Lumentum
Vertically-integrated component + systems supplier: designs and fabs InP/EML laser chips and optical transceivers (Cloud & Networking) plus telecom systems and industrial lasers; sells to hyperscalers, networking OEMs, and module assemblers. Revenue is hardware unit sales, not recurring subscription — a cyclical, capacity-gated chip business currently in an up-cycle.
The thesis on this name
State of AI Compute
Purest laser-scarcity rent with the highest GM (47.9%) and steepest growth (+90%) — but verify flagged it as a cyclical at a cyclical peak wearing a compounder label (8.4x fwd sales vs 3.4x median, self-reported >30% gap, capacity being added by the rent-holders); buy ONLY on a pullback, treat as a momentum/event trade with a hard stop on gap-narrowing.
Earnings, margins, COGS & capex
Lumentum is in a violent AI-driven up-cycle: revenue went $480.7M → $533.8M → $665.5M → $808.4M across Q4 FY25 to Q3 FY26 (+90% YoY in Q3), with non-GAAP gross margin recovering to 47.9% and non-GAAP operating margin to 32.2% as cloud-transceiver and EML laser-chip volume leverages fixed costs. Q4 FY26 guidance of $960M-$1.01B implies ~22% sequential growth and >85% YoY, with non-GAAP op margin guided to 35-36% [fact]. The company is spending heavily on capex (~$125M/qtr) to expand InP wafer fab and the Thailand transceiver plant, so reported FCF is still thin-to-negative even as P&L margins inflect [fact]. The June-2025 self-reported >30% premium-to-median multiple plus added capacity is the cyclical-peak tell.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~52¢ is cost of goods and ~16¢ operating expense, leaving ~32¢ of operating profit (~28¢ net).
Revenue trend
Margins
up (vs 35.2% in Q3 FY25) [fact]
up sharply (vs 10.8% Q3 FY25); guided 35-36% Q4 [fact]
up [fact]
up (vs 28.8% Q3 FY25) [fact]
depressed by capacity capex; should turn positive as revenue scales [fact/est]
COGS structure
COGS is dominated by semiconductor fab + module assembly: indium-phosphide (InP) and gallium-arsenide wafer fab depreciation and yields, EML laser-chip processing, sub-component purchases (DSPs from Broadcom/Marvell, optics, fiber, gold/precious-metal packaging), and labor at the Thailand transceiver plant and Japan EML fab. Margin is gated by InP wafer yields and mix toward high-value 200G EML / cloud transceivers; capacity being fully allocated (Japan EML fab fully booked) is what lets it hold premium pricing [fact].
Capex
~$125M in Q3 FY26 (~15% of revenue); FY25 capex was $231M (up from $136M). Funds InP wafer-fab expansion (the scarce upstream chokepoint), the Nava Nakorn Thailand transceiver facility, EML laser-chip capacity in Japan, and OCS/CPO systems capacity — i.e. buying the scarcity it currently rents [fact].
Latest earnings
Beat — revenue above the guidance midpoint and ahead of consensus; Q4 guide came in above estimates [fact].
Q4 FY26: revenue $960M-$1.01B (midpoint $985M, ~22% QoQ, >85% YoY), non-GAAP op margin 35-36%, non-GAAP EPS $2.85-$3.05. Management reiterated a $2B-quarterly-revenue ambition [fact].
