Lumentum

Phase A — Understand the business

Lens 1 · Company Overview

Lumentum makes the light for the internet — the lasers, optical chips, and photonic modules that move data through cloud data centers and telecom networks, plus an industrial-laser business. Plain-terms model: it sells hardware (point-in-time revenue recognition, ~12-month warranties, much of it through vendor-managed-inventory hubs), priced almost entirely in USD, manufactured increasingly outside the US (UK, China, Thailand, Japan).

The company self-describes as "a leading provider of optical and photonic products… recognized as an industry leader based on revenue and market share," whose products are "essential to a range of cloud, artificial intelligence and machine learning (AI/ML), telecommunications, consumer, and industrial end-market applications".

Reporting structure changed. Through FY2025 the company ran two reportable segments — Cloud & Networking and Industrial Tech. As of the FY26 10-Qs it reports one reportable segment and disaggregates revenue only by product type (Components vs Systems) and geography. That re-segmentation itself is a tell: the business is now overwhelmingly one thing — optics for AI/cloud data centers.

• Products (Components, 66% of Q3 FY26 rev): the building blocks — semiconductor laser chips (the crown jewel: EML/electro-absorption-modulated lasers and CW-DFB lasers), laser sub-assemblies, line subsystems, wavelength-management (ROADM) systems. These feed module-makers and network-equipment OEMs.
• Products (Systems, 34%): complete products — optical transceiver modules (incl. the cloud/datacom transceivers), and the new optical circuit switch (OCS).
• Main customers: hyperscalers, AI/ML infrastructure providers (NVIDIA prominently), and network-equipment manufacturers. Concentration is high — see Lens 4.
• Suppliers / manufacturing: Lumentum is "heavily dependent on a small number of manufacturing sites" and relies on independent contract manufacturers, "for many products a particular contract manufacturer may be the sole source of the finished good," with significant supplier exposure to China and a manufacturing footprint in Thailand.
• Competitors: Coherent (the old Finisar/II-VI), InnoLight, Fabrinet (assembly), Marvell/Broadcom (DSP/switch silicon adjacency), plus Chinese module makers.

The headline contract structure: in March 2026 NVIDIA made a $2.0B Series A convertible-preferred investment plus a multi-billion-dollar purchase commitment and future capacity-access rights for advanced laser components — funding a new US fab. This converts the customer relationship into something closer to a strategic supply-lock. Lumentum is also described as the sole-sourced supplier of co-packaged optics (CPO) for NVIDIA's upcoming switch platforms.

Lens 2 · Supply Chain

Map, upstream → Lumentum → end demand, named where the filings/web name them:

Upstream inputs:

• Compound-semiconductor wafers (Indium Phosphide, GaAs) — Lumentum is unusually vertically integrated here: it fabs its own InP laser chips (the Oclaro/NeoPhotonics legacy). This is the moat (Lens 3).
• Contract manufacturers / OEM assembly — Lumentum both runs its own fabs and depends on third-party CMs, several of them sole-source for a given finished good, concentrated in Asia (notably Thailand and China). The mirror-image of this is Fabrinet, which assembles for the broader industry.
• Equipment / capacity: the NVIDIA $2B is explicitly earmarked for "R&D, future capacity and operations" and a new US fab in Greensboro, North Carolina — i.e., Lumentum is being financed to add upstream capacity it could not self-fund fast enough.

Lumentum (the node): laser chips + sub-assemblies (Components) and modules + OCS (Systems).

Downstream / end customers:

• Module makers & OEMs buy Lumentum's EML chips and integrate eight 200G/lane EMLs into one 1.6T transceiver. So Lumentum sells picks-and-shovels into its own competitors' modules — InnoLight and Coherent build modules using lasers that, in part, originate from Lumentum.
• Hyperscalers / AI buyers: Google (named as using bi-directional transceivers with OCS), NVIDIA (the anchor), and the broad cloud build-out.
• Geographic ship-to (Q3 FY26): Thailand 21.9%, US 23.0%, Hong Kong 16.8%, Mexico 12.8%, China 8.7%. Critical caveat from the filing: "Net revenue is assigned to the geographic region and country where our product is initially shipped" to a contract manufacturer — so these are CM ship-to locations, not end-demand geographies. The China line (8.7%) understates neither real China end-demand nor the US-export-control exposure cleanly.

