Layer 4 · LITE — Lumentum Holdings
One-line thesis
Lumentum is the under-shipped hyperscaler optics name with a +90% YoY record quarter, Nasdaq-100 inclusion bid layered on top — own it as the photonics-tightness play with a passive-flow tailwind into a structurally short market, sized for the 1.6T transition through 2027.
What Lumentum physically does
Lumentum manufactures the optical components that move data inside and between AI data centres — laser diodes (VCSELs and EELs), receivers (photodiodes), and the integrated transceiver modules that plug into the back of every AI server's networking port. Inside a frontier training cluster, every node-to-node connection runs over fibre at 400G, 800G or 1.6T optical speeds, and the bidirectional optical link at each end of every fibre is a transceiver. Lumentum sells the silicon photonics and III-V semiconductor components that go inside those transceivers, and increasingly the fully-assembled transceiver modules themselves.
The AI-stack layer is Layer 4 — interconnect, photonic — and it is the second binding constraint after memory in current-cycle AI build-outs. As GPU clusters scale past a few thousand nodes, the back-end fabric becomes the bottleneck, and the back-end fabric is optical. The migration from 400G to 800G to 1.6T transceivers — driven by Nvidia, Broadcom, Marvell and AMD silicon roadmaps — is structurally pulling Lumentum's content per port from $200-300 to $600-1,200, and the unit volume is exploding alongside.
Lumentum's competitive position inside this market is anchored by two product lines: indium-phosphide laser diodes at the highest-speed nodes (where Lumentum has historic technology leadership against AAOI, Source Photonics and Inphi) and the integrated transceiver assembly business that consolidates the III-V components with packaging and test. The recent Cloud Light acquisition added scale in datacom transceivers and is now contributing meaningfully to the revenue line.
The physics of why indium phosphide matters at the highest speeds is worth understanding. Silicon photonics can serve 400G and most 800G applications with modulator-based architectures, but the highest-density 1.6T and beyond transceivers require external modulated lasers (EMLs) built on indium-phosphide substrate rather than silicon. Lumentum is one of three suppliers worldwide with the InP epitaxy capacity and process know-how to produce these lasers at scale — Coherent and a handful of Chinese vendors are the others. The InP supply chain itself is a bottleneck (which is why AXT, the InP substrate vendor, has been a separate AI-photonics trade); Lumentum's vertical integration into InP gives it both supply security and a margin advantage at the highest-speed nodes.
The financial print
Q3 FY26 (reported May 7) printed record revenue $808.4 million, +90% YoY, EPS $2.37 beat consensus, and Q4 guide of $960 million-$1.01 billion with EPS $2.85-$3.05 — well above sell-side. Operating margin printed at 32.2%, a structural high for Lumentum and confirmation that operating leverage in the AI-datacom mix is now flowing.
The critical disclosure on the call was that Lumentum is under-shipping demand by approximately 30% — capacity is the gating constraint, not demand. That is the optical-photonics equivalent of the HBM-sold-out story, and it is what justifies the FY27 consensus walk-up that the sell-side has been doing weekly.
The stock closed $953.90 on May 17. Nasdaq-100 inclusion announced for May 18 (replacing CSGP), which adds a passive-flow tailwind of roughly $4-6 billion of forced buying across the QQQ ecosystem in the days following inclusion. Q4 FY26 reports August 7. Median sell-side price target is $1,100 with high $1,400 and low $600.
The Q3 print mechanics deserve a closer look. Revenue acceleration from sub-$500 million quarterly run-rates two years ago to $808 million in Q3 with $960M-$1.01B guided for Q4 implies a roughly doubling of revenue inside an 18-month window. The 32.2% operating margin print is structurally higher than Lumentum's historical 18-22% through-cycle range — a step-change that reflects both the AI-datacom mix shift and the under-shipping pricing power. The combination of revenue doubling and operating margin expanding by 1,000+ basis points produces the kind of EPS compounding that is normally only seen in software-class businesses. The Q4 EPS guide of $2.85-$3.05 implies annualised run-rate EPS approaching $12, materially ahead of consensus FY27 estimates that still anchor to $9-10.
