Coherent Corp COHR

Thesis

Coherent Corp (COHR) is the post-merger combination of legacy II-VI and the original Coherent Inc., creating a vertically integrated supplier of compound-semiconductor materials (InP, SiC, GaAs), photonic components, and high-end industrial/scientific lasers. The bull thesis is straightforward: hyperscaler AI buildouts require ever-faster optical interconnects (400G to 800G to 1.6T pluggable transceivers), and Coherent is one of perhaps three credible Western suppliers with end-to-end capability — wafer to module. Datacom is now the single largest growth driver, and in good quarters the optical communications segment is growing 30-50% year over year. The problem is that none of this shows up in the long-run economics. The scorecard tells the truth: 10-year average ROIC of 4.16%, share count up 10.08% over a decade (almost entirely the Finisar and Coherent equity issuance), TTM P/E of 779x, and EV/FCF of 359x. Owner earnings of roughly $447M against a current enterprise value implied by the EV/FCF ratio gives a price-to-IV ratio of 0.0023 in the scorer — a mechanical artifact, but the directional message is unambiguous: every dollar of optimistic 2027-2030 earnings is already discounted into today's quote. The composite score of 72 is respectable, driven mostly by valuation methodology generosity and FCF conversion of 1.44x; the underlying business quality (profitability bucket: 17) is mediocre. At $329.50, with the scorer's base-CAGR clamped from 178.8% to 14% and maintenance capex flagged as uncertain by more than 50%, there is no margin of safety. Owning this requires believing AI optics scales linearly with model sizes for a decade — a tech-adoption forecast Munger would refuse to make. Thesis: avoid until either the price halves or the company demonstrates two consecutive years of double-digit ROIC at scale.

Moat

Coherent operates in a sector where moats are narrow and perishable. Apply the five-moat checklist:

1. Pricing power. Coherent sells into hyperscalers (Microsoft, Meta, Google, Amazon), Apple's VCSEL supply chain, and industrial OEMs. These are the most sophisticated, most concentrated, most ruthless buyers on earth. They run dual-source procurement by policy. There is essentially no pricing power — gross margins in datacom transceivers compress with every product transition (100G to 400G to 800G), and the only way to defend dollar margins is to ship faster speeds at lower cost per bit. This is the opposite of See's Candy [4], where the moat is built around "the lush earnings of the business" reinvested into similar businesses. Verdict: minimal.

2. Switching costs. In compound-semiconductor materials (SiC substrates, InP wafers) there is some qualification stickiness — once a customer qualifies a supplier into a power-electronics or optical product, requalifying takes 12-18 months. But the customer ultimately controls the design and can — and does — multi-source. This is unlike Microsoft Office's switching cost moat described by Damodaran [2], where the end-user pays the cost; here, the OEM pays nothing meaningful to switch.

3. Network effects. None. Photonics is not a two-sided market.

4. Intangibles — patents, trade secrets, process know-how. This is Coherent's strongest claim. Decades of accumulated process IP in epitaxial growth, laser cavity design, and wafer-bonding produce real know-how that competitors cannot replicate quickly. Damodaran's framework [1, 2] notes that patent moats matter most when R&D is productive, not merely large — and Coherent's cumulative R&D has produced a real catalog. But patents in optoelectronics expire, are designed around, and are increasingly threatened by Chinese state-funded entrants (Accelink, Hisense, Innolight). The cumulative 10-year ROIC of 4.16% is the empirical proof that the intangibles, in aggregate, are not generating excess returns.

5. Cost advantages. Coherent has scale in SiC and InP wafer production. Damodaran's Home Depot example [6] notes scale-driven cost advantages can be durable. But scale in semiconductor manufacturing is only a moat if utilization stays high — and this business is sharply cyclical, so utilization swings wildly. Coherent has had multiple quarters of negative gross margin variance from underutilization. Vertical integration (wafer to module) is a theoretical cost advantage but in practice it traps capital in non-economic stages of the value chain.

Competitor stress test ($10B + 5 years). Hand $10B and five years to a determined entrant — Marvell, Broadcom on the silicon-photonics side; Sumitomo, Lumentum, Fabrinet on the module side; or a Chinese national champion — and Coherent's positions in 800G/1.6T modules and SiC substrates would erode meaningfully. The ASIC vendors (Broadcom, Marvell) are already pulling integration up the stack with co-packaged optics (CPO), which threatens to disintermediate the discrete pluggable module — Coherent's biggest growth driver — over the 2027-2030 horizon. This is exactly the dynamic Damodaran [3] warns about: "the presence of these excess returns will attract new entrants and imitation will push excess returns down."