Chokepoints / single-source dependencies:

1. Lumentum's own EML fab is the chokepoint for the industry — it is the single most important supplier of 200G/lane EMLs, which is precisely why NVIDIA paid $2B for capacity rights.
2. Contract-manufacturer sole-sourcing is a chokepoint for Lumentum — a labor strike at a former CM once "threatened … our ability to fulfill our obligations to our customers".
3. Geopolitics: higher US tariffs on goods from Thailand/China, and export controls (EAR/ITAR; the Huawei/FiberHome Entity-List history) sit directly across the chain.

Lens 3 · Competitive Advantages (moats)

The moat is process + IP in the laser chip, not the module.

• EML chip leadership. Lumentum "is the world leader in 200G-per-lane EML laser chips" and "currently controls about 50% to 60% of the entire world market for these advanced lasers". EMLs are hard: InP epitaxy, yield, and reliability at 200G/lane are a genuine technical wall. Every 1.6T DR4 transceiver needs eight of these chips. That is a scale + process + IP moat, reinforced by a decade of consolidation (Oclaro 2018, NeoPhotonics 2022) that swept up the best InP laser IP.
• Vertical integration. Owning the laser fab lets Lumentum capture the highest-margin layer (Components gross margin pulls the corporate number up as mix shifts to chips) and supply its own module competitors — a structurally advantaged position: it wins whether the module is made by InnoLight, Coherent, or in-house at a hyperscaler.
• OCS / MEMS. The optical circuit switch is built on "field-proven MEMS technology" (R300 up to 300×300 ports, compact R64 64×64), with backlog already "well beyond $400 million" and a Google design-win pattern. This is a second moat lane — switching, not just transmission.
• The NVIDIA lock. Sole-source CPO for NVIDIA's next switch platforms + a $2B capacity prepayment + multi-billion purchase commitment is a customer-switching-cost moat that did not exist 12 months ago.

Bargaining power — honest read: mixed. Over its laser-chip customers, Lumentum currently has strong pricing power (sold-out capacity, 540bps QoQ gross-margin gain — Lens 5). But the 10-Q is candid that "large OEM and end-user service providers … have considerable bargaining power," can demand "most-favored-nation or exclusivity provisions," and dictate terms that affect revenue timing. With one customer at 26% of revenue, the buyer has leverage too. Net: Lumentum holds the technical whip hand today because capacity is scarce; that flips if EML supply catches up to demand.

Versus rivals on perceived value: Lumentum is positioned as the component/laser leader, while InnoLight is the module-share leader (">50% of NVIDIA's optical-module procurement wallet") and Coherent ~20%+. So at the module layer Lumentum is a follower; at the laser-chip layer it is the leader. The bull case requires the value to keep migrating to the chip.

---

Phase B — Measure performance

Lens 4 · Segments

Lumentum now reports one segment, disaggregated two ways. All figures unless noted.

By product type (fiscal Q3 FY26, quarter ended 2026-03-28):

Nine-month FY26: Components $1,356.2M (+70.4%), Systems $651.5M (+76.8%), total $2,007.7M (+72.4%).

What drove it (from MD&A): Components growth = "the ramp of laser chip and laser assembly product shipments… across intra-data center, data center interconnect, and long-haul," plus a slight ASP uplift from the shift to 200G lane speeds. Systems growth = cloud transceiver product lines up >$137.0M in the quarter (>$268M over 9 months), plus the initial OCS ramp (>$25M in Q3, >$38M over 9 months) — explicitly "on track for manufacturing expansion over the coming quarters".

