For a decade after spinning out of JDS Uniphase, Lumentum was a mid-cap optics company whose fortunes rose and fell with telecom capex and Apple’s 3D-sensing orders. Then the AI build-out turned the lasers inside data-center transceivers into one of the scarcest inputs in computing — and within a year the stock rose roughly elevenfold, NVIDIA wrote a $2 billion check, and Lumentum joined the Nasdaq-100.

In Q3 FY2026 (ended March 28, 2026) Lumentum reported record net revenue of $808.4M, up roughly 90% year-over-year, with non-GAAP gross margin of 47.9%and GAAP net income of $144.2M ($1.50 diluted)[1]. On March 2, 2026 NVIDIA announced a strategic partnership and a $2B equity investment plus a multi-billion-dollar purchase commitment[2]. At ~$863 on June 5, 2026 the stock carried a ~$67B market cap and a trailing P/E near 164[3]. The question is not whether AI demand is real — it is whether this is a durable franchise or a cyclical spike on a historically lumpy base. This study lays out both cases.

Four quarters of acceleration

The clearest picture of the story is the quarterly revenue line: from $480.7M in Q4 FY2025 to a record $808.4M in Q3 FY2026, with Q4 FY2026 guided to a midpoint near $985M[12][1][22]. The slope is steep — and the debate is whether it is the front edge of a multi-year AI super-cycle or a supply-constrained spike.

Q4 FY2026 is the guidance midpoint, not a reported figure[22].

The three questions this case study turns on

The balance of evidence, at a glance

Lumentum carries telecom-bubble heritage (Uniphase/JDS Fitel → JDS Uniphase, 1999) and was spun out of JDSU in August 2015[4]. Its history is a string of cyclical swings and acquisitions — Oclaro, NeoPhotonics, Cloud Light — punctuated by a failed Coherent bid and a brutal FY2024 trough, before the AI build-out reset the story[5][6].

Lumentum (“lumen” for light) is the descendant of two 1980s optical pioneers that merged into JDS Uniphase in 1999 — a symbol of the telecom bubble. In August 2015 JDSU split in two: the components business became Lumentum, and the test-and-measurement parent became Viavi Solutions[4]. For its first years as a standalone company, Lumentum was defined as much by 3D sensing — where Apple was roughly 30% of revenue at the peak — as by telecom lasers[5].

The company grew by acquisition: Oclaro (~$1.8B, 2018), a failed 2021 bid for Coherent, Inc. that still paid Lumentum a $218M break fee, then NeoPhotonics (2022) and Cloud Light (~$750M, 2023) to re-enter datacom modules[5]. That datacom re-entry is precisely what positioned it for the AI wave — but the path there ran through a painful FY2024.

A timeline of the swings

CEO commentary, FY2025 results[30].

LightCounting projects the total optical-transceiver market to exceed $23B in 2025, roughly 50% above 2024, with Ethernet transceivers alone near $17B (+60%)[7]. The choke point is the laser chip: the EML market — under $200M in revenue but made at scale by fewer than five firms — gates every high-speed module, and AI demand is pushing combined EML/CW-DFB capacity toward ~50.7M units/month in 2026[8][9].

Every AI server cluster needs vast numbers of optical interconnects to move data between GPUs, switches and racks. That demand has roughly doubled the optical-transceiver market in a year, with LightCounting projecting it to exceed $23B in 2025 — up ~50% — led by Ethernet transceivers near $17B[7]. Inside each high-speed module sits an externally-modulated laser (EML) or a CW-DFB pump laser. That chip layer is far smaller in dollars than the module market — the EML chip market was about $182M in 2024, projected to ~$352M by 2032[8] — but it is the constraint everything else depends on.

Boom — or looming overcapacity?

The demand is not in dispute; its durability and the supply response are. Bulls see a multi-year structural ramp in which the scarce input is exactly what Lumentum makes. Bears note that the same incentives that pull Lumentum, Coherent and Chinese second-tier players to add capacity also set up a classic overshoot, with module pricing likely to come under pressure as 1.6T volume ramps in 2026-2027[41].

