--- title: "From optical interconnect veteran to all-in 'picks-and-shovels' supplier: Why LITE?" type: "Topics" locale: "en" url: "https://longbridge.com/en/dolphin/post/44059432.md" description: "The stock price surged nearly 10x within a year, pushing market cap above $70bn. It drew NVDA to invest $2bn for an equity stake in Mar 2026." datetime: "2026-09-24T10:28:06.000Z" locales: - [en](https://longbridge.com/en/dolphin/post/44059432.md) - [zh-CN](https://longbridge.com/zh-CN/dolphin/post/44059432.md) - [zh-HK](https://longbridge.com/zh-HK/dolphin/post/44059432.md) author: "[Dolphin Research](https://longbridge.com/en/dolphin.md)" generator: "portal-rs" --- # From optical interconnect veteran to all-in 'picks-and-shovels' supplier: Why LITE? $Lumentum(LITE.US) Dolphin Research noted in its optical connectivity review that regardless of where optical modules sit in the stack, optical engines with high value content and defensible moats remain firmly positioned. Their strategic importance does not change with architectural shifts. This piece starts with a veteran in the lane, Lumentum (LITE). **Its share price has surged nearly 10x over the past year, pushing market cap above $70bn, and it drew a direct $2bn investment from Nvidia in Mar 2026.** The company has moved back into the spotlight. Break down an optical link in a data center: data transmission must go through O/E conversion. Electrical signals are first 'translated' into light, carried over fiber, then 'translated' back to electrical signals. The component that does this translation is the optical module. **Inside it, the most critical, technically demanding and expensive part is the laser die that emits light — think of it as the 'bulb' of the entire link.** Lumentum sells this 'bulb', and is among the few players globally that can build from indium phosphide wafers in‑house. But selling light‑source dies alone does not explain the current re‑rating. The real shift is that the AI supercycle is rewriting pricing for optical components. Demand and pricing logic have structurally changed. Legacy optical components were a classic, cyclical business: demand tracked cloud capex and traditional data center builds, with balanced supply/demand, full competition and steady price erosion. The AI cluster build‑out upended that structure. **Massive lateral GPU interconnects unleashed exponential demand for high‑speed optics, and the move from 800G to 1.6T further lifted per‑box laser die content.** **On supply, the mismatch is stark — high‑end high‑speed laser dies (**200G and up**) face very high volume‑production barriers. Only a handful of global players can expand capacity, and expansions rely on self‑built fabs, making near‑term supply rigid and mid‑to‑long‑term ramps lengthy.** **Exploding demand with constrained capacity weakens prior cycles and shifts the industry to a supply‑constrained seller’s market.** Top capacity owners simultaneously benefit from product premia, long‑term fixed‑price contracts, and rising margins. This is why Dolphin Research tags Lumentum as a core name in AI optical interconnects. In this note, we break down Lumentum’s businesses and focus on two questions. What matters most for the growth story? **1) In an 800G/1.6T module, why are lasers the bottleneck for both value content and yield?** **2) Across Lumentum’s five product lines (EML, CW, telecom components, optical modules, OCS), where do they sit in AI clusters and how is competition shaping up?** We primarily focus on the first two product lines here. Below is the main text. We proceed section by section. **I. From discrete optics to an integrated photonics platform** 'Yi‑Zhong‑Tian' is a nickname Chinese investors use for three local optical module vendors (InnoLight, Accelink, and TFC). In the optical comms industry, however, the real veteran is **Lumentum**. Uniphase (founded 1979, commercial lasers) and JDS Fitel (1981, fiber components) merged in 1999 to form JDSU, a star in optical networking during the dot‑com era. In 2015, JDSU split into two. **The comms and data center optics business listed independently as Lumentum, while the remaining test and measurement and specialty optics business became Viavi (VIAV).** That separation set the stage for targeted growth. Post‑listing, Lumentum’s growth has leaned heavily on M&A. a) In 2018 it acquired Oclaro for about $1.8bn, **adding InP lasers and PIC capacity, the core technical base for today’s AI high‑speed optics.** b) In 2022 it completed two deals: bought NeoPhotonics to add narrow‑linewidth tunable lasers (ITLA, the core light source for coherent), high‑speed coherent PICs, and coherent ROA components (TROSA). In parallel, it acquired IPG’s telecom transmission unit, gaining coherent DSP companion ASIC self‑design capability and full coherent modules. **Together these built a full coherent transmission stack for Lumentum, enabling end‑to‑end coherent solutions for DCI, metro and long‑haul.