Hey everyone,
This is the fifth post in our Investing Themes series. We've covered the power and cooling behind AI, cybersecurity, biotech, and the memory shortage. This one is about a part of the AI buildout that most people never see: the light inside the data center. Photonics, the technology of moving information with light instead of electricity, is quietly becoming the wiring of AI. It's also one of the most violent corners of the market right now, so I want to be especially clear about both the opportunity and the risk.
A timing note, because it matters. I'm writing this on November 4, 2026. Optical stocks have just been through a brutal stretch: the group started sliding in mid-October on worries about how AI buildouts get funded, and the three companies I picked are each roughly 55% below their 52-week highs. On November 2, Marvell also agreed to buy Lumentum, one of the best-known names in the space, which I'll explain below. All numbers are approximate and move daily.
A quick heads up on how this post works: parts 1 to 6 are open to everyone. Parts 7 and 8, where I go deep on my three picks (what each company does, the bull case, the bear case, and what I'd watch next) and walk through the risks and how I'd own it, are free too, but you'll need to enter your email to unlock them. It takes about five seconds, costs nothing, and there's no password or credit card.
1. What is photonics, and why does AI need it?
Computers move information two ways inside a data center: as electrical signals through copper, and as pulses of light through glass fiber. Copper is cheap and simple, but it has a physical limit. The faster you push data down a copper wire, the more the signal fades and the shorter the distance it can travel, and the more power you burn trying to fix it. Light doesn't have that problem. It can carry far more data, over far longer distances, using less energy per bit.
For decades that didn't matter inside a building. Fiber handled the long hauls between cities, and copper handled the last few feet between servers. AI changed that. Training a large model means thousands of chips acting as one giant computer, and they have to exchange enormous amounts of data constantly. When the chips can't talk fast enough, they sit idle, and idle AI chips are a very expensive way to waste electricity. So the connections inside and between data centers are being pushed toward optics.
That's the heart of the theme: as AI clusters get bigger, the network that ties them together becomes as important as the chips themselves, and more of that network is becoming light.
2. How does an optical link actually work?
You don't need an engineering degree, just the chain of five steps:
- A laser makes the light. A tiny semiconductor laser, no bigger than a grain of sand, produces a steady beam. Many of the best lasers are made from a material called indium phosphide, which is harder and slower to manufacture than ordinary silicon chips.
- A modulator writes the data onto it. The beam is switched or dimmed billions of times per second to encode ones and zeros.
- Fiber carries it. The light travels through hair-thin strands of ultra-pure glass. Corning, which invented low-loss optical fiber in the 1970s, is the best-known maker.
- A detector catches it. A photodetector at the other end turns the light back into an electrical signal.
- A chip cleans it up. A digital signal processor (DSP) fixes errors and hands the data to the server or switch.
Most of these pieces are packed into a plug-in module called an optical transceiver, which looks like a small metal stick and plugs into the front of a switch. Transceivers are sold by speed. Data centers moved from 400 gigabits per second to 800G, and the industry is now ramping 1.6 terabit modules. Each speed jump needs better lasers, better chips, and tighter manufacturing, which is why the supply chain is hard to scale quickly.
3. What's new: co-packaged optics and the three kinds of AI network
Two ideas explain most of the news in this industry.
First, AI networks have three scales, and optics is spreading into each one.
- Scale-up: chips talking inside a single rack. This was copper's home turf, and it's the next big frontier for optics.
- Scale-out: racks talking to other racks across a data center. This is where pluggable transceivers and fiber already dominate.
- Scale-across: whole data centers talking to each other across a campus or a region, as AI clusters outgrow a single building. This is where optical networking systems from companies like Ciena come in.
Second, there's a push to move the optics closer to the chip. Today's pluggable modules sit at the edge of the box, and electrical signals still have to travel from the chip out to them, which wastes power. Co-packaged optics (CPO) puts the optical engine right next to the chip, and near-packaged optics (NPO) sits just beside it. Nvidia, Broadcom, and others are investing here, and reports say Nvidia put roughly $4 billion into the photonics ecosystem. Coherent has launched a platform for CPO and NPO called PhotonLink, Tower Semiconductor has announced NPO work with Innolight, and Credo moved into silicon photonics by acquiring DustPhotonics. A related product, the optical circuit switch, uses mirrors to steer light between ports without converting it to electricity, and it's a fast-growing niche.
Why this matters to an investor: it means the technology map is still being redrawn. Today's winners in pluggable modules may or may not be tomorrow's winners in co-packaged ones, which is one of the risks I'll come back to.
4. Why is optical capacity so tight?
The stocks have been strong for one reason: demand is outrunning supply, and it takes years to fix that. A few signals from recent earnings reports:
- Lumentum said its production capacity is largely spoken for through early 2029.
- Corning said availability of its newest high-density optical products is tight, and it expects about $2 billion of capital spending in 2026, mostly to add optical capacity.
- Ciena exited its latest quarter with an $8.5 billion backlog, up $800 million in three months, and expects it to pass $10 billion by year-end, mostly because lead times have stretched.
- Fabrinet, which manufactures optical products for many of the big names, reported record revenue up 45% in its fiscal fourth quarter and about 51% in the quarter after.
The reason it's hard to scale is that these are precision products. Indium phosphide lasers, high-density fiber, and 1.6T modules all need specialized factories, careful testing, and customer qualification that can take many months. That gives suppliers pricing power while shortages last. It also, as we'll see, is exactly how overbuilding starts.
5. The layers of photonics, and who plays where
| Layer | What it does | Names to know |
|---|---|---|
| Light sources and modules | Lasers and the transceivers built from them | Coherent, Lumentum, Applied Optoelectronics, Chinese makers like Innolight |
| Fiber and connectivity | The glass and cabling that carries the light | Corning |
| Manufacturing | Builds the modules for the big brands | Fabrinet |
| Optical networking systems | Links data centers and cities together | Ciena |
| Chips and switches | The silicon that drives and manages the light | Broadcom, Marvell, Credo, Nvidia |
I chose one company from three of these layers: a source and module maker, a fiber maker, and a manufacturer. I'll also explain in part 7 why I'm leaving out Lumentum and Ciena for now.
6. What happened last time, and how I'd judge a photonics stock
We've been here before. In 1999 and 2000, optical networking was the hottest theme in the market, and the big names rose many times over. Then the telecom buildout overshot, orders disappeared, and the leaders fell by more than 90% from their peaks. Corning, JDS Uniphase (the predecessor of Lumentum), and Ciena all went through it, and some took years to recover. The technology was real and still is. The lesson is that a great technology can still be a terrible stock if you pay the peak price for peak demand.
Here is the checklist I use now:
- Backlog and lead times. A growing backlog and stretching lead times mean real demand. Watch for them to shrink, which can be the first sign of a turn.
- Customer concentration. Optical companies often depend on a handful of giant buyers. The more of their revenue comes from one or two customers, the more fragile the story.
- Capacity and capital spending. Heavy spending to add capacity is a vote of confidence, and also next year's oversupply if demand slows.
- Gross margin and product mix. Higher-speed products and new technologies usually carry richer margins. Watch whether margins hold as volumes ramp.
- Competition, especially from China. Chinese suppliers are strong in transceivers and can compete on price.
- The price you pay. Compare a stock's earnings multiple to growth, and ask what happens if profits fall by a third. Our DCF course and P/E explainer can help.
- Who funds the buildout. These companies sell to hyperscalers and AI labs. If those customers can't afford to keep spending, the orders slow.
7. My top three picks in the theme
Here are the three companies I'd put at the center of this