Lumentum reports two segments. In FY2025, Cloud & Networking (Telecom + Datacom) rose ~30% to $1,410.8M — about 88% of revenue — while Industrial Tech (legacy lasers + 3D sensing) fell 14.6% to $234.2M[10]. By product, Q3 FY2026 split into Components $533.3M (+77%) and Systems$275.1M (+121%)[11]. The model is vertically integrated — Lumentum grows its own indium-phosphide and makes the lasers inside its modules — which is the source of both its margin and its concentration risk.

The revenue mix, two ways

Cloud & Networking includes both Telecom and Datacom; Components (lasers, ROADMs) vs Systems (transceiver modules) is the product cut[10][11].

How Lumentum makes money

• Components — EML, pump and narrow-linewidth lasers, ROADMs and optical chips sold to module makers and system OEMs. The scarce, high-margin layer; +77% YoY in Q3 FY2026, led by narrow-linewidth and pump lasers[11].
• Systems — finished 800G/1.6T cloud transceivers (the re-entry that Cloud Light enabled) plus optical circuit switches. +121% YoY in Q3 FY2026, cloud transceivers up ~40% sequentially[11].
• Industrial Tech — fiber and ultrafast lasers for manufacturing plus 3D-sensing (consumer) optics. The shrinking legacy: −14.6% to $234.2M in FY2025[10].

Vertical integration: moat or concentration?

Owning the laser chip is what lets Lumentum capture margin a pure assembler cannot — and why NVIDIA pre-paid for its capacity. But the same model concentrates the business in one segment sold to a few buyers.

The five largest transceiver suppliers — Coherent, Lumentum, Broadcom, Accelink and InnoLight — captured ~50% of 2025 revenue, with Chinese assembler InnoLight reaching #2 for the first time[13]. InnoLight and Eoptolink together took an estimated ~60% of NVIDIA’s incremental 800G orders, leaving Coherent and Lumentum the remaining ~40% — favored for vertically integrated laser supply[14]. The twist: the EMLs and DSPs inside those Chinese modules still come from Lumentum, Coherent, Broadcom and Marvell[15].

Where the value pools: chips vs. modules

Split is a PhotonCap/LightCounting estimate of incremental volume, not audited share[14].

Positioning: who owns what layer

Mapping rivals by vertical integration (do they own the laser chip?) against AI-datacenter leading edge places Lumentum near Coherent in the “owns the chip, leading-edge” quadrant — but with a narrower portfolio and a heavier telecom/3D-sensing legacy.

Five forces on the optics value chain

The structural pressures are real on every side: concentrated, powerful buyers; fragmenting global rivalry; and architecture shifts that could reroute where value sits. Click each force.

The bull case is concentrated in one number: an estimated ~50-60%share of the advanced 200G-per-lane EML chips inside every 1.6T transceiver — a chokepoint NVIDIA pre-allocated, pushing rivals’ lead times past 2027[16][17]. The bear case is also one history: this is the same company whose revenue collapsed 23% to a $1.36B FY2024 trough when telecom OEMs (Ciena, Cisco) destocked and Apple 3D-sensing demand fell, with margins bottoming at 30.2%[18][19]. Both are true at once.

The bull pillar: the EML chokepoint

Share is a PhotonCap / Global Equity Briefing estimate — directional, not audited[16][17].

Executive framing of the partnership[31].

The bear pillar: the legacy has crashed before

The same business that is now supply-constrained was, two years ago, drowning in customers’ excess inventory. For years Apple was the single-largest customer at >25% of revenue; revenue fell from $1.77B in FY2023 to $1.36B in FY2024 (−23%) as Ciena and Cisco sat on telecom stockpiles and 3D-sensing softened, dragging non-GAAP gross margin to a 30.2% trough[18][19]. Hyperscaler AI capex is itself cyclical; the question is whether the customer set has changed enough to break the pattern.

