sit outside the public record and outside this company's control. That is where the analysis ends, at too-hard.

First, the business's due — because the temptation is real. This is a compounding machine, not a melting ice cube. Book value per share went 1,853.68 → 2,652.76 over five years [F31–F35], and the owner collected dividends of ¥48, 61, 90, 99, 126 on top [F118–F122]. The cleanest series is ordinary profit, which strips the one-off gains and rises every single year: 4,227 → 8,320 [F6–F10]. Intrinsic value is compounding [L35]. And management passes the trail-of-actions test [L52] handsomely: they cancelled 1,200k shares F494, sold all three megabank cross-holdings [F531–F533][E134–E136], upgraded the dividend policy to 60% / 6.0% DOE E87, and wrote in their own filing that they "were unable to invest sufficiently in growth" E22 — candor most Japanese managements never commit to paper. Clean auditor, 63 years, no officer over ¥100m, one outside director unpaid E119E120E128. This is a rational allocator. My verdict is not a criticism of these people.

Now the boundary. To predict earnings power in 2036 [L1], three variables decide it, and each is a problem. The first is Fujitsu — simultaneously the dominant supplier (¥18,965M, ~61% of procurement F483F216), the number-two shareholder at 12.80% E81, a related party E139, and the source of an outside director E109. If Fujitsu changes its policy, delivery methods, or procurement terms, earnings may be affected — the company says so E37. I cannot underwrite Fujitsu's channel strategy over a decade; it is exogenous to this company and invisible from outside, and it sits upstream of a large share of gross profit. The second is the ¥40bn strategic-investment program F236E28 — M&A, alliances, and VC, funded by "combining use of cash and debt," against a ¥72.8bn cap D4 and ¥35bn of net cash D1. They are telling me in advance they intend to convert more than half the company's value into acquisitions that do not yet exist, at prices not yet set. My whole discipline is to trace every retained yen [L21]; I cannot do that for capital not yet spent on targets not yet named. This converts a knowable, cash-rich compounder into an unknowable roll-up, precisely in the window I'm asked to predict. The third is softer but real: management itself writes that AI and in-house development are eroding "simple dev-build" E19, and its answer is the "AI Native" pivot to a service-centric model E24E25 — yet service revenue was essentially flat this year, +0.8% F442E51, while the record came from hardware and dev-build volume and margin management, not the service engine that is supposed to carry the next decade.

One more structural fact [L30]: Aso holds 23.97% E80 under a covenant to maintain its voting ratio E69 — a permanent control anchor, approved with the interested director recused and judged minority-friendly E71, which I take at face value, but which caps how much the reforming capital policy can ultimately do for an outside minority. Could the price rescue this? No. At ¥3,995 the stock is 1.51× book D3 and ~11.2× earnings F40 — and the honest ROE is 12.3%, not 14.0% E30. Quality at a fair price, not a dollar at fifty cents [L18]. My method requires that acknowledged unknowns be absorbed by the discount [L5]; here there is essentially no discount to absorb the Fujitsu dependence or the ¥40bn of unpriced M&A. So the verdict is too-hard, and the specific trigger is the knowledge bar [L1]: two of the three decisive variables are structurally unknowable to an outside minority, and the price offers no margin of safety to carry them.

What a student should take from this: a company can compound, be cash-rich, and be run by honest allocators, and still be too-hard — because the knowledge bar is about predictability, not quality or cheapness. When the two variables that most determine the next decade live outside the public record and outside the company's control, no record of past excellence and no fair multiple can manufacture the confidence you don't have. Admire it, put it on the shelf, and wait until either the unknowns resolve or the price falls far enough to absorb them.

