Lumentum: Capacity Crisis, What Other Core Skills Does the All-Purpose "Water Seller" Have?

In the article "From Optical Interconnect Veteran to 'Universal Water Seller,' What Makes Lumentum Stand Out?", Dolphin Jun broke down the two core optical component businesses of Lumentum: EML securing the basic profit foundation of pluggable optical modules, while high-power discrete laser UHP CW bets on the incremental growth of CPO/NPO next-generation architectures.

The underlying logic of both lines ultimately converges—the extreme mismatch of supply and demand for high-end laser chips allows top manufacturers with self-built production capacity to simultaneously enjoy price hikes, long-term orders, and rising gross margins.

However, whether it's EML or CW, they all have to be "cut" from the same piece of indium phosphide wafer. The slow growth of indium phosphide crystals, lack of foundries, and the fact that new capacity takes two to three years to ramp up means that Lumentum's growth will sooner or later hit the wafer capacity ceiling. So, does Lumentum have any other cards to play?

In this piece, we continue to explore, focusing on two questions:

1. Why does Scale-across enable traditional telecom device businesses to "revitalize"?

2. Why does Lumentum step into the low-margin module business?

Detailed Analysis Below

I. Other Telecom Components & Industrial Business: New Scale-across Scenarios "Revitalize" the Old Tree

Lumentum’s telecom business has been a typical stock business in recent years—the demand side is global operators, with CAPEX at a steady pace and fluctuating with the cycle, growth relying on network expansion. This round of AI upended this narrative for a simple reason—there isn’t enough power.

You have to know, the power supply capacity a single data center park can obtain is limited. When the size of training clusters exceeds the power a single park can handle, training tasks must be distributed across parks—another data center tens to hundreds of kilometers away, which needs to collaborate with the local data center as if they’re one machine. This link is called Scale-across.

As distance increases, existing internal DC optical interconnect tech fails:

Within data centers (Scale-up & Scale-out), Intensity Modulation-Direct Detection (IM-DD) is widely used: it relies on optical signal intensity to carry data, with receivers using photodiodes to directly detect optical strength, converting it into electrical signals. This approach is low-cost and power-saving, but the downside is that only the intensity is captured at the receiver—phase information (the particular state of the light wave at specific points/times) is lost during opto-electrical conversion.

This has limited impact for short distances (compensated through DSP in the electrical domain), but for long-distance fiber transmissions, dispersion occurs; light of different wavelengths travels at different speeds, pulses widen and ultimately interfere with each other.

Therefore, inter-park, long-distance interconnect must use coherent optical communication, using phase and polarization dimensions of light to encode information. The receiver uses a local oscillator light mixed with signal light for demodulation and recovery of the full amplitude and phase information, with coherent DSP compensating for dispersion, nonlinear damage, and signal error correction.

However, coherent solutions greatly increase link complexity: the system needs ultra-narrow linewidth tunable lasers, high-precision coherent modulators, dual polarization receivers, and every tens of kilometers, optical amplifiers (EDFA) must be deployed to make up for fiber optical power attenuation.

For Lumentum, its accumulation in telecom long-haul and metro optical devices and transmission engineering in the coherent photonics field can be reused in AI data center inter-park interconnect (Scale-across/DCI) scenarios.

Let’s look specifically at Lumentum’s two most representative components in coherent light connectivity:

1. Tunable Laser (ITLA)—the "Clean Light Source" for Coherent Links

ITLA (Integrated Tunable Laser Assembly) is the light source in a coherent module, with only one purpose: to continuously output a "clean" continuous light beam. Usually, only one ITLA is installed in each ZR coherent module; the light is split in two—one path is modulated and sent to carry data, the other is kept locally as the reference light for demodulation (local oscillator). So if the light source quality is problematic, both transmission and reception are simultaneously affected.

"Clean" means meeting three conditions:

a. Ultra-narrow linewidth: The color of light is determined by wavelength. Even a seemingly single-color beam stretches a bit around a center wavelength; this width is called linewidth.

Ultra-narrow linewidth means the phase carrying information remains regular for a long time. Since phase is used to transmit data, if the source's phase is unstable, the receiver can't distinguish meaningful changes from noise.

b. The wavelength must be precisely tunable to any channel in the C-band (1310/1550nm communication band—lowest fiber loss).

c. Long-term stability of light source quality

Unlike EML lasers (which require a redesign with each speed generation and manufacturing advances like secondary epitaxy matching), ITLA’s challenge is not the chip architecture, which doesn’t need to track transmission rate upgrades—it’s matured through decades of iterations.