Legacy segment trend (FY history, for context):

The story in one line: Industrial Tech (the old 3D-sensing/iPhone-VCSEL + industrial-laser business) has shrunk every year (consumer 3D-sensing lost share to competition), while Cloud & Networking has become the entire company and is now re-accelerating violently on AI. The FY24 trough ($1,359M, down 23% YoY) was telecom-inventory destocking; FY25 recovered +21%; the 9-month FY26 run-rate (~$2.0B in three quarters) annualizes well above any prior full year.

By geography (Q3 FY26 ship-to, %): US 23.0, Thailand 21.9, Hong Kong 16.8, Mexico 12.8, China 8.7, Other-APAC 6.5, EMEA 6.5, Japan 3.6. Asia-Pacific total 57.5%. Reminder: ship-to ≠ end-demand.

Lens 5 · Earnings Result (latest print — fiscal Q3 FY26, reported 2026-05-06)

GAAP, from the filing:

• Net revenue $808.4M, +90.1% YoY (vs $425.2M), a record.
• Gross profit $357.0M; GAAP gross margin 44.2% vs 28.8% a year ago (+1,540bps).
• Income from operations $174.5M (GAAP op margin 21.6%) vs a $(37.7)M loss.
• Net income $144.2M vs $(44.1)M loss. GAAP diluted EPS $1.50 (basic $1.99) vs $(0.64).
• R&D fell to 11.2% of revenue (from 17.9%) and SG&A to 11.2% (from 26.3%) — operating leverage is the whole margin story: opex barely moved in dollars while revenue nearly doubled.

Non-GAAP, from the earnings release:

• Non-GAAP gross margin 47.9%, non-GAAP operating margin 32.2%, non-GAAP diluted EPS $2.37. Gross margin +540bps QoQ, operating margin +700bps QoQ.

Guidance / outlook (Q4 FY26):

• Revenue $960M–$1.01B (midpoint $985M = another record), non-GAAP op margin 35.0–36.0%, non-GAAP diluted EPS $2.85–$3.05. Management framed >85% YoY growth. The sequential margin guide (32.2% → ~35.5% op margin) says leverage is still building.

Balance-sheet flags:

• Cash $2,617.8M (up from $520.7M at FY25 end) — the jump is the $2B NVIDIA preferred + the $1.265B 2032 convert (Lens 9/10).
• Inventories $632.8M (up from $470.1M) — building ahead of the ramp; watch this vs revenue (it's growing slower than revenue, which is healthy).
• Accounts receivable $441.6M (up from $250.0M) — tracking revenue.
• Current portion of long-term debt $3,238.6M — the convertibles reclassified to current because the stock is >130% of conversion price, so holders can convert (cash-settle principal). This is the single biggest balance-sheet flag — see Lens 10.
• 9-month operating cash flow context: net income $226.6M + D&A ~$193M + SBC $129.4M; the company is now self-funding operations comfortably.

Market reaction / what was priced in: the stock fell ~5.3% on the NVIDIA $2B announcement (dilution + "sell the news"), yet is up ~7% on other AI-order/Nasdaq-100 headlines. Net: the tape now reacts less to the print (beats are expected) and more to customer/architecture signals and index flows.

Lens 6 · Earnings Calls (sentiment trend)

No transcripts on the research layer (transcripts/ empty) — this lens is.

• Tone trajectory (last ~3 calls, Q1→Q3 FY26): decisively more confident. Management moved from "recovery + AI inflection" language to "sold-out capacity," "record," and raising the bar every quarter (Q2 FY26 rev $665.5M → Q3 $808.4M → Q4 guide ~$985M).
• Recurring phrases now: "200G per lane," "1.6T," "optical circuit switch / OCS," "co-packaged optics / CPO," "capacity expansion," "US-based manufacturing," "NVIDIA partnership." The narrative has fully pivoted from telecom/3D-sensing to AI optics.
• What they stopped saying: the FY24-era language of "customer inventory normalization," "underutilized manufacturing capacity," and demand "fluctuations" has largely dropped out — the destocking cycle is over [contrast research-layer: filings/10-k-2025-q2.md MD&A, which still discussed FY24 normalization].
• New disclosures volunteered: OCS backlog ">$400M," an "incremental multi-hundred-million-dollar CPO order deliverable 1H CY2027," and the Greensboro fab. Management is signaling a multi-year, multi-product runway, not a one-quarter spike.

