Every conversation about AI infrastructure eventually arrives at the same two names: NVIDIA's GPUs and the data centers that house them. But there's a third layer that almost nobody talks about, and it might be the most consequential bottleneck in the entire AI buildout.
Light.
Specifically — how you move data between thousands of GPUs at speeds that electricity-based copper cables simply cannot achieve, without melting the wires or burning through a small city's worth of power. The answer is photonics. And the company at the center of that answer is a San Jose-based manufacturer called Lumentum Holdings (NASDAQ: $LITE).
On March 2, 2026, NVIDIA wrote a $2 billion cheque to Lumentum. Jensen Huang doesn't write cheques to photonics companies because it feels nice. He wrote it because optical components are a genuine bottleneck to AI infrastructure scaling — and he wants Lumentum locked in as the supplier.
This is Part 1 of a five-part deep dive into $LITE. Before we get to valuations, before we talk about whether you should buy the stock, we need to understand what this company actually does. Because without that foundation, the rest of the analysis is just numbers on a page.
First, a Quick Primer: Why Light Beats Electricity for Data
If you're not a telecom or networking professional, this concept might feel abstract. Let me make it concrete.
Copper wire (electrical signal): Think of it like a pipe carrying water. It can carry a lot of information, but as you increase the volume or speed, friction builds up, heat is generated, and the signal degrades. Copper cables start to struggle at data rates above 400 gigabits per second over anything but very short distances.
Optical fiber (light signal): No friction. No electrical resistance. Light travels through glass fiber at, well, the speed of light. You can push 1.6 terabits per second through a single fiber strand — four times what copper can handle — and do it over kilometers, not just meters. It also uses a fraction of the power.
In a world where NVIDIA's Blackwell GPU clusters need to move petabytes of training data between thousands of processors in milliseconds, copper doesn't cut it anymore. The entire spine of a modern AI data center is being rewired with optical fiber and optical components. This is not a trend. This is a fundamental architectural shift in how computing infrastructure works.
Lumentum makes the lasers and optical components that sit at the heart of this shift. Every time data goes from electrical (how your CPU processes it) to optical (how it travels between servers) and back again, there are Lumentum components doing that conversion.
The Business: What Lumentum Actually Sells
Lumentum operates in two segments. Here's the honest version:
Segment 1: Cloud & Networking (~90%+ of Revenue)
This is the engine. Everything important about Lumentum's future lives here. Within this segment, the key products are:
EML Chips (Electro-Absorption Modulated Lasers)
This is the crown jewel product. An EML is essentially a tiny semiconductor laser — typically a few millimetres long — that converts electrical data signals into precisely modulated pulses of light at extremely high speeds. Modern EMLs run at 100G (100 gigabits per second per channel) and increasingly at 200G per channel. Inside every high-speed optical transceiver in a hyperscale data center, there are typically 4 or 8 of these chips. Lumentum holds leading positions in both 100G and 200G EML chip production.
The simple version: if a data center is the brain, and optical fiber is the nervous system, EML chips are the synapses.
Optical Transceivers (post-CloudLight acquisition)
A transceiver is a complete module that both transmits and receives optical signals. Think of it as a plug-in card that slots into a switch or server and handles the electrical-to-optical conversion. Before November 2023, Lumentum sold EML chips to transceiver makers. After the CloudLight acquisition, they now make and sell complete transceivers directly to cloud customers. More on why this pivot changes everything, shortly.
Narrow Linewidth Lasers for Data Center Interconnect (DCI)
These are specialised lasers used for connecting data centers that are geographically separated — a hyperscaler might need to move data between its Mumbai campus and its Pune campus at 100+ terabits per second. The "narrow linewidth" refers to the extremely precise frequency of light, which allows signals to travel longer distances without degrading.
Optical Circuit Switches (OCS)
This is the newest and potentially highest-value product in the portfolio. An OCS is essentially a photonic traffic director — it routes optical signals between different paths inside a data center without ever converting them back to electrical form (which adds delay and consumes power). Think of it like a highway interchange that operates entirely in the light domain. Lumentum has a $400 million+ OCS backlog as of early 2026, and this is the product that could expand margins most significantly over the next two years.