Weighing the two

Annual revenue ran $1.77B (FY2023) → $1.36B (FY2024 trough) → $1.65B (FY2025), then quarterly revenue accelerated to $665.5M (Q2 FY2026, +65.5%) and a record $808.4M (Q3 FY2026, +90%)[21][1]. Q4 FY2026 is guided to $960M-$1.01B with non-GAAP EPS $2.85-$3.05 — implying a ~$3.7B annualized run-rate[22]. Non-GAAP gross margin climbed from a 30.2% FY2024 trough to 47.9% in Q3 FY2026[19][1].

Annual revenue: the V, then the breakout

FY26E (~$3.4B) is an estimate built from reported H1 FY2026 + the Q3 actual + the Q4 guidance midpoint — not a reported full-year figure[21][22].

Gross margin: out of the trough

GAAP gross margin troughed near 16.2% in Q3 FY2024; management targets a long-term 45-50% non-GAAP range, though the recovery rests on a premium 1.6T/OCS mix in a supply-constrained window[42][19].

Financing the ramp: the NVIDIA cash and Project Vanguard

On March 2, 2026 Lumentum sold 2,876,415 shares of new Series A Convertible Preferred Stock to NVIDIA at $695.31/share, raising $2B in a private placement[23]. That capital partly funds “Project Vanguard,” a $3.8B capex plan (Greensboro expansion plus a second US indium-phosphide fab) — which the market did not greet warmly.

Among optics peers Lumentum is the smallest by revenue ($1.65B FY2025) yet trades at a ~$67B market cap, second only to Coherent ($5.81B revenue, ~$73.8B cap)[26][3]. Consensus is Buy with an average target near $1,105 (range $600-$1,400)[25], and LITE joined the Nasdaq-100 in May 2026 after an ~1,100% one-year run[24]— but bears note the >90x forward multiple is one usually reserved for software, not hardware[38].

Revenue: Lumentum is the small one

Fiscal years differ; Broadcom (~$63.9B) is omitted from the bar so it doesn’t flatten the others[26].

Market cap: repriced toward the top

Mega-caps Broadcom (~$1.8T) and Marvell (~$230B) are shown in the table below, not the bar[3].

The peer table

Revenues are most-recent fiscal year; caps are point-in-time (~June 5, 2026); fiscal years differ[3][26].

The bear case is not that AI demand is fake — it is that nearly everything has to go right at a price that leaves no margin of safety. At ~99x forward normalized EPS the stock prices a durable super-cycle[27]; one independent thesis assigns a 20-25% probabilityto a $450-600 bear case (27-45% downside) on a hyperscaler capex pause, an EML quality issue, or a Chinese breakthrough[40].

SWOT

Bracketed ids reference the per-claim sources listed in the Sources section.

The four risks bears press hardest

1 · Architecture shift — CPO and LPO

Co-packaged optics moves the optical engine onto the switch package and could make pluggable transceivers redundant at the highest bandwidths; it is ~0.5% of AI-data-center modules in 2026 but projected to ~35% by 2030 (TrendForce)[35]. Linear pluggable optics (LPO) strips the DSP out of the module, cutting 800G power from 14-17W to 7-8.5W (~40-50%)[39]. Lumentum sells CPO and OCS as opportunities it is winning — but if the architecture shifts faster than its design wins, the merchant pluggable pool thins[20].

2 · Customer concentration

~86-88% of revenue is Cloud & Networking sold to Google, Amazon, Microsoft and Meta, with the OCS book “almost entirely dependent on Google”; a single hyperscaler capex pause or qualification failure creates a disproportionately large revenue impact[36][10].

3 · Chinese localization

Chinese assemblers already take ~60% of incremental 800G module volume and are pushing into in-house EMLs; today their 25G+ chip localization is below ~15%, but a faster-than-expected close on 100G/200G EMLs would erode the moat[14][37].

4 · Capex and valuation

The $3.8B Project Vanguard fab build (full capacity ~late 2028) drew an 11% one-day sell-off on “capex trap”/overcapacity fears and could weigh on free cash flow[34][38]. The 2030 DCF in one bear write-up shows only ~93% upside even under generous assumptions (45% CAGR, 40% margins, 40x exit), implying much may already be priced in[28].

The counter to the bears