Lens 7 · Comps

Peers: the closest pure comps are Coherent (COHR) and Fabrinet (FN); module-leader InnoLight (SZSE: 300308) and silicon/DSP adjacents Marvell/Broadcom are context. Multiples are ****, dated; where I could not source a clean figure I mark n/a. Do not treat these as precise — they are point-in-time pulls from aggregators (GuruFocus, stockanalysis, MacroTrends) in early June 2026.

Sources:.

Read: the entire optical-AI complex is trading at ~44–52x forward earnings and EV/EBITDA multiples 3x+ their own historical medians — these are priced as secular-growth semis, not cyclical optics. LITE sits at the top of the peer P/E range (~52x fwd vs COHR ~44–49x, FN ~44x). TTM P/E (~159x) is meaningless given the loss-year base — forward is the only honest lens. The premium is defensible only if Lumentum's laser-chip mix gives it higher sustainable margins/growth than module-centric peers; that is the crux of the whole call.

Lens 8 · Stock-Price Catalysts (moves >5%, last ~5 years) —

The five-year tape splits into three regimes:

1. 2021–2022 — consolidation + 3D-sensing peak, then de-rate. The NeoPhotonics deal (announced Nov 2021, closed Aug 2022, $918M/$16.00 a share) and Oclaro legacy made LITE the InP-laser consolidator; but the stock was still anchored to Apple-iPhone VCSEL 3D-sensing demand, which the market began to discount.
2. 2023–mid-2024 — the telecom trough. Revenue fell to the FY24 bottom ($1,359M, GAAP op loss $(434)M, EPS $(8.12)) on customer destocking and consumer-3D-sensing share loss. The stock de-rated hard. Catalysts that moved it: earnings misses/guides, Cloud Light acquisition (Nov 2023, $750M — the strategic AI pivot), and macro/rate moves.
3. Late-2024 → mid-2026 — the AI re-rating. The stock is up ~450%, "the vast majority… between late 2024 and early 2026". Discrete >5% catalysts: AI-driven earnings beats + raises (each of Q1–Q3 FY26), the NVIDIA $2B deal (Mar 2, 2026; -5.3% on the day on dilution, then higher on follow-through), S&P 500 and Nasdaq-100 inclusion (index buying), OCS/CPO order announcements, and analyst target hikes (Needham $1,040 on 2026-06-09; Northland to $1,200).

What the pattern reveals: for most of its life LITE traded on telecom-cycle inventory and Apple-3D-sensing demand. Since late 2024 it trades on AI-capex sentiment, hyperscaler/NVIDIA signals, and index flows — a higher-beta, narrative-driven regime where the marginal buyer is an AI-thematic fund. That makes it vulnerable to any wobble in the AI-capex narrative, regardless of its own print.