Erosion risk. High. The product cycle is 18-24 months. Each transition resets pricing. Chinese competition is structural and subsidized.

Moat verdict: NARROW.

Management & Capital Allocation

Capital allocation at Coherent has been the defining story — and not a flattering one. The 5-choice framework:

1. Reinvestment in the business. Coherent reinvests heavily — capex routinely runs 8-12% of revenue, and R&D another 9-11%. The problem is the return on that reinvestment. The scorecard reports a 10-year average ROIC of 4.16% and notes ROIIC is 'not meaningful' because NOPAT declined. Reinvesting at sub-cost-of-capital rates is value-destroying by definition. Buffett's 2007 letter [4] is explicit: "Truly great businesses, earning huge returns on tangible assets, can't for any extended period reinvest a large portion of their earnings internally at high rates of return." Coherent has the inverse problem — it reinvests heavily at low rates because the industry demands constant capacity refresh.

2. Acquisitions. This is the central capital-allocation event. The $7.1B Finisar acquisition (closed 2019, by legacy II-VI) and the $7B Coherent Inc. acquisition (closed 2022) together transformed the company from a niche materials player into a photonics conglomerate. The headline issue is that both deals were funded with a mix of stock, term loans, and preferred equity — culminating in roughly $4-5B of net debt and a high-cost preferred (originally with Bain Capital and others) that diluted common holders. Share count is up 10.08% over the trailing decade. The merger thesis (vertical integration from wafer to module) is intellectually coherent but the integration has been operationally messy: gross margins compressed for several quarters post-close, and the EBITDA multiple paid was elevated. The current CEO, Jim Anderson (ex-Lattice Semiconductor), inherited the integration in 2024 and has begun a portfolio review including divestiture of non-core assets.

3. Debt management. The scorecard reports net debt/EBITDA of 0.0044, which is implausibly low for a post-merger entity and reflects either a methodology artifact (perhaps a TTM EBITDA blow-up) or recent debt paydown after segment sales. Reported debt is in the $4B+ range against an EBITDA run-rate that has been recovering. Interest coverage is null in the scorecard. The company has been refinancing the term loan B and paying down preferred — directionally healthy, but the balance sheet is not Buffett-grade.

4. Buybacks. Effectively none. The share count trajectory is dilution, not retirement. There is no repurchase program of consequence. This earns no credit.

5. Dividends. None on common (a small preferred dividend exists). Reasonable for a reinvesting business — but only if reinvestment earned high returns. It does not.

Communication quality. Investor communications under Anderson have improved — clear segment reporting, explicit margin bridges, and a focus on AI datacom unit economics. Earlier (Mattera-era II-VI) communications were promotional and acquisition-focused.

The full record: two large equity-funded acquisitions, persistent dilution, sub-cost-of-capital ROIC, no buybacks, and a balance sheet that took years to repair. The new CEO is moving in the right direction (portfolio focus, debt paydown), but a Buffett-Munger grade is the ten-year record, not the trailing twelve months.

Capital allocator: C.

Industry Structure

Photonics and optical communications is a structurally Average-to-Poor industry through Porter's Five Forces:

1. Rivalry among competitors — HIGH. The optical-component market is fragmented and aggressive. In datacom transceivers Coherent competes with Lumentum, Marvell (Inphi), Broadcom, Innolight, Eoptolink, Accelink, Hisense, Source Photonics, and Fabrinet-built modules. In industrial lasers it competes with IPG Photonics, Trumpf, nLight, and a long Chinese tail. In materials (SiC, InP) it faces Wolfspeed, onsemi, Sumitomo, and II-VI's own former captive customers. Product cycles are 18-24 months and price/bit declines 20-40% per generation. This is a brutal pricing environment — closer to Damodaran's airlines example [3] than to See's Candy [4].

2. Threat of new entrants — MEDIUM-HIGH. Capital intensity provides some barrier (a SiC fab is $1B+), but Chinese state-funded entrants have demonstrated repeatedly that they will fund through losses for a decade to win share. In silicon photonics, the vertical integration of hyperscaler ASIC vendors (Broadcom, Marvell, Nvidia) into co-packaged optics is the existential new-entrant threat — it doesn't bring a new company in, it brings new capability into customers who were previously buyers.