The core mass production challenge lies in downstream micro-packaging and full-wavelength calibration, which are key to outputting the "clean" light source required for coherent communications.

a. Micro-packaging: For EML and CW, the light path is essentially straight—the chip outputs light, which passes through lenses and isolators into fiber, with wavelength fixed by the chip, and the optical elements arranged in series.

As you can see from the diagram below, ITLA adds a "self-correction" feedback loop: after the chip outputs light, a small portion is sent to a wavelength locker to continually compare the current wavelength with the standard grid, feeding back to adjust the chip’s drive current.

Therefore, ITLA’s packaging not only protects the chip and extracts the optical signal; crucially, in a non-highly integrated setup, every discrete optical component inside (lens, isolator, prism, etc.) must be precisely positioned, fixed, sealed, temperature-controlled, powered, and optically connected.

This new feedback optical path in the ITLA brings in a set of components that need micron-level optical alignment, and their relative positioning must remain unchanged despite temperature fluctuations and long-term aging, greatly increasing packaging difficulty.

b. Full-wavelength calibration: EML and CW lasers operate at a single fixed wavelength, so calibration is simple: at target temperature, calibrate drive current, output power, extinction ratio, etc. one by one and you're done.

But for coherent transmission, using Dense Wavelength Division Multiplexing (DWDM): dozens of signals transmit simultaneously in a single fiber, each on a precise wavelength channel with ultra-narrow spacing, so the ITLA must be able to switch on command among dozens to nearly a hundred channels across the C-band.

Ensuring stable output at a given channel requires simultaneously setting values for the reflection section, phase section (via injected current or heating power), gain section current, and chip temperature—all interrelated and non-linear. Thus, a one-size-fits-all table can't cover all chips.

So each ITLA must be individually fully parameter-calibrated before leaving the factory—across the full operating temperature range, scanning all combinations of control variables, and recording the output wavelength and power to create a multi-dimensional "control parameter—wavelength" map for the device. The precision here determines whether wavelength switching works instantly; lack of precision causes drift and even channel cross-talk.

However, overall, the two craft challenges of ITLA—micro-packaging and full-wavelength calibration—lie in engineering experience in manufacturing and testing, not fundamental design or material science. The real gap is yield, long-term stability, and calibration model accuracy—all of which can theoretically iterate as shipments scale.

Thus, there are more players in this track: Oversee competitors include Lumentum, Coherent, Sumitomo Electric, and Fujitsu; in China, the Huawei group and Accelink have independently developed ITLA, mainly supplying domestic operator networks.

For Lumentum, its core advantage is long-term stable qualification in overseas equipment vendor supply chains (while Chinese manufacturers' reliability certification is mainly domestic, overseas validation is still limited). But as coherent pluggable module penetration rises and domestic companies gradually push into overseas certifications, Dolphin Jun judges that Lumentum’s supply chain advantage will gradually weaken.

2. Pump Laser—The "Excitation Light Source" Injecting Power into EDFA Optical Amplifiers

Coherent signals continuously attenuate in fiber transmission. If it’s like a long-range car trip, you lose fuel and need gas stations and fuel trucks along the way.

This is where EDFA (Erbium-Doped Fiber Amplifier, a special fiber doped with erbium) acts as the gas station, and the pump laser is the fuel truck—pump light injects power into the fiber, the erbium atoms absorb it (get "fueled up").

So when C-band signal light with data passes by, the energized erbium atoms are "triggered" to emit identical light, amplifying the signal.

You can see, the 980 pump laser is an energy light source, not a data carrier, with low requirements for color purity, but key for high power—since amplifier output depends on how much energy is fed in.

For a laser to emit light, electrons drop from higher to lower energy states, releasing photons—the greater the energy gap, the more energy released.

This energy gap is determined by semiconductor material: the active emission layer is grown layer by layer via epitaxy on the substrate; substrate choice thus defines compatible emission materials and the achievable emission wavelength range.

The previously mentioned EML, ITLA, and CW all require C-band and thus need indium phosphide (InP) substrates; but 980nm pump lasers (since erbium ions in EDFA only absorb 980nm, 1480nm, and 980nm is mainstream for long-haul) can only be matched with gallium arsenide (GaAs).

With this fundamental substrate difference in mind, let's analyze the core technical barriers for pump lasers:

After the laser chip is polished, epitaxied, and manufactured, unlike silicon photonics, it’s cleaved along the crystal lattice for atomically smooth edges—this is the "cleaving" process in photonic chip production.