Claude — watch, buy below ¥3,526 (implied ¥3,799)

I registered four hypotheses figures-blind and a central prior that this is fair-not-cheap quality behind a governance ceiling — the mistake to guard against being "good company mistaken for good minority investment." The ledger moved three of the four, and not all the way I expected. The honest resolution is watch: the business is better and modestly cheaper than my prior allowed, but there is no margin of safety at ¥3,995, the "quality" this year is a hardware-cyclical peak rather than a service annuity, and the net-cash floor I was counting as downside insurance is an announced ¥40bn M&A war-chest E28 about to be spent by an unproven allocator.

Margin of safety — overturned toward cheap, but not far enough to buy. Normalized owner earnings — I take management's own "real ROE" 12.3% ex-extraordinary E30, which reconciles to a tax-effected ordinary profit of ¥8,320M F10 — land at ~¥5,700M, a 7.8% owner-earnings yield on the ¥72,763M market cap. Strip deployable cash and the ex-cash EV/OE is 8.5×; a reverse-DCF at a 10% hurdle shows the price embeds only −1.6% perpetual growth on the operating business F10F76. That is not a price that has capitalized the 14.0% ROE F51 — the market is pricing decline. Net cash of ¥35,072M D1 is 48% of market cap and BPS ¥2,652.76 puts the tag at 1.51× book F35. So my figures-blind "already priced" was wrong: on cash-adjusted normalized earnings this is cheap-to-fair. But "cheap business" is not "margin of safety at this price." My downside case, built from the company's own history — revenue reverting toward the FY2025 ¥98,263M print F4 as the hardware spike unwinds, margin reverting to a ~5.5% mid-cycle between the FY2022 3.5% and FY2026 8.0% range F6F10 — yields ~¥3,300/share including deployable cash, a −17% draw. The price at which even that bear case clears an 8% owner-yield is ¥3,799 (my implied buy-below) — below the ¥3,995 stamp.

Service mix — overturned unfavorably, and unconfirmable in the margins. This was the hinge I most wanted the ledger to settle, and it settled against the re-rate case. The direction is the opposite of a service pivot: Service revenue rose +0.8%, Service orders fell −1.7% F199F203. The entire record year is the Hardware box-spike — Hardware orders +24.3%, backlog +62.7% F201F206, 63% of the +32.9% backlog surge — a public-sector/financial server cycle E49, not a growing annuity. Worse, the group is a single reportable segment E143, so there are no per-model gross margins disclosed — I am structurally blind to whether the service leg carries the margin the "professional-service company" narrative E23E25 claims. The AI-Native story remains a plan, exactly as my prior warned bare slogans would.

Capital return + M&A — split, and the M&A leg is now a live risk to the downside. The capital-return upgrade is genuine and dated: a 1,200k-share cancellation E76, all three megabank cross-holdings sold [E134–E136], and a step to 60% payout / 6.0% DOE from FY2027 E87 — ~4.7% forward yield, you're paid to wait. But the ¥40bn strategic-investment ambition E28 exceeds all net cash and is explicitly funded by "cash + debt." So the net-cash floor is not a permanent cushion — it is a declared war-chest management intends to spend and lever, by an unproven allocator that conceded it "could not invest sufficiently in growth" E22. A buy that relies on this cash as downside insurance is underwriting a floor management has announced it will remove.

Minority not trapped — confirmed as a ceiling, not a trap-door. Aso 23.97% E80 + Fujitsu 12.80% E81 + Fuso 4.08% + employee association 3.02% + trusts + treasury ≈ a ~49% allegiant block E82E83; true float ~39%, foreign 6.7% E79; special resolutions blocked by the blocks E102. Aso's anti-dilution covenant E69 constrains the very equity-funded M&A flexibility the growth leg would want, and Fujitsu is the sharpest point — 61% of procurement, a 12.8% owner, and a board director E109E139, with terms set "per the sales-partner agreement" E68 and not independently benchmarked. This is not left-tail risk; it is a ceiling on the re-rate and a channel through which margin could leak to the vendor.