** This broadened market coverage materially. c) In Nov 2023 it bought Cloud Light for about $750mn, a supplier of high‑speed optical modules to hyperscalers. **This marked Lumentum’s move beyond chips/components to shipping complete modules externally.** In sum, these deals **took Lumentum from a discrete optics vendor to an integrated photonics platform, spanning short‑reach AI interconnects, long‑haul coherent, and high‑speed module systems.** Its product scope now covers the full stack. **II. Product and biz. matrix breakdown** To understand Lumentum’s lines, start with the tech: a traditional pluggable high‑speed module (800G/1.6T) has two core hardware categories along the O/E chain. **They are 1) optics and 2) electrical chips.** **1) Optics: handle O/E conversion** **① Laser die** — the continuous, stable light carrier, essentially the 'light source' of the link. It is the fundamental emitter. **② Modulator** — writes the electrical data onto the light carrier, generating the optical signal. This is the encoding stage. **③ Photodiode die** — the receive‑side device that converts the optical signal back into a small electrical current. It enables downstream electronics. **2) Electrical chips: handle signal processing and gain** **④ DSP** — the module’s 'digital brain', repairing and correcting signals in the electrical domain, with relatively high power draw. It is critical for link integrity. **⑤ Driver and TIA** — the module’s 'analog muscle'. The TX driver boosts DSP output to drive the modulator; the RX TIA amplifies the photodiode’s small current into a usable voltage. Critically, optics and electricals differ fundamentally in manufacturing models and capacity elasticity. They do not scale the same way. Electrical chips like DSPs are silicon CMOS, with most vendors fabless and taping out at foundries like TSMC. While advanced nodes face capacity quotas, once secured, foundry lines can flex wafer starts, giving good capacity elasticity without self‑built fabs. Design houses focus on architecture and tapeouts. Optics follow different physics. **Because silicon does not emit light efficiently, ultra‑high‑speed lasers rely on compound semis like InP; yet, unlike CMOS, compound optical chips lack a mature, standardized third‑party foundry ecosystem akin to TSMC.** Source: Plutosemi **As a result, leading players are vertically integrated IDMs — self‑built fabs, in‑house epitaxy, wafer processing and coatings.** Their model is capital and process intensive. **Capacity adds require build‑out, debug and yield ramps over 2–3 years, far less elastic than electricals.** There are a few InP foundries that can handle small‑lot tapeouts for smaller clients, but they cannot support large‑scale, stable shipments of high‑end, high‑speed optical dies. With that optics vs. electricals context, Lumentum’s positioning is clearer. The differences drive very different supply dynamics. The company’s core is optics. **Capacity hinges on self‑built InP wafer lines, which means even if AI demand pulls modules sharply higher, Lumentum cannot lift shipments as fast near term, leaving structurally weaker expansion elasticity.** Given that backdrop, we turn to Lumentum’s product lines. This frames the business mix. Per financial disclosures, Lumentum revenue has two major segments. **1) Components**: mainly EML high‑speed lasers, CW continuous‑wave lasers, and legacy telecom optics. **2) Systems**: includes Cloud Light’s high‑speed modules and OCS optical circuit switches. It also has an industrial laser business, which is non‑comms and smaller in scale. The chart shows that **system revenue mix has been rising sequentially, driven by rapid growth in module shipments.** Mix is shifting toward systems. Because both components and systems include AI‑related and non‑AI legacy businesses, the reporting mix obscures underlying drivers. It is harder to unpack by GAAP buckets. Below, Dolphin Research departs from the reporting view and uses five core product lines — EML, CW, telecom components (these three under components), and optical modules and OCS (these two under systems). **We discuss each by technical moat, competition, and growth runway.** We start with the core base: high‑speed laser dies. The mainstream high‑speed laser architectures diverge into three routes based on how emission and modulation are integrated. Each has distinct trade‑offs. **a) EML (electro‑absorption modulated laser) — integrated, edge‑emitting:** emission and modulation are monolithically integrated on a single InP die, outputting high‑speed modulated light directly. It is the mainstream solution for high‑speed pluggable modules. **b) CW (continuous‑wave laser) — separated, edge‑emitting:** focuses only on continuous, stable, high‑power emission, with modulation handled by an external silicon photonics die. This is the core light source for silicon‑photonics and CPO packaging. **c) VCSEL (vertical‑cavity surface‑emitting laser) —** **low‑cost, surface‑emitting:** a special vertical‑emission structure enables low‑cost, arrayed mass