The real "kung fu" is dealing with the exposed facet after cleaving—the laser cavity facet. After "cleaving," two main problems emerge:

a. 980nm pump lasers are high-power energy sources with small exit apertures; the cavity surface undergoes immense power pressure and is easily damaged:

b. With GaAs as substrate, gallium is highly reactive, and the external materials for pump lasers contain aluminum, which oxidizes rapidly after cleaving, causing surface defects.

Oxidation absorbs light, generates more heat, leads to local melting of the facet, and kills the laser—this is so-called catastrophic optical mirror damage (COMD).

The solution is to operate in ultra-high vacuum and deposit a protective film on the facet—facet passivation—to prevent oxidation and isolate from air.

Unlike ITLA micro-assembly (equipment and components can be bought and the difference lies in assembly yield), facet passivation is more material and process know-how: film material and thickness, vacuum conditions, interval between cleaving and film deposition—every parameter determines the facet’s power tolerance, representing a higher technical barrier.

Lumentum is an industry leader in facet passivation: Besides passivating the facet, it uses an internal non-absorption window structure to eliminate facet absorption, greatly improving the chip's COMD tolerance.

In industry competition, telecom-grade 980nm pump lasers—with strict beam shape requirements—remain highly concentrated among a few overseas firms: Lumentum (technology lineage from former pump leader SDL), Coherent, and Furukawa Electric. According to Lumentum management in earnings calls, the company holds about 70%-80% market share here—a high monopoly.

Besides the two representative components analyzed above, Dolphin Jun summarized Lumentum’s full coherent optical product line and main competitors in the table below.

The table shows a clear rule: the closer to optical physics (emission and amplification), the higher market concentration (these core steps depend on compound semiconductor material epitaxy and chip processing, with much higher technical barriers), while moving upstream towards modulation and systems, the competitive landscape grows more diverse and complex.

Overall, Lumentum is selectively extending downstream, integrating modules/devices only in the segments where it maintains dominance in core optical chips (e.g., 800ZR/1.6T ZR coherent modules, Nano-ITLA assemblies), retaining natural gross margin and supply security advantages, and avoiding excessive involvement in areas lacking optical chip barriers and dominated by DSP giants like Broadcom or Marvell.

The current telecom business is $1.2 billion, close to 40% of company revenue. According to research, the part directly related to AI park-to-park interconnect is roughly $300 million (25%), with the rest ~$900 million from traditional operator demand, basically steady.

In terms of timing, DCI coherent modules are being upgraded from current large-scale deployment of 400ZR (single-wavelength 400G coherent module) to 800ZR.

According to Cignal AI, the general trend is for 800ZR mass adoption around 2026–2027, and from 2028, 1.6T ZR will gradually be commercialized (the 1.6T ZR standard was released in September 2026, with sampling expected in 2027 and ramp-up in 2028), when the value of a single light source will see a second leap.

At the current YoY shipment growth rates of 80%–130% (pump laser shipments up 80+%, tunable narrow-linewidth laser shipments up 130% in the latest quarter), the explosive growth of AI inter-park interconnect will drive traditional telecom components into a second growth curve.

From the above, it’s clear that, whether EML as the basic profit pillar, CW (including UHP‑CW) as the new compute cluster increment, or DCI coherent as the telecom segment growth unlocked by AI inter-park interconnect, they all compete for the same supply-constrained indium phosphide wafer capacity.

For Lumentum, with indium phosphide wafer supply constraints, maximizing performance elasticity means either continuously increasing value per wafer by making higher value chips (maximizing revenue and profit per wafer) or developing entirely new businesses not dependent on indium phosphide wafers, breaking free from the capacity ceiling.

Let's continue the analysis:

4. Optical Modules: From Selling Chips to Modules, Completing Modular Delivery Capability

After acquiring Cloud Light in 2023, Lumentum obtained the ability to deliver complete 800G/1.6T optical modules directly to Google, AWS, etc. These products also serve Scale-out inter-cabinet networking, only now the delivered product moves up from chips to full system modules.

From a competitive landscape perspective, pluggable optical modules are a highly mature, fiercely competitive market. According to LightCounting’s latest rankings, seven of the world’s top ten module makers are from China.

Chinese manufacturers have significant advantages in capacity and manufacturing cost. This means Lumentum does not compete for share on commodity standardized SKUs through capacity and cost, but instead focuses on high-speed, highly technical, custom modules.