---

Phase C — Judge people & books

Lens 9 · Management

• CEO: Michael Hurlston — appointed President & CEO effective Feb 7, 2025, succeeding Alan Lowe (CEO since 2015, retired). This is the single most important management fact and it cuts deep: Hurlston previously ran Finisar and oversaw its sale to II-VI (now Coherent) — i.e., he built and sold Lumentum's biggest competitor's optical franchise. Before that, CEO of Synaptics (2019–2025) and 17 years at Broadcom (EVP Worldwide Sales). EE degrees + MBA from UC Davis. Track record: a proven semis operator and dealmaker who knows the datacom-optics customer set and competitive map cold. The board hired a closer/operator precisely as the AI window opened — a deliberate, high-conviction choice.
• Tenure & skin in the game: Hurlston is ~16 months in. Insider ownership is modest (professional-manager profile, not founder). Insider activity is a yellow flag: president Wupen Yuen and director Ian Small sold/repositioned >$23M in the past three months with no insider purchases. In a stock up 450%, some selling is rational diversification, but a net-zero buy signal from insiders at ~52x forward is not a vote of value-conviction.
• Capital-allocation history: the LITE record is acquisition-led — Oclaro ($1.8B, 2018), NeoPhotonics ($918M, 2022), Cloud Light ($750M, 2023) — plus the IPG telecom-transmission product lines. It also ran a buyback up to $1.2B that expired May 2025. Verdict: the deals consolidated the InP-laser supply base that is now the moat (good strategic logic), but were funded with cheap converts and came through years of GAAP losses — value-creation is only now showing up in the P&L. ROE/ROIC on management's watch was negative FY23–FY24; the test of this team is whether the AI cycle converts the accumulated R&D and acquired IP into durable returns.
• Red flags (governance): none egregious at the issuer level. The historical insider-trading ring (Lens 10) involved a former CISO trading ahead of deals — the company was the victim, and it cooperated. No related-party or comp scandal surfaced.
• Founder vs professional manager: firmly professional manager / hired operator. Implication for this stage: good — you want a disciplined semis-supply-chain operator (capacity, yield, customer contracts, a $2B fab build) running an arms-supplier through a demand surge, not a visionary founder. The risk is that professional managers optimize for the current cycle and the NVIDIA relationship rather than long-horizon independence.

Lens 10 · Forensic Red Flags

Acting as a forensic analyst. The accounting itself looks clean (Deloitte, auditor since 2016, unqualified opinion on FY25 financials and ICFR; the only Critical Audit Matter is inventory valuation / excess-and-obsolete — appropriate for a cyclical hardware maker). The risks are in the capital structure and the quality/durability of the surge, not in fraud signals.

• The convertible stack is the dominant flag. Lumentum carries four converts: $1,050M 0.50% due 2026, $861M 0.50% due 2028, $603.7M 1.50% due 2029, and $1,265M 0.375% due 2032 (the 2032s issued Sept 2025), plus an undrawn $400M Wells Fargo senior secured revolver (Dec 2025). As of Q3 FY26 the stock traded ≥130% of every series' conversion price for the required window, so all converts are now convertible at holder option with principal cash-settled — which is why $3,238.6M reclassified to current liabilities. The 10-Q explicitly warns: mass conversion "could adversely impact our ability to continue as a going concern". Mitigant: $2.6B cash covers a lot, and in April 2026 the company exchanged ~5.7M shares of common for portions of the 2026 + 2029 Notes (equitizing the converts, leaving only small stubs outstanding). Still, the diluted share count is the live issue: it jumped from 71.5M basic to 96.2M diluted in Q3 because the if-converted converts and the NVIDIA preferred are now dilutive. Per-share math must use the fully-diluted ~96M+ base, not the ~72M basic — bulls quoting EPS off basic shares overstate the equity value materially.
• GAAP vs non-GAAP gap. Non-GAAP EPS $2.37 vs GAAP $1.50 in Q3 — a ~37% wedge. The bridge is mostly SBC ($129.4M over 9 months) and acquired-intangible amortization ($102.2M over 9 months). SBC at ~6–8% of revenue is real dilution flattered out of non-GAAP — legitimate to track, standard for the sector, but not free.
• Tax-driven FY25 "profit." FY25 GAAP net income of $25.9M (EPS $0.37) was entirely a $198.0M income-tax benefit — pre-tax was a $(172.1)M loss. The headline "first profitable year" was a tax artifact; the operating turn is real only from FY26.
• One-offs propping comparisons: FY25 included a $34.9M gain on sale of facility and the 9-month FY26 included a $27.5M escrow settlement in other income. Strip these for clean run-rate.
• Cash flow vs earnings: broadly converges — 9-month net income $226.6M against strong non-cash add-backs; no obvious receivables/inventory blow-out (both grew slower than the 72% revenue growth). Healthy.
• Inventory is the auditor's flagged risk and the right thing to watch into a capacity ramp: if AI demand hiccups, excess-and-obsolete write-downs hit a business that just built inventory to $633M.