Resolving the tension → watch. The business is genuinely good and, on cash-adjusted normalized earnings, modestly cheaper than I expected — the reverse-DCF pricing −1.6% growth is a real disconfirmation of "already priced." But three facts keep it from a buy: my downside arithmetic implies a buy-below of ¥3,799, so there is no margin of safety at ¥3,995; the record year is a hardware-cyclical peak with a stalling service leg I cannot margin-audit; and the net-cash cushion is an announced M&A spend by an unproven allocator behind a locked register. That is the textbook profile of a good business I'd happily own cheaper — a watch, with the upgrade being a lower price (≤¥3,526, where the bear clears 9%) or hard evidence the mix is actually shifting to services. This is not too-hard: the load-bearing unknowns are resolvable by the next 1–2 filings.

What a student should take from this: a reverse-DCF is the fastest way to separate "the business is good" from "the price is good" — here it showed a 14%-ROE compounder priced for −1.6% perpetual decline, genuinely cheap-ish, yet the verdict is still only watch. Cheapness on normalized earnings is necessary but not sufficient: the margin of safety must survive the bear case (mine implied a buy-below below the market price), the "quality" must be the durable kind (this year's record was a cyclical hardware spike, not the growing service annuity the story sold), and a net-cash "floor" management has publicly earmarked for M&A is not a floor at all.

Synthesis

Where the five lenses agree

For the first time in the record, every lens respects the business. The facts are not in dispute and no one calls the company bad. Tsuzuki earns a 14.0% ROE F51 on net cash of ¥35,072M D1 with a 55.1% equity ratio F46; it is growing (fourth straight record, operating profit +26.2% F364F192); it is genuinely shareholder-friendly on capital allocation (a share cancellation E76, every megabank cross-holding sold [E134–E136], a dividend step to 60%/6.0% DOE E87); and its governance is clean on paper (8/10 outside board E92, clean 63-year auditor with a single dev-build KAM E119E148). Buffett calls the balance-sheet fact "the strongest single fact" [B60]; Munger concedes it is "a good business… the balance sheet is fortress-clean" [M37]; Pabrai builds a hard floor on it [P1]; Li Lu calls it "a compounding machine, not a melting ice cube" [L35]; Claude's reverse-DCF finds it priced for decline [C34]. There is no bear among the five — and no buyer at ¥3,995 either.

Where the lenses diverge

The split is three-way — watch (Buffett, Pabrai, Claude) vs pass (Munger) vs too-hard (Li Lu) — and it is a disagreement about what kind of "not yet" the record supports: wait-for-a-price, good-but-not-good-enough-at-a-fair-price, or can't-underwrite-the-ten-year-minority-outcome.

Munger will not wait for a price, because a lower price would not fix his objection: "This is priced correctly as a good business — which is exactly the quadrant that earns a pass. Name the moat mechanism or admit it does not exist [M18]. Hardware at 42% D6 is intermediary economics, gross margin 24.1% is integrator-thin F360, and the capital-return upgrade E87 and backlog F212 are visible to every analyst after a 3× run from ¥1,215 F174. Great-at-a-fair-price, not fair-at-a-great-price [M37][M44]." Li Lu will not wait for this price either, but for a different reason — the knowledge bar, not the moat: "Two of the three variables that decide 2036 — Fujitsu's channel strategy E37E139 and the outcome of a ¥40bn acquisition program not yet begun E28 — are unknowable to an outside minority, and at 1.51× book D3 there is no discount to absorb them [L1][L5]. A fine business I cannot predict is a pass, however much I admire the people running it."