- Components revenue (Q3 FY26)
- $533.3M (66% of total)
- Systems revenue (Q3 FY26)
- $275.1M (34%; +121% YoY)
- OCS backlog
- >$400M
- Cash + ST investments
- $3,172.3M (after NVIDIA $2B)
Growth drivers
- AI-datacenter optics demand — cloud transceivers (lion's share of growth) up >40% QoQ in Q3 FY26 as Thailand capacity ramps [fact]
- 200G EML laser chips — Japan fab fully allocated, shipments doubled YoY; targeting >50% EML unit growth Dec-26 vs Dec-25; 200G EML revenue +128% QoQ [fact]
- NVIDIA $2B strategic investment (Mar 2026) + multibillion-dollar component purchase commitments anchoring demand [fact]
- Optical Circuit Switch (OCS) — backlog well beyond $400M, scaling rapidly [fact]
- Co-packaged optics (CPO) — incremental multi-hundred-million-dollar order deliverable 1H CY2027 [fact]
- Telecom/systems recovery — Systems revenue $275.1M in Q3 FY26, +121% YoY [fact]
Reported financials — SEC EDGAR
Audited GAAP figures pulled from SEC filings · latest filing 2026-08-17. The audited primary-source spine — not financial advice.
Revenue — annual (GAAP)
Margins & balance sheet — FY’26
Bull & bear
Lumentum is the highest-margin, fastest-growing Western owner of the scarce AI-optics input — InP/EML laser chips — and is converting that scarcity rent into 90% revenue growth, 48% gross margins, and an NVIDIA-anchored multi-year demand book that the multiple, while rich, can grow into.
- Scarcity rent is real: Japan EML fab fully allocated, 200G EML +128% QoQ, and the upstream chip is the chokepoint the whole transceiver chain depends on [fact]
- Demand visibility is unusually concrete — NVIDIA $2B investment + purchase commitments, OCS backlog >$400M, CPO multi-hundred-M order for 1H CY2027 [fact]
- Margin inflection compounding with revenue: non-GAAP op margin 10.8% → 32.2% YoY, guided to 35-36% — operating leverage is steep [fact]
- Guidance points to >85% YoY growth into Q4 FY26 and a stated $2B/quarter revenue ambition — run-rate makes the ~17x forward-sales multiple defensible if it holds [fact/est]
- Vertical integration + Western-supplier preference gives it a structurally better position than Chinese module-only assemblers for hyperscaler-critical links [fact/est]
This is a cyclical at a cyclical peak wearing a compounder label: ~17x run-rate / ~26x TTM sales for a hardware business whose revenue is driven by a single wave of hyperscaler AI capex, with the rent-holder and the whole industry adding capacity into the very demand that justifies the price — a classic setup for both estimate and multiple compression.
- Valuation-vs-history gap: a self-reported >30% premium to the median multiple (8.4x fwd sales vs 3.4x median) for a business that historically traded low-single-digit sales multiples through prior optics cycles [fact/est]
- Capacity is the self-inflicted risk — Lumentum, Coherent, and Chinese vendors are all adding InP/transceiver capacity; the scarcity rent erodes precisely when supply catches demand [fact]
- Customer = competitor: NVIDIA, Google, and Amazon are building in-house optics; the $2B NVIDIA stake also dilutes and ties fate to one buyer's roadmap [fact]
- FCF is negative and debt is heavy (~$3.28B, mostly converts) — the growth is being funded by capex and balance-sheet leverage, not self-funding cash generation yet [fact]
- Down-cycle math is brutal: a hyperscaler capex pause would hit a name priced for perpetual 85%+ growth, where any deceleration compresses both EPS estimates and the sales multiple simultaneously [est]
What it is worth
Forward EV/Sales and P/E cross-check against optics-cycle history. On annualized Q4-FY26 run-rate (~$3.9B) the stock trades ~17x sales; ~26x TTM (~$2.49B) — vs low-single-digit sales multiples in prior optics cycles and a self-reported >30% premium to the ~3.4x peer/median.
$300-$600 (hyperscaler capex digestion + industry capacity catches demand → GM reverts and multiple compresses to mid-single-digit sales; Street low ~$600) [est]
~$850-$1,100 (~consensus ~$1,100 target; growth decelerates but stays strong, GM ~45-48%) [fact/est]
$1,100-$1,400 (Street high
assumes $2B/qtr run-rate, GM holding 48%+, OCS/CPO scaling — multiple stays premium) [fact: analyst targets]
Priced for sustained 85%+ growth and durable ~48% gross margins; the multiple only holds if the scarcity rent and AI-capex wave persist through the capacity build — otherwise it de-rates on both estimates and multiple.