Regulatory findings (required sub-section).

• SEC LR / AAER naming the issuer: None. Per the pre-fetched file, "No LR found" and "No AAER found" for Lumentum across 2021-06-18→2026-06-18 via EDGAR EFTS.
• SEC enforcement adjacent to the company (web): there is a real SEC matter, but Lumentum was the victim, not the respondent — the agency charged Amit Bhardwaj, former Lumentum CISO, and associates (Patel, Kakkera, Saeedi, Chitor) with insider trading ahead of Lumentum's planned acquisitions of Coherent and NeoPhotonics, >$5.2M illicit profit (filed 2022); downstream tippees (e.g., Umapathi Kakkera, Oct 2024) and a former employee (Andre Wong, final judgment Jan 2025) settled. Implication for LITE shareholders: none adverse — it reflects on individuals who misused MNPI, not on the issuer's accounting or controls; if anything it shows the deals were tightly held inside-information events.
• Export-control settlement: Lumentum recorded a $7.8M settlement for export-control violations (charge taken Q2 FY23, "Litigation Matters," pending court approval). Small dollar, but confirms the export-control risk in the risk factors is not hypothetical.
• 10-K Item 3 (Legal Proceedings): the company's own posture is the standard "subject from time to time to various legal proceedings and claims that arise in or outside the ordinary course," with patent-infringement / IP litigation called out as a recurring industry risk (incl. patent-troll exposure) — no single material proceeding quantified as company-threatening in the FY25 10-K.
• Net: No material regulatory or accounting-enforcement finding against the issuer — verified via SEC EDGAR EFTS (LR, AAER), web search, and 10-K Item 3 as of 2026-06-18. The only "findings" are (a) a historical insider-trading ring victimizing the company and (b) a small export-control settlement.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection (EPS, next three fiscal years) —, inputs labeled

Base building blocks (all per-share figures on the fully-diluted ~96M share base, which is the honest count post-converts/preferred):

• FY2026 (ending ~June 2026): 9-month GAAP EPS (diluted) $2.59 + Q4 guide. Non-GAAP: 9-month non-GAAP EPS not cleanly on the research layer; Q4 non-GAAP guide $2.85–$3.05. FY26 non-GAAP EPS ≈ $9.0–$9.5. Note the consensus conflict: one aggregator showed "2026 EPS $2.36" (a stale/GAAP or single-quarter figure) against "non-GAAP ~$8" and a Q3 print of $2.37 — I treat the non-GAAP ~$9 annualized as the operative number and flag the discrepancy rather than average them.
• Drivers: AI-optics TAM "low-teens $B with mid-teens CAGR"; Lumentum growing well above market on EML share (~50–60%) + the 200G→400G/lane roadmap + OCS + CPO (sole-source NVIDIA, 1H CY2027 deliveries). Operating leverage is the swing factor — op margin guided to 35–36% and likely still climbing.

Three-year non-GAAP EPS path, fully-diluted:

The honest takeaway: at ~$970 the stock is ~52x FY26 / ~37x base-FY27 / ~30x base-FY28 non-GAAP EPS. The multiple compresses fast if the base/bull revenue path holds — i.e., the bull case is "earnings grow into the multiple in 18–24 months." The bear case is not a profit collapse (the franchise is real) but a growth deceleration that de-rates a 52x stock to 25–30x, which alone is a ~40% drawdown.

Per --watchlist rules, no Brier forecast logged (forecast.ts create skipped in breadth mode). A natural forecast to log later: "LITE FY27 non-GAAP EPS ≥ $13, resolves ~2027-08-15."