production, but with shorter reach. Current use is mainly consumer 3D sensing and short‑range automotive LiDAR. Note that EML and CW are both edge‑emitters (EELs), requiring cleaving and facet coatings with complex steps, and significantly higher process barriers than VCSEL. That gap shows up in costs and yields. These routes differ sharply in process difficulty, ASPs, margins and competitive landscape. Dolphin Research compares them in the table. (VCSELs are mostly for consumer/industrial uses, under 10% of Lumentum revenue and not a core AI comms driver, so we do not expand here. The focus is on EML and CW lasers.) **1) EML laser dies: the Scale‑out foundation** EML is the mainstream for 800G/1.6T pluggable modules. A 1.6T module typically needs eight 200G EMLs. That drives substantial per‑module laser content. By deployment, **EMLs are used for Scale‑out interconnects across racks in AI clusters, linking GPU racks to leaf switches and leaf to spine (tens of meters to ~500m).** They primarily serve intra‑data hall links. Downstream customers include InnoLight, Accelink and Coherent among top module makers, ultimately serving North American hyperscalers. This is Lumentum’s core optics revenue base. Customer concentration is typical in this tier. **Industry checks indicate EMLs are hard because the DFB emission region and EAM modulation region are monolithically integrated on an InP substrate, requiring extremely complex second‑epi growth, with high process barriers and difficult yields.** **Only Lumentum, Mitsubishi Electric, Sumitomo Electric and Broadcom can stably mass‑supply 200G EMLs at high yields to top module customers today, with a combined 80%+ share. Lumentum is in the first tier by capacity scale.** Dolphin Research maps key EML vendors: Chinese suppliers have caught up in 100G EMLs, with Source Photonics, Lumentech and others in stable mass production and ~35% domestic share, and price competition has begun. **But for 200G high‑end EMLs, domestic self‑sufficiency is still under 5%, with most in sampling and small trial runs.** Also, **two Japanese vendors (Mitsubishi and Sumitomo) experienced overcapacity pain in the last 5G cycle, so they are conservative in this round, limiting new 200G EML capacity. Most incremental 200G supply comes from Lumentum and Broadcom.** Given that, **Lumentum is actively shifting mix toward higher‑value 200G EMLs. Management guides that by mid‑2027, 200G EMLs will exceed 50% of its EML shipments.** Industry checks suggest Lumentum’s EML laser revenue is about $640mn (roughly 20% of company revenue), with both volume and pricing up. The business is in a high‑cycle phase. **On one hand, AI cluster build‑outs are driving a surge in 800G/1.6T pluggables, lifting EML unit shipments to new highs; on the other, mix is shifting from 100G to 200G EMLs, roughly doubling per‑die ASPs and magnifying revenue elasticity.** Profitability is strong with composite GPM above 60% for EMLs. It is currently the largest profit contributor with the strongest pricing power. This underpins group margins. **2) CW/UHP lasers: the Scale‑up growth engine** Unlike EMLs that integrate emission plus modulation, CW lasers take the separation route. They act as the 'bulb', generating a high‑power, stable, data‑free beam, with an external silicon photonics die handling modulation. This underpins silicon photonics and CPO paths. ![IMG_256](https://pub.pbkrs.com/uploads/2026/34da1f8ebd42f8aa89740818909e8a2b?x-oss-process=style/lg) In general, the farther the light source is from the silicon photonics modulator, the higher the link loss, which raises the CW output power requirement. **That increases per‑die process barriers (noise, heat) and value content.** By power class, CWs cover three core scenarios. **a) 70–100mW — silicon photonics pluggables for Scale‑out** Yole estimates silicon photonics approaches ~30% penetration in data center modules, potentially topping 60% by 2030. This is a meaningful route share shift. **For Lumentum, this is a substitution within the installed base (**800G silicon displacing 100G‑EML, and 1.6T silicon displacing 200G‑EML**).** The value, however, is not equal. **For a 1.6T module, an EML solution needs eight 200G EMLs (ASP $15–25 each), while a silicon solution needs only four CWs (ASP $3.5–4 each).** Switching to silicon cuts total light‑source content per module. The company has said this tier is not a focus — unit growth exists, but per‑module value and earnings elasticity trail EMLs. Allocation follows returns. **b) 120–200mW — integrated light engines for NPO (near‑package optics), a Scale‑up path** NPO is a bridge from pluggables to CPO, placing the module on the PCB to speed signals between the switch and the optical engine. It improves system‑level bandwidth. Versus 70–100mW CWs in silicon pluggables, **this class has higher specs for output power, noise suppression and high‑temp reliability, meaning significantly higher per‑die value and technical requirements. It is pure incremental demand and does not cannibalize EMLs.