SWOT
Strengths
- Owns the scarce upstream chokepoint — InP wafer fab + 200G EML laser chips — with Japan fab fully allocated, conferring pricing power that pure module assemblers lack [fact]
- Highest gross margin among Western optics peers (47.9% non-GAAP) and the steepest growth (+90% YoY) in the AI-optics cohort [fact]
- Vertical integration from laser chip → transceiver → OCS/CPO system, capturing more of the value chain than Chinese module-only players [fact]
- NVIDIA $2B strategic investment + purchase commitments — capital and a demand anchor from the dominant AI-compute platform [fact]
Weaknesses
- Structurally cyclical hardware business carrying a compounder-grade multiple (~17x run-rate / ~26x TTM sales) — peak-cycle valuation risk [est]
- Thin-to-negative free cash flow while capex runs ~15% of revenue to fund capacity [fact]
- Heavy convertible-debt load (~$3.28B total debt) and equity dilution from NVIDIA stake/converts [fact]
- Customer concentration in a handful of hyperscalers and NVIDIA; order timing is lumpy [fact/est]
Opportunities
- 1.6T transceivers and 200G/lane EMLs as next AI-cluster speed node — secular content growth per GPU [fact]
- OCS and co-packaged optics (CPO) as new systems categories with multi-hundred-million-dollar early orders [fact]
- Telecom/networking recovery layering on top of the cloud cycle (Systems +121% YoY) [fact]
- Western-supplier preference / supply-diversification away from Chinese module vendors for hyperscaler-critical optics [fact/est]
Threats
- Demand air-pocket if hyperscaler AI capex digestion arrives — the classic optics down-cycle that a high multiple cannot survive [est]
- Hyperscalers (Google, Amazon) and NVIDIA developing in-house optical interconnect, turning customers into competitors [fact]
- Chinese module leaders (InnoLight, Eoptolink) dominate 800G/1.6T volume (~60% share) and could move upstream into chips [fact]
- Capacity being added across the whole industry — Coherent, Chinese vendors, and Lumentum itself — risks over-supply and margin reversion [fact/est]
Moats, dependencies & bottlenecks
Moats
InP/EML laser-chip fab scarcity (vertical integration to the scarce input) real today (Japan fab fully allocated) but the whole industry is racing to add InP capacity; rent erodes as supply catches up The single most important moat and the single biggest bear risk — it is the rent being competed away [fact]
leading-edge node lead is a few quarters, not years; Chinese vendors closing the gap on modules [fact]
powerful demand anchor but single-buyer dependence; NVIDIA dual-sources with Coherent and builds in-house [fact]
Scale + Western-supplier qualification for hyperscaler-critical links qualification is sticky but hyperscalers actively multi-source and reshore [fact/est]
qualification cycles create some lock-in per design win, but each new speed node re-opens the socket [est]
Dependencies
The dominant revenue driver; a capex digestion phase is the central downside [fact]
Demand anchor and capital source, but concentration + customer-could-go-in-house risk [fact]
Self-supplied chokepoint; capex-dependent, yield-gated [fact]
Marvell) and packaging materials Transceivers need merchant DSPs; gold/precious-metal and optics inputs in COGS [fact/est]
Cloud-transceiver growth hinges on this facility's yield and scaling [fact]
Advantages
- Highest Western-peer gross margin (47.9% non-GAAP) and steepest growth (+90% YoY) in the AI-optics cohort [fact]
- Vertical integration from laser chip to OCS/CPO system — captures more value than module-only assemblers [fact]
- Owns the scarce upstream EML/InP chip that the whole transceiver chain consumes [fact]
- NVIDIA $2B strategic investment + purchase commitments providing capital and demand visibility [fact]
- New systems categories (OCS >$400M backlog, CPO orders) extending TAM beyond modules [fact]
Weaknesses
- Cyclical hardware economics dressed in a compounder multiple — high de-rating risk [est]