Lens 12 · Bull vs Bear

Bull case. Lumentum is the toll-booth on AI optics at the hardest-to-replicate layer — the laser chip. It owns ~50–60% of 200G/lane EMLs, eight of which go into every 1.6T transceiver; the roadmap to 400G/lane and 3.2T is its to lose. NVIDIA validated this by paying $2B for preferred + capacity rights + a multi-billion purchase commitment and sole-sourcing CPO to Lumentum for next-gen switches — the strongest possible third-party signal, and a multi-year demand floor. Two new product lanes (OCS, backlog >$400M; CPO, 1H CY2027) extend the runway beyond transceivers. Operating leverage is ferocious: revenue +90% with opex roughly flat drove op margin from negative to 32% non-GAAP in a year, guided to 36%. A proven datacom-optics operator (Hurlston) runs it. If AI-capex compounds, FY27–28 EPS grows into the multiple and the stock works.

Bear case (permanent-impairment lens).

1. Customer + architecture concentration. Two customers = 38% of Q3 revenue (26% + 12%), 40% over 9 months. A single hyperscaler's capex pause, second-sourcing decision, or a pivot away from pluggable optics toward an architecture Lumentum doesn't lead would gut the growth.
2. The moat is a supply-scarcity moat, not a forever moat. Today's pricing power exists because EML capacity is sold out. Coherent, Chinese entrants, and the hyperscalers' own efforts are racing to add 200G EML capacity. When supply catches demand, ASPs and margins normalize — exactly the cyclicality that produced the FY24 trough.
3. Valuation = priced for perfection. ~52x forward (top of the peer range), TTM ~159x, vs a 5-yr median ~36x. The stock already discounts the base/bull path; the bear doesn't need a recession, just deceleration.

Pre-mortem (18 months out, thesis broke — what happened?): AI-training capex growth slowed in CY27 as hyperscalers digested 2025–26 buildouts; 1.6T transceiver demand pushed out two quarters; EML supply from Coherent + Chinese vendors arrived, so Lumentum's ASPs fell and gross margin slipped from ~48% toward ~40%; NVIDIA, having secured capacity, dual-sourced the next node; the converts that didn't equitize plus the NVIDIA preferred left ~100M diluted shares; and a 52x stock re-rated to ~28x on a flat-to-down EPS — a 45%+ drawdown despite the business still being fine.

Are multiples too high? Relative to history, yes; relative to the growth, only if growth disappoints. This is the entire tension. At 30x base-FY28 EPS the stock is not absurd for a structural AI-optics leader; at 52x FY26 it has no margin of safety.

Contrarian view (what the market refuses to see): the bulls treat Lumentum as "NVIDIA's optics partner" and price it like a captive winner — but the $2B preferred + capacity rights is as much a leash as a gift. NVIDIA now has board-adjacent economics, a purchase commitment that ties Lumentum's capacity to NVIDIA's roadmap, and the option to direct where the next fab's output goes. Lumentum may be trading independence and pricing power for demand certainty — great for two years, potentially margin-capping thereafter. The market is pricing the certainty and ignoring the leash.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case.