Customer mix
Hyperscaler concentration is heavy and increasing. The top four hyperscalers (AWS, Microsoft, Google, Meta) plus the AI-focused tier (Oracle, CoreWeave, x.AI) account for the bulk of Lumentum's datacom revenue. Apple is a separate large customer through the consumer-3D-sensing line, but that segment is now a fraction of the AI-datacom business. The acquired Cloud Light business deepened relationships at AWS specifically and is now the principal volume engine for the 800G transceiver line; the integration has executed cleanly and is contributing margin accretion ahead of the original deal model.
The customer book is contracted and tightening. Lumentum has disclosed multi-year supply commitments with hyperscaler customers for 800G and 1.6T transceiver production through 2028, which is the kind of visibility that justifies capex acceleration. The customer mix is heavily skewed to the AI-datacom thesis — a positive durability signal in the up-cycle and a concentrated downside if hyperscaler capex blinks.
Competitive context
The optical transceiver market is more fragmented than the memory market but is consolidating. Direct competitors include Coherent (the closest peer, larger but more diversified), AAOI (the high-velocity small-cap challenger with Amazon as anchor customer), Ciena (more diversified network-equipment vendor), Innolight (the dominant Chinese transceiver assembler), and the captive silicon-photonics teams at Broadcom and Marvell. The combined market structure is a mid-single-digit handful of credible suppliers at the highest speeds and a longer tail at the lower-speed legacy nodes. The competitive structure is therefore more accurately a tight-oligopoly at the leading edge and a fragmented commodity market at the trailing edge, which is the precondition for sustained pricing power at the leading edge.
Lumentum's competitive moat is the indium-phosphide laser-diode technology at the highest-speed nodes (1.6T and beyond), the integrated transceiver assembly capability post-Cloud Light, and the hyperscaler relationships at the architect level. Against Coherent, Lumentum runs a more focused datacom mix; against AAOI, Lumentum has scale and is qualified across more customers. The competitive context is favourable in the up-cycle and probably extends through the 1.6T transition.
The pricing dynamic in optical transceivers has flipped over the past 18 months in a way that mirrors the HBM and HDD pricing flips. Through 2023-24 transceiver pricing was negotiated quarterly with the customer holding leverage; through 2025 pricing moved to annual contracts; through 2026 multi-year LTAs are emerging at fixed-pricing escalators. Lumentum has explicitly captured this transition in its commentary. The under-shipping disclosure (shipping 70% of demand) is not just a capacity story; it is a pricing-power story, because demand-above-supply is precisely the precondition for sustained margin expansion. The Q3 FY26 print at 32% operating margin is not a peak — it is a structural step-up that should hold through the 1.6T transition.
Terminal risk
The terminal risk is silicon photonics co-packaging — the move from pluggable transceivers to optically-coupled silicon directly on the switch ASIC. If silicon photonics CPO (co-packaged optics) replaces pluggable transceivers in the 2027-29 window, the entire transceiver value chain compresses and Lumentum's pluggable-module revenue line goes structurally lower. The mitigation is that Lumentum's component-level III-V semiconductor business (laser diodes, photodiodes) is needed inside CPO modules too, so the displacement is partial rather than total — but the value capture per port is lower.
The secondary terminal risk is the Chinese transceiver assemblers (Innolight, Hisense Broadband) capturing more share at the volume tier, which compresses pricing at the 800G node before the 1.6T transition fully ramps.
A third terminal risk worth flagging is the inventory-cycle pattern in datacom optics. Historically, hyperscaler optical inventory builds for 9-12 months at a time before customers pause to absorb deployed capacity. The current cycle has been running roughly 18 months of continuous ramp without a meaningful pause, which is unusual and probably reflects the AI-cluster scale-up rather than normal capacity-refresh patterns. The risk is that hyperscalers eventually pause to digest deployed inventory, which would produce a one-to-two-quarter air pocket in transceiver orders even if the structural demand curve continues higher. That pattern would be visible in the monthly customer-mix commentary at Q4 FY26 print or the Q1 FY27 print.