The three watchers agree the business is worth owning at some price and split only on what that price is — and the dispersion is striking. Pabrai and Buffett sit near book: "At 1.5× book and ~10× earnings you have a fair price, not a fifty-cent dollar [P53] — I'd swing near ¥2,400, half the low end of my IV band," says Pabrai; Buffett wants "roughly ¥2,650, around book, where the owner-earnings yield pushes toward double digits and the Fujitsu question becomes an option rather than a wager [B89][B93]." Claude sits far higher, near the stamp: "The reverse-DCF prices −1.6% perpetual growth [C34] — this is cheap-ish, not already priced — so my buy-below is ¥3,526, and my implied downside threshold ¥3,799 lands just below the stamp." The gap between ¥2,400 and ¥3,526 is not a disagreement about the facts; it is a disagreement about whether FY2026 earnings are a cyclical hardware peak or normalized, and how much margin of safety the quality warrants. Claude and Munger say peak — the record is the Hardware box-spike (+62.7% backlog F206) with a stalling service leg (orders −1.7% F203), so mid-cycle the multiple is ~14× not ~10× [C40]; Buffett and Pabrai normalize down to the same conclusion by a different route (strip the ¥2,419M securities gain F372, anchor on "real ROE" 12.3% E30) and simply demand a wider discount for the quality they can't fully attribute.

Three shared threads run through every lens's reasoning. (a) The Fujitsu supplier-and-shareholder duality — 61% of procurement, a 12.8% owner, and a board director at once F483E139E109 — which Buffett frames as "whose franchise is it?" [B23], Munger as a structural conflict [M18], Li Lu as an unknowable exogenous variable [L1], Pabrai as a position-size cap [P12], and Claude as an un-benchmarked margin-leak channel [C78]. (b) The ¥40bn M&A ambition that spends the net-cash floor E28F236 — the one fact that turns the balance-sheet strength every lens admires into a live risk: Pabrai's floor, Claude's downside insurance, and Li Lu's knowable-compounder all depend on cash management has announced it will spend and lever. (c) The ~49% Aso/Fujitsu locked register E80E81E82 — a ceiling on any re-rate (no takeover premium, minorities in the back seat) that Aso's voting-maintenance covenant E69 entrenches. Every lens weighed all three; they differ only on whether those threads cap the price you'd pay (the watchers) or the analyzability of the name itself (Munger, Li Lu).

Self-distance note. The Claude lens holds one of the five verdicts compared above (watch) and wrote this synthesis; it also built the dual-blind reconciled figure table and evidence ledger (two independent passes each, reconciled against page-delimited source text) that all five lenses consumed. That is an unusual concentration of authorship in one model: the answerer, the ledger-builder, and one of the five voters are the same system — a same-model concentration. Read the synthesis with that in mind.

Prediction-vs-actual: VOID. This was an autonomous headless cycle; the human blind prediction is voided (void: no-human-prediction, never forged). No prediction-vs-actual scoring applies to this study.

Verdict accounting (fixed ex-ante)

• A buy-below-¥X verdict is price-falsifiable against the unadjusted stamp. The three watch thresholds recorded here — Pabrai ¥2,400, Buffett ¥2,650, Claude ¥3,526 (plus Claude's implied downside threshold ¥3,799) — are all below the ¥3,995 stamp; they are the prices at which each watcher would re-engage.
• pass / watch / too-hard are recorded but unscored in any future review. Munger's pass and Li Lu's too-hard carry no buy-below by construction.
• The original verdict counts at its original stamp regardless of later corrections.
• On a stock split, reverse split, or consolidation, the buy-below threshold restates mechanically by the announced ratio (corporate-action disclosure cited); the stamp itself never restates.

What would change our minds

Pre-registered falsifiers, per lens issuing a buy-below or watch. Future review notes score against these, not hindsight.