• What structurally breaks the model: a shift in AI networking architecture away from where Lumentum leads. If co-packaged optics or linear-drive/LPO displaces pluggable transceivers faster than expected, or if silicon photonics (integrating the laser differently) erodes the discrete-EML value, the eight-EMLs-per-module math weakens. Lumentum is in CPO, but CPO also reduces the chip count per port over time.
• Revenue concentration: 26% in one customer, 38% in two. The NVIDIA relationship is simultaneously the bull thesis and the single point of failure. If NVIDIA's own optics strategy (it is investing across the stack, incl. Coherent — "$4B photonics play, Lumentum vs Coherent" ) shifts share to a second source, Lumentum's premium evaporates.
• Why the moat is weaker than bulls think: it is a capacity + process-yield lead, not a patent monopoly. Compound-semi laser fabs can be built (that's literally what the $2B funds); Coherent has the Finisar/II-VI base; China is investing aggressively. A 50–60% share in a sold-out market is the most fragile kind of share — it only goes down as the market adds capacity.
• Most dangerous competitor bulls underestimate: Coherent (COHR) — bigger ($70B), broader, and run by people who… used to run alongside Hurlston. And InnoLight, which already owns >50% of NVIDIA's module wallet and is integrating backward.
• Worst capital-allocation / incentive risks: years of GAAP losses funded by serial converts; the converts are now an in-the-money dilution overhang the company is scrambling to equitize (5.7M shares issued in April 2026 ). The NVIDIA preferred dilutes another ~2.88M as-converted shares at $695.31 — below the current ~$970 price, so NVIDIA is already in the money on a sweetheart entry. Insiders are selling >$23M with zero buys.
• Assumptions that must hold for today's price: (1) AI-capex compounds through CY27–28 without a digestion air-pocket; (2) Lumentum holds ~50%+ EML share as capacity floods in; (3) gross margin holds ~46–48%+; (4) the OCS/CPO lanes inflect on schedule; (5) no NVIDIA second-sourcing. All five.
• If growth disappoints 20–30%: a 52x-forward stock on a 25% revenue miss to the base case likely halves — multiple compression (52x→~28x) times lower EPS.
• Single scenario that permanently impairs: an architecture transition (CPO/LPO/silicon-photonics) that structurally cuts discrete-EML content per port at the same time a hyperscaler second-sources — turning Lumentum from scarce-supplier-with-pricing-power back into a cyclical optics vendor. Plausibility: moderate, and rising with time — not a 2026 risk, a 2027–28 risk.

Lens 14 · Management Questions (ordered by information value)

1. As 200G/lane EML capacity is added across the industry (including the fab NVIDIA is funding), what is your internal assumption for EML ASP and gross-margin trajectory through CY2027, and at what share level do you expect pricing power to normalize?
2. Quantify the NVIDIA arrangement: what are the purchase-commitment dollars and duration, what capacity is contractually reserved for NVIDIA, and what governance/information rights did the Series A preferred convey?
3. With CPO reducing discrete-optics content per port over time, does co-packaged optics expand or cannibalize your dollar content per switch — and what is the net EML-equivalent demand under full CPO adoption?
4. What share of Q4-FY26 revenue is the top one and top two customers, and where do you expect that concentration to be in 12 and 24 months?
5. How much of the OCS >$400M backlog and the multi-hundred-million CPO order is firm/non-cancellable vs forecast, and what are the revenue-recognition timelines?
6. What is the fully-diluted share count you are managing to after the April 2026 note exchanges and the NVIDIA preferred — and is further convert equitization planned?
7. What are the return-on-invested-capital targets for the Greensboro fab and the broader US-manufacturing buildout, and over what horizon?
8. How exposed is the cost structure to tariffs and export controls given Thailand/China manufacturing and contract-manufacturer sole-sourcing — quantify the gross-margin sensitivity.
9. Where are you on the 400G/lane and 3.2T roadmap relative to Coherent and Chinese competitors, and what protects the lead beyond capacity?
10. What is the plan for the 2028 Notes ($861M, 0.50%) at/before maturity given they're now in-the-money — cash, refinance, or equitize?
11. Industrial Tech has shrunk three years running — keep, divest, or harvest?
12. What single architecture shift (LPO, silicon photonics, optical I/O) most threatens your discrete-laser content, and how are you hedged?
13. How do you think about capital returns now that you're cash-generative and the prior buyback expired — given the dilution overhang, is a buyback even appropriate?
14. What would have to be true for you to walk away from a hyperscaler's terms (MFN/exclusivity) to protect margin?
15. Beyond NVIDIA, which two customer wins would most de-risk the concentration, and what's the pipeline there?

---