Bull case
The three-to-five-year bull case sees Lumentum riding the 1.6T transition through FY28, with revenue running $5-6 billion versus FY26 run-rate closer to $3.5 billion, operating margins holding in the low-30s, and EPS in the $14-16 range. A 25-30x multiple — earned through the AI-datacom-photonics-bottleneck moat plus Nasdaq-100 passive flow — gets the stock to $1,200-1,500 over a 24-month view.
The stretch bull case adds CPO becoming a real architecture that Lumentum is the principal III-V supplier into, plus continued under-supply through 2027, plus the Apple-3D-sensing line stabilising. That path takes the stock to $1,800+ on a 36-month view.
The under-shipping disclosure deserves an additional layer of analysis. When a manufacturer publicly states that it is shipping 70% of demand, the message to customers is that capacity is being rationed and that pricing will reflect scarcity. Lumentum's competitive set — Coherent, AAOI, the Chinese assemblers — has not made equivalent disclosures, which suggests Lumentum is either uniquely positioned at the highest-speed nodes (1.6T) where capacity is structurally tightest, or is more transparent about supply-demand dynamics for strategic-customer-management reasons. Either reading is positive for through-cycle pricing power. The Nasdaq-100 inclusion combined with the supply-tightness disclosure creates a particular kind of technical setup — passive index buying meeting fundamental supply scarcity — that historically produces extended momentum runs.
Gap / bear case
What the market may be missing is the under-shipping disclosure. Lumentum is shipping 70% of what customers want — that is a structurally tight market where pricing power persists through the next 24 months. Consensus FY27 revenue at $4-4.5 billion likely understates the trajectory if capacity expansion stays disciplined.
The market is right about the +90% YoY print being a hard number to grow on. The base-effect compression in FY28 is real, and a deceleration to +30-40% YoY in FY28 is mathematically inevitable. The bear case is that the stock has priced perfect execution and any operational stumble (capacity delays, qualification slips, hyperscaler order rephasing) is punished disproportionately.
A specific consensus gap worth flagging is the operating-margin trajectory assumption. Most sell-side models assume operating margin compresses from the Q3 32% print back toward the 25-28% historical range over FY27-28 on competitive pressure and mix normalisation. The bull-case alternative is that the 32% margin holds or expands because the under-supply position persists and the 1.6T mix shift pulls ASPs higher. If operating margin holds at 32% rather than compressing to 28%, that is roughly $1.50-2.00 of additional FY27 EPS that consensus does not capture — a meaningful asymmetry on a stock trading at $953.
Optionality
Three options. First, CPO — silicon photonics co-packaged optics — is an inevitable architecture transition; Lumentum's III-V component business is positioned for it even if the pluggable-module business compresses. Second, the Apple 3D-sensing line stabilising or expanding (LiDAR-class consumer applications, AR/VR headsets) adds a non-AI revenue diversification. Third, the Nasdaq-100 inclusion is a discrete passive-flow event that creates ~$4-6 billion of mechanical buying — the price action over the 30 days post-inclusion is independent of fundamentals. A fourth option is the indium-phosphide supply chain — Lumentum's vertical integration into InP provides margin protection and could become a revenue line if InP capacity is licensed to third-party transceiver assemblers. The combination of these four optionality legs is unusual for a single name and supports the conviction sizing.
The trade
Entry: $920-980 — acceptable; aggressive add below $850. Size: 3% portfolio target. Stop: $700 (200-day MA). Catalyst date: Nasdaq-100 inclusion May 18 (today); Q4 FY26 print August 7. Trim/exit: trim 25% at $1,200, 50% at $1,400; full exit on hyperscaler capex blink or CPO displacement evidence. Conviction: 8.5/10.