3. Bargaining power of suppliers — MEDIUM. Coherent owns much of its own materials supply (vertical integration in InP and SiC) which neutralizes upstream supplier power. But for laser diodes, optics, and packaging the company is exposed to a small set of specialized suppliers.

4. Bargaining power of buyers — VERY HIGH. This is the single most damaging force. The top 5 hyperscalers represent the majority of datacom transceiver demand. Apple is a concentrated VCSEL buyer. Industrial laser buyers (Tier 1 auto, electronics OEMs) are sophisticated repeat purchasers. Hyperscalers announce dual-sourcing as a procurement principle. Buyers extract every cent of supplier-side scale benefits.

5. Threat of substitutes — MEDIUM. Within optical, the substitution risk is intra-industry (CPO replacing pluggables, silicon photonics replacing InP). Across the broader interconnect stack, electrical (copper) is being displaced by optical, which is favorable. The risk is that Coherent's specific technology choices get substituted, not that optics gets substituted.

Value pool location and trajectory. Historically the value in optics has accrued to (a) ASIC/DSP vendors (Broadcom, Marvell), (b) the EDA and silicon side rather than packaging/modules, and (c) hyperscalers themselves through procurement leverage. The module/component layer where Coherent operates is the lower-margin layer. The trajectory is for value to migrate further toward integrated silicon-photonics solutions over the next 5-10 years.

Industry Verdict: Average. The AI datacom tailwind raises near-term unit volumes, but it does not change the structural economics of being a component supplier to hyperscalers.

Inversion (Bear Case)

I am now playing the short. I do not soften.

1. The single event that kills this. Co-packaged optics (CPO) ramps faster than expected at one of Nvidia, Broadcom, or Marvell. The first hyperscaler — likely Microsoft or Meta — qualifies a CPO-based switch in volume in 2027. Pluggable transceiver demand at 1.6T and beyond is curtailed. Coherent's biggest growth segment loses 30-50% of its 2030 TAM. Concurrently, a second-source Chinese transceiver vendor (Innolight or Eoptolink) wins a major hyperscaler 800G/1.6T design slot, resetting Coherent's pricing and unit share. The stock that traded at 30x forward 'AI infrastructure' multiples re-rates to 12x trough datacom-equipment multiples. The narrative breaks in a single quarter.

2. Why the moat is narrower than bulls think. Bulls cite vertical integration (wafer to module) as the moat. The 10-year ROIC of 4.16% is the empirical refutation. Vertical integration in a product-cycle business with hyperscaler buyers is a capital trap, not a moat: it forces capacity additions into every cycle peak that then sit underutilized into every trough. The patent estate is real but expiring and design-around-able. The customer relationships are real but every hyperscaler runs dual-source by mandate. Damodaran's [3] excess-return-decay framework predicts exactly this: a sub-15% ROIC business cannot have a wide moat by definition, because moats are the cause of high ROIC. The narrative argues the moat exists and the returns will follow. In ten years, that has not happened. It will not happen now.

3. Why management is worse than it appears. Two large acquisitions (Finisar, Coherent Inc.) funded with stock and high-cost preferred. Share count up 10% in a decade. Zero meaningful buybacks. Multiple post-close margin disappointments. The new CEO Anderson is competent (Lattice was a real success), but he inherited a balance sheet that absorbed years of free cash flow just to pay down debt and preferred — capital that, in a Buffett business, would have compounded for owners. Management at this company has spent the last decade building scale, not value. The portfolio review now underway is an admission that the empire was over-assembled. Investors should ask why, after a $14B combined acquisition program, the consolidated 10-year ROIC is 4%.

4. What bulls are extrapolating that won't hold. Bulls extrapolate (a) AI datacom unit volumes growing 30%+ annually for five years, (b) 800G and 1.6T pricing holding in a way that prior generations did not, (c) Coherent maintaining or gaining hyperscaler share against Innolight/Eoptolink/Lumentum, (d) gross margins expanding 500-800 bps as datacom mix grows, and (e) free cash flow conversion holding at the recent 1.44x while capex is ramping. Each of these extrapolations is plausible alone. Multiplied together they require a near-perfect path. The base rate for component suppliers hitting all five at once is approximately zero. The TTM P/E of 779x prices something close to all five.