- Negative/thin free cash flow during the capex ramp [fact]
- ~$3.28B debt load (mostly converts) plus dilution from NVIDIA stake/converts [fact]
- Concentration in a few hyperscalers + NVIDIA; lumpy, order-timing-sensitive revenue [fact/est]
- Chinese module leaders own 800G/1.6T volume share (~60%) and could move upstream into chips [fact]
- Whole-industry capacity additions (incl. its own) threaten the scarcity premium it is priced on [fact/est]
Bottlenecks
- InP wafer-fab capacity and yields — the binding constraint on EML/laser-chip output and the source of the scarcity rent [fact]
- EML laser-chip capacity in Japan, currently fully allocated — limits near-term upside and is being expanded via capex [fact]
- Thailand transceiver-plant ramp/yield as the gate on cloud-transceiver volume [fact]
- Capital intensity / FCF — capex ~15% of revenue funded partly by converts and the NVIDIA raise, not self-funding cash yet [fact]
- Merchant DSP availability (Broadcom/Marvell) for the highest-speed modules [fact/est]
Top signals & trends
Top signals
The demand source; any capex-pause language is the first crack in the thesis [fact/est]
Watch for the supply-catches-demand inflection that ends the rent [fact]
Holding/expanding GM is the proof the scarcity rent persists; reversion is the bear's confirmation [fact]
New-category revenue de-risks the modules-only cyclicality [fact]
The self-reported >30% premium-to-median is the valuation tell [fact/est]
FCF inflection would justify the quality re-rating; continued cash burn keeps it a leveraged cyclical [fact]
Trends
The secular wave driving 90% growth; content-per-rack rising each speed node [fact]
>$400M backlog; a new systems TAM [fact]
New order (1H CY2027) is upside, but CPO could disrupt pluggable-transceiver volume long-term [fact]
The supply response that historically ends optics up-cycles [fact/est]
Largest customers building in-house and dual-sourcing [fact]
Favors Lumentum/Coherent for hyperscaler-critical links vs Chinese module vendors [fact/est]
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Merchant DSPs / retimers for high-speed transceivers; silicon-photonics inputs [fact]
DSPs and coherent-optics electronics used in modules [fact]
Semiconductor capital-equipment & InP/GaAs wafer / materials vendors Fab tools and III-V wafer/precious-metal packaging inputs feeding the laser-chip COGS [est]
Contract assembly / Thailand (Nava Nakorn) plant inputs Optics, fiber, connectors, and labor for transceiver assembly [fact]
$2B strategic investor + multibillion-dollar component purchase commitments; anchor AI-compute customer/partner [fact]
Google/Alphabet, Amazon, Meta) Buy cloud transceivers/EMLs for AI clusters; Google/Amazon also building in-house optics [fact/est]
Networking OEM; Lumentum is one of Ciena's largest component suppliers [fact]
Networking systems customer for telecom/transport optics [fact/est]
Chinese vendors) Buy Lumentum EML/laser chips to build their own transceivers — Lumentum sells chips to the cohort it also competes with downstream [fact]
Closest Western peer — vertically-integrated optics + lasers, ~25% transceiver share, FY25 rev $5.81B; also a NVIDIA dual-source partner. The direct comp for AI-optics share and margin [fact]
Chinese module leader; InnoLight + Eoptolink ~60% of 800G and projected 50-60% of 1.6T modules — the volume threat (not a US-listed recommendation) [fact]
Chinese 800G/1.6T module leader alongside InnoLight; volume/price competitor downstream (analysis only, not a recommendation) [fact]
Merchant DSP + silicon-photonics/CPO supplier — both a key COGS supplier and a competitor as optics integrate onto the switch [fact]
DSPs + 400ZR coherent optics; competes in the electronics-for-optics layer and custom AI silicon [fact]
Acacia (Cisco-owned) coherent optics + pluggables; competitor in coherent/networking optics [fact]