Based on research, Lumentum’s current annual module revenue is about $720 million, roughly 24% of total, making it the company’s single largest product line by revenue—but gross margin is just around 30%, significantly lower than that of previously analyzed device businesses.

The question is, if the pluggable module track is fiercely competitive with low gross margins, why did Lumentum acquire Cloud Light? Dolphin Jun believes there are two main reasons:

a. Internalizing self-developed light sources to improve module business profitability

In the past, Cloud Light had to purchase external lasers, but the company is now gradually installing in-house CW into its own modules (currently about 20% use self-developed light sources, with this ratio set to increase). By increasing the proportion of in-house sources, profits once paid externally are retained, boosting the gross margin of the module business.

Considering that the light source accounts for about 17%-18% of the module price, a simple calculation: if Lumentum’s self-supply rises from 20% to 50%, module gross margin can increase by about 3 percentage points.

b. Completing module-level packaging, full-system verification, and cloud vendor delivery qualification for NPO/CPO business preparedness

According to company call info, some NPO and CPO project customers require vendors to have subsystem/full-system delivery capability. If you only supply bare chips—even with qualified UHP-CW performance—you cannot directly win ELS system orders. Cloud Light’s packaging, system testing, and cloud qualification capabilities complete Lumentum's chip-to-module delivery chain, effectively securing ELS system supply chain entry.

5. OCS Optical Circuit Switch: A Byproduct of Technology Downscaling

Finally, Lumentum’s OCS business—a purely optical switch (replacing electric switches at the data center SPINE layer), delivering ultra-low latency and transparent transport.

Why does Lumentum engage in OCS? It's easy to understand: the growth logic is the same as DCI coherent devices—it’s a reuse of Lumentum’s existing technology.

The MEMS chip with hundreds of micro-mirrors in an OCS is the same optical path scheduling tech as in the wavelength selective switches (WSS) in the telecom business for over twenty years—both use micro-mirrors to precisely direct light beams to selected ports. The difference is WSS switches by wavelength, while OCS switches by physical port in the optical path.

Therefore, when Google replaced electric switches in TPU clusters with OCS and needed an external supplier, Lumentum was among the few in the industry able to provide hundreds-of-port scale, with its MEMS micromirror path validated by decades on real telecom networks.

Currently, there are five competing OCS routes: Lumentum's 3D-MEMS (micromirror deflection), Coherent’s LCOS silicon-based liquid crystal, Polatis’ piezo-ceramic, iPronics’ silicon photonics, and Telescent’s robotic fiber patching.

So far, OCS tech direction is not fully settled, but to meet both "hundred-port scale + hyperscale AI cluster network validation," only the MEMS route is realized, having passed Google TPU cluster scale trials.

Lumentum’s OCS advantage mainly comes from being first and having large-scale real-net validation—the basic technical barrier is not high, the real hurdle is mass production and cluster-level engineering validation.

Additionally, there’s a major risk, as Google is both Lumentum OCS’s largest customer and possesses in-house R&D—potentially a future competitor.

Further business growth will depend on whether the second-tier hyperscalers—Microsoft, Meta—follow suit with mass deployment. For now, only Google has achieved full deployment; Microsoft, Meta, AWS remain at small-prototype testing and not yet in volume orders.

Summary: The Only All-Scenario 'Water Seller' in Optical Connectivity

Through the analysis in these two articles, it is clear that every iteration and expansion in AI computing topology creates demand for Lumentum:

Horizontal expansion (Scale-out, between racks): Relying on high-end EML laser chips to guard the ultra-high-margin profit base of traditional pluggable optical modules;

Vertical density (Scale-up, within racks): With next-generation CPO/NPO architecture, explosive growth through UHP ultra-high power CW chips;

Cross-regional extension (Scale-across, inter-park): Adapting to distributed cluster demand prompted by 'electricity bottlenecks,' repurposing telecom-grade coherent tunable sources (ITLA), pump lasers, and ROADM/WSS for long-haul DCI links, turning stock assets into 'revitalized old trees';

Network topology (Spine architecture): With the MEMS OCS optical switch attacking all-optical networks, unlocking a high-margin, large-scale second growth curve not constrained by indium phosphide wafers.

In other words, no matter which level the AI optical network evolves to, Lumentum holds an indispensable position at the foundational layer. In the next article, we will provide performance estimates for each product segment and see what Lumentum is really worth—stay tuned!