Pump Lasers
Used in optical amplifiers along long-haul fiber routes to boost the signal without electronic conversion. A more commoditised product, but a steady revenue contributor.
Segment 2: Industrial Tech (~10% of Revenue)
This segment covers commercial lasers for industrial manufacturing (cutting, welding, materials processing), 3D sensing (facial recognition technology — Lumentum was a key supplier for Apple's Face ID), and scientific applications. This segment has been in steady decline, primarily because Apple began diversifying its 3D sensing supply chain. Management is not in denial about this — they've been gradually pivoting resources toward Cloud & Networking.
The chart above shows how dramatically Cloud & Networking has come to dominate the revenue mix, and how the AI-driven acceleration from FY2024 onward dwarfs everything that came before it.
The CloudLight Pivot: The Most Important Thing That Happened to This Business
In November 2023, Lumentum acquired a company called Cloud Light for approximately $750 million. This is the strategic inflection point that most casual observers missed, and it fundamentally changed what kind of company Lumentum is.
Before CloudLight: Lumentum was a components business. They sold laser chips and sub-components to transceiver makers like InnoLight, Accelink, and others. They were one step removed from the actual hyperscaler relationship. The hyperscaler (Google, Amazon, Microsoft) would buy transceivers from InnoLight, not from Lumentum directly. Lumentum was important, but somewhat invisible.
After CloudLight: Lumentum can assemble complete transceiver modules and sell them directly to hyperscalers. They've vertically integrated — moved up the value chain. Now they sit on both sides of the market: selling chips to transceiver makers (as before), and also competing with those transceiver makers in the direct hyperscaler market.
The diagram above shows exactly where Lumentum sits in the optical value chain, and how the CloudLight acquisition extended their reach one step closer to the end customer.
Why does this matter?
Margin: When you sell a chip to a transceiver maker for $30, and they sell the completed module to Google for $150, the transceiver maker captures the difference. Post-CloudLight, Lumentum captures more of that spread.
Relationship: Direct hyperscaler relationships mean earlier access to design requirements, longer-term purchase commitments, and greater product co-development. Lumentum now sits in engineering meetings with Google and Microsoft, not just in procurement conversations.
Revenue visibility: When Google signs a multi-year purchase commitment directly with Lumentum, that's a completely different quality of revenue certainty than selling chips spot-market to a transceiver assembler.
The downside? More capital intensity, more manufacturing complexity, and more exposure to the hyperscaler capex cycle. But on balance, this was the right move.
The Revenue Story in One Chart
Before we go deeper, let's anchor on the numbers.
This chart tells you everything about the business cycle. From the trough of $308M quarterly revenue in Q4 FY2024, Lumentum climbed to $808M in Q3 FY2026 — a 162% increase in six quarters. For context, this company was doing $425M/quarter just a year before the $808M print. The acceleration is not gradual. It's parabolic.
What drove the trough? Two things:
- Inventory digestion across the telecom/datacom supply chain. After the COVID-era overbuild, networking companies were sitting on excess inventory and stopped ordering components for 18 months.
- Apple's 3D sensing diversification. A meaningful chunk of Industrial Tech revenue evaporated.
What drove the recovery and subsequent surge? Entirely the AI data center buildout. When hyperscalers started building GPU clusters for AI training at scale — NVIDIA Hopper, then Blackwell — the optical networking requirements exploded. Every new generation of AI hardware roughly doubles the optical content per server rack. Lumentum was already qualified and ready to supply.
How Lumentum Makes Money: The Simple Version
It's a hardware components business with a capital-intensive manufacturing model. There are no subscriptions, no recurring software revenues, no services to speak of. Revenue is recognised when components ship. This means:
- Revenue is lumpy: A large hyperscaler order can move a quarter meaningfully.
- Margins leverage on volume: The fab (semiconductor fabrication facility) has high fixed costs. As revenue increases, margins expand rapidly because the incremental cost of the next chip is low. This is exactly what we're seeing now — gross margins expanding from ~27% at the trough to 35%+ today.
- Capex cycles matter: When hyperscalers decide to spend, Lumentum sees it in orders 6–12 months before it shows in quarterly revenues. When they pause, the pipeline dries up fast.