** **c) 350–400mW (UHP) — for ELS in CPO/NPO, the Scale‑up core** ELS (external laser source) is the mainstream external light source in CPO architectures, serving Scale‑up interconnects for rack‑level GPU direct links. It supports chassis‑level optical power delivery. In CPO, the silicon engine and switch die are co‑packaged. **Lasers are the most failure‑prone and heat‑sensitive component; co‑packaging them means a single failure can take down the entire engine.** Hence the industry consensus for ELS: make CW lasers pluggable at the front panel, delivering light via PMF into the internal silicon engine. This design improves serviceability. On value, a CPO switch typically needs 16–18 ELS modules, each with eight UHP CW dies, implying 128–144 CWs per switch. That is a large per‑system bill of materials. **An ELS module sells for about $500–600, with eight CW dies at roughly $240, or 40–50% of the module value.** **In other words, one ELS brings CW vendors revenue equivalent to about 15 traditional silicon pluggable modules.** This is a substantial uplift per unit. Overall, versus lower‑power CWs, **UHP CWs have the highest technical barriers and are a pure AI‑driven incremental market that does not erode EML demand. They are Lumentum’s largest medium‑term growth lever.** On competition, **low‑power CWs are a red ocean: InP epi and chip structures are now mature, with the challenge shifting to low‑cost, high‑yield delivery.** Chinese vendors (Lumentech, Lumentum China peers, Yunling Optics, etc.) and low‑cost IDMs like AAOI are expanding, many having passed top module vendor quals. Downstream customers multi‑source to push prices down. For high power (350–400mW UHP), **only Lumentum and Coherent can stably mass‑produce today, and they take very different technical routes.** **Coherent: single‑die power (BH‑DFB)** — akin to a big single‑cylinder engine, using one buried‑heterostructure DFB to push 400mW. It is extremely compact with small die area and high die per wafer. **The trade‑offs are thermal, linewidth and long‑term reliability concentrated on one die, making early yield ramps slow.** This raises execution risk upfront. It seeks payoff post‑yield ramp. **Lumentum: two‑stage separation (DFB seed + SOA)** — like a 'small mature engine plus turbo'. A clean, low‑noise DFB seed is followed by an on‑die SOA to amplify to 400mW. Both sub‑blocks operate within mature process windows. Under this approach, **reliability is higher and the same platform can scale down to mid‑power, improving R&D efficiency.** The cost is a much longer die, slashing dies per wafer **from ~4,000–5,000 for normal CW to just ~400–500 at UHP, roughly a 10x drop.** In short, Coherent bets that after yields improve, smaller die area will drive cost leadership, while Lumentum trades die area for higher reliability. They optimize on different axes. This means that in a tight‑supply phase, hyperscalers prioritize delivery and stability over cost, favoring Lumentum’s area‑for‑reliability engineering path. **Checks suggest current ELS shares are roughly Lumentum ~60% and Coherent ~40%.** Over the longer run, **if Coherent lifts yields, its cost edge will emerge and pressure Lumentum.** Competition will likely intensify. For mid‑power, players capable of 400mW can readily make 150–200mW, making it a 'by‑product scale‑out' zone for high‑power leaders. Supply at this tier piggybacks on UHP platforms. **Conversely, for Chinese or second‑tier vendors stepping up from low power, 150–200mW is the gateway. Without breaking into high power, they cannot gain die‑per‑wafer economies across power bins, making mid‑power wafer costs uncompetitive versus leaders’ 'by‑product mode'.** Industry checks indicate current CW laser revenue is about $50mn, roughly 2% of company revenue, still small. It mainly reflects initial silicon pluggable demand and CPO sampling. This base can scale with architectures. That said, as the route and power tiers show, **the real CW inflection depends entirely on NPO/CPO adoption. As the industry moves from pluggables to CPO, CW demand will grow exponentially.** (We will quantify in the next note.) \ Risk disclosure and statement: [Dolphin Research disclaimer and general disclosure](https://support.longbridge.global/topics/misc/dolphin-disclaimer) ### Related Stocks - [03308.HK](https://longbridge.com/en/quote/03308.HK.md) - [300308.CN](https://longbridge.com/en/quote/300308.CN.md) - [IPG.US](https://longbridge.com/en/quote/IPG.US.md) - [300502.CN](https://longbridge.com/en/quote/300502.CN.md) - [VIAV.US](https://longbridge.com/en/quote/VIAV.US.md) - [LITE.US](https://longbridge.com/en/quote/LITE.US.md) - [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md) - [300394.CN](https://longbridge.com/en/quote/300394.CN.md) - [NVDD.US](https://longbridge.com/en/quote/NVDD.US.md) - [NVDL.US](https://longbridge.com/en/quote/NVDL.US.md) --- > **Disclaimer: This article is for reference only and does not constitute any investment advice.**