5. Valuation trap — multiple compression / regime change. TTM P/E of 779x and EV/FCF of 359x are not investment metrics — they are momentum metrics. The 10-year average P/E of 26.84x is a more reasonable terminal multiple. If owner earnings of $447M scale to, say, $1.2B in 2028 (a 28% CAGR — already optimistic) and the multiple compresses to 18-22x — appropriate for a cyclical hardware business — the equity is worth roughly $22-26B. Today's market cap is north of $50B. That implies 50%+ downside even on a constructive scenario. In a regime change where AI capex pauses for 4-6 quarters (which has happened in every prior tech-capex cycle), the multiple goes to 12-15x trough earnings of $700-900M and the stock prints below $100.

If I am right, the stock could be worth $110-150 within 2-3 years.

Lollapalooza Bias Check

Active biases the analyst should call out:

Recency bias — strongly active. The last six quarters have been excellent for Coherent's datacom segment. Stock-price action has been violently positive. There is a strong pull to extrapolate that pattern. Antidote: weight the 10-year ROIC of 4.16% at least as heavily as the trailing six quarters.

Authority bias — active. Sell-side analysts have placed COHR on "AI picks" lists. Buy-side AI thematic funds have made it a core holding. The 779x trailing P/E is socially sanctioned because the names involved are sanctioned. Antidote: ask whether any single member of those analyst lists could explain why a sub-5% ROIC business should trade at >50x normalized earnings.

Social proof / commitment — active. Owning AI infrastructure names is the consensus 2024-2026 trade. Going against it requires explaining yourself at every dinner. The lollapalooza of social proof + authority + recency combines into a narrative-stock effect that Munger explicitly warned about.

Anchoring — active in two directions. First, anchoring to the current $329.50 quote — "surely the market is roughly right." Second, anchoring to the $143.78B base-case IV in the scorecard, which is itself a function of optimistic clamped CAGRs and uncertain maintenance capex (the scorer flagged the >50% capex spread). Antidote: re-derive IV from owner earnings of $447M times a normalized 18-22x — a much smaller number.

Confirmation bias — active. Once an analyst calls something "Too Hard" or "Avoid," it is psychologically easy to find disconfirming data and dismiss it. Antidote: explicitly enumerate the strongest bull arguments (AI capex durability, vertical-integration synergy, CPO is over-hyped, Chinese competition is over-stated) and weigh them honestly.

Deprival superreaction — latent. A bull who already owns will resist Trim/Sell because of fear of missing further upside. This is the classic loss-aversion-on-paper-gain pattern. Antidote: think in terms of expected value, not regret.

Incentive bias — meta-level. Compounder analysts are rewarded for finding compounders. There is a subtle incentive to issue Buy ratings to justify the work. The honest call is often "Too Hard" or "Avoid," which requires resisting the incentive structure of the framework itself.

Net: at least four biases are pulling toward a more bullish call than the underlying numbers support. The right move is to bias the conclusion in the opposite direction.

10-Year Outlook

Same fundamental business model in 10 years? Probably not. Today's product mix — pluggable optical transceivers, SiC substrates, industrial fiber lasers, materials — is plausibly very different from the 2035 mix. CPO is likely to displace a meaningful portion of pluggable transceivers. SiC may be commoditized or partially displaced by GaN in some power applications. Industrial laser pricing will continue its grind. The company will still exist, will still sell photonic components, but the specific revenue mix that justifies today's price is unlikely to be recognizable.

Customer base larger? Yes, almost certainly. Optics and compound semiconductors are growth areas in absolute terms. AI, autonomous driving, EV power, telecom, sensing all expand the addressable customer base.

Profit per customer higher? Doubtful. Hyperscalers are not getting less powerful as buyers; they are getting more powerful and more concentrated. Apple is more aggressive on supplier margins, not less. Industrial OEMs continue to pursue dual-sourcing. The structural buyer-power problem will compound, not relent.

Moat wider? Unlikely. The patent base will turn over. Process know-how will leak (employees move, Chinese competitors replicate). Vertical integration becomes a liability when interfaces standardize. The most likely path is moat erosion, not deepening.

Single biggest threat? Co-packaged optics adoption combined with Chinese transceiver competition. Both are visible today; both are accelerating; both attack the highest-growth, highest-margin part of the current mix.

Apply Buffett's 2007 letter [4] standard: "Long-term competitive advantage in a stable industry is what we seek." Photonics is not a stable industry. The competitive advantage is narrow. The 10-year picture is reasonably likely to feature lower margins, more competition, a different product mix, and a balance sheet that has spent another cycle being repaired. None of that is consistent with paying 779x trailing earnings today.

CONFIDENCE: low