This is a cyclical business with a powerful secular overlay — the AI capex story provides a long runway, but the component supplier is never fully insulated from the spending cycles of a handful of very large customers.
The Moat: How Hard Is This Business to Replicate?
This is the question that determines whether $LITE deserves a premium valuation or just looks expensive.
Let me explain the two highest-rated moats:
1. Switching Costs — The 12-18 Month Qualification Moat
When a hyperscaler or network equipment maker decides to use Lumentum's EML chips in their transceiver design, they run an extensive qualification process — testing the chip across thousands of operating conditions, validating yield consistency, ensuring it meets their specific optical specifications. This typically takes 12–18 months and costs the customer significant engineering resources.
Once a chip is qualified, the hyperscaler does not want to restart that process with a different supplier unless there is a compelling reason (price catastrophe, supply failure, or new technology generation). This creates a natural lock-in period that typically spans the entire product generation cycle — 2–4 years.
This is the photonics equivalent of a switching cost moat, and it's real.
2. Intangible Assets — InP Process Knowledge
Lumentum's lasers are built on Indium Phosphide (InP) — a semiconductor material that is fundamentally different from the silicon used in conventional chips. InP is much better for optical applications (it can emit light natively; silicon cannot), but it's also much harder to work with. The yield optimization, defect control, and manufacturing process knowledge accumulated over decades of InP wafer fabrication is not written in a textbook. It lives in the engineers, the process recipes, and the accumulated data sets of a company that has been building InP lasers for 20+ years.
A new entrant trying to compete with Lumentum on 200G EML chips would need to build an InP fab, hire experienced engineers who are genuinely rare, run through 3–5 years of yield improvement cycles, and then still go through 12–18 months of customer qualification. That's a 5–7 year head start that Lumentum holds.
The honest caveat on the moat: This is a narrow moat, not a wide one. Coherent — the closest competitor — has comparable InP capabilities and has also received a $2B NVIDIA strategic investment. The US government's preference for non-Chinese optical supply chains benefits both Lumentum and Coherent simultaneously. This is not a monopoly. But it is a well-defended oligopoly in a market where demand is currently running ahead of industry-wide supply.
The Competitors: Who's in the Race?
Coherent (NYSE: COHR) — The most direct full-portfolio rival. Also received a $2B NVIDIA investment on the same day as Lumentum. Strong in coherent transceivers and datacom components. Larger company (revenues approaching $7B annually), but more diversified — which means less pure-play AI optical exposure.
Applied Optoelectronics (NASDAQ: AAOI) — Smaller, focused more on transceivers than on chip-level components. Has had a spectacular run in 2026 (up 440%+ YTD at some points) driven by similar AI demand tailwinds, but operates at a lower technology tier.
Chinese Players (Accelink, EOLITE, etc.) — Technically capable but constrained by US export control concerns and hyperscaler preference for non-China supply chains. This is Lumentum's geopolitical tailwind — Western hyperscalers are actively trying not to be dependent on Chinese optical components for their AI infrastructure.
What Comes Next in This Series
You now understand the business. That's the hardest part of any analysis — everything else builds on it.
In Part 2, we'll go deep on the financial story — the revenue ramp, the margin recovery, the balance sheet, and why the gap between GAAP losses and non-GAAP profits tells you something important about how to read this business.
In Part 3, we tackle the headline story: NVIDIA's $2B investment, the 1.6T transceiver demand surge, the OCS backlog, and what Project Vanguard (Lumentum's US manufacturing expansion) means for the geopolitical thesis.
In Part 4, we do the valuation math honestly — bull case, base case, bear case — and tell you exactly what assumptions the market is pricing in at current levels.
In Part 5, we bring it home for the Indian investor: the LRS route, the currency math, the tax structure, whether any Indian-listed company gives you similar exposure, and what an entry strategy actually looks like.
If you found this useful, share it with one person who talks about AI stocks but has never heard the word "photonics." That's exactly who this series is written for.
Manu | Curious Investing Insights
Applying institutional analytical depth to markets that matter for Indian investors
This is not investment advice. Do your own research. The author may hold positions in securities discussed.
Tags: AI Infrastructure · Optical Networking · Lumentum LITE · US Stocks for Indian Investors · Photonics · Data Center