• Buffett (watch ¥2,650). The verdict must be re-underwritten if Fujitsu ceases to be both the dominant supplier (¥18,965M procurement F483 via the auto-renewing sales-partner contract E68) and a 12.8% holder F485 — i.e. the relationship is lost (the earnings leg collapses) or bought out/normalized (a real independent franchise emerges). Either way the "is-it-a-franchise-or-a-reseller" question changes.
• Pabrai (watch ¥2,400). Buy if the price falls to ~¥2,400 — roughly 50% of the low end of the conservative IV band ~¥4,700–5,200/sh D1F10 — making it a genuine fifty-cent dollar. Downgrade to pass if the ¥40bn M&A war-chest F236E28 is deployed into a sizeable unproven acquisition at a full multiple — that spends the net-cash floor that makes the downside safe.
• Claude (watch ¥3,526 / implied ¥3,799). Upgrade to buy if FY2027 first-half data show Service revenue AND Service order intake both growing >5% YoY (mix genuinely shifting, not a hardware-spike year) with price at/below ¥3,526. Downgrade to too-hard if the ¥40bn program executes its first material deal with no post-deal ROIC disclosure and net cash falls below ¥15bn while Fujitsu procurement terms remain un-benchmarked E28E139.
• Munger (pass) and Li Lu (too-hard) carry no price falsifier by construction — they issued neither a buy-below nor a watch, so there is nothing of that kind to pre-register. We say so plainly rather than manufacture one.

The single observable most lenses converge on is whether FY2026's record is a durable service annuity or a hardware-cyclical peak — resolved by the FY2027 first-half order mix (Hardware roll-over vs Service intake F201F203).

What this taught the checklists

Queued for the next study (F2 revision proposals; see docs/process/evolution.md), attributed per lens:

• Buffett — a "supplier that is simultaneously a major shareholder + director" item is missing. B37 (supplier dependence under franchise/commodity), B78 (owner alignment) and B84 (blockers) each touch part of it, but nothing squarely handles a party that is at once the dominant supplier, a major shareholder, and a competitor's partner E37E68E109E81. The Fujitsu case fell between B23, B26 and B78 rather than being caught cleanly. Consider an explicit "captured-franchise / related-party-supplier" item — and note that B76 (acquisition record) can only answer data-insufficient against a company announcing its first big program E28.
• Munger — two ledger-gap notes under single-segment reporting. M18/M27 need customer-concentration and churn/retention data the filing omits, and M8/M22/M24 need at least two named competitors' metrics to be answerable rather than perpetually data-insufficient. And M9/M19/M40 need per-segment (per-business-model) margin data: with 42% hardware / 17% dev-build / 41% service [D6–D8] hidden inside one reported segment F444, the moat/commodity answer differs by model and the checklist should demand the split.
• Pabrai — P12 (revenue concentration) is half-blind to a distributor/SI whose binding dependence is a dominant supplier who is also a related-party shareholder F483E139; add a companion clause treating that as concentration risk even absent customer disclosure. And P38/P48 (replication vs acquisition) sit far apart in the checklist but are really one question for a fortress balance sheet announcing a large M&A envelope F236 — a cross-reference would have saved assembling it by hand; it was the single most decision-relevant pairing here.
• Li Lu — three. (1) L1 needs a "supplier-is-also-shareholder-is-also-related-party" sub-prompt — the Fujitsu situation is a distinct species of dependence the current L1/L11/L37 phrasing scatters across three items. (2) An "announced-but-unspent forward capital program" item is missing — L21–L24 audit retained capital historically; nothing handles a large forward commitment (¥40bn vs a ¥72.8bn cap F236D4) that converts a knowable business into an unknowable one prospectively. (3) A mid-window-divestiture scoring note so L2/L3/L38/L39 don't misread incomparability as weakness: the Restar spin E7E9 breaks the revenue series (124,856 → 98,263 F3F4) cleanly enough that "downturn record" and "growth decomposition" return data-insufficient for the current entity.
• Claude — two. (1) A "mix-shift margin proof" item: when a company narrates a product-to-service pivot E25 but reports a single segment E143, nothing forces the analyst to demand the service leg's gross margin (or downgrade the mix-shift claim from evidence to narrative) — that gap was load-bearing here, since H2 turned on it. (2) A C47 sharpening that earmarked/M&A-committed cash is not a downside floor — re-run the bear case (C33) with the cash spent, since a net-cash screen counting the cash at face is wrong when management has publicly committed it to M&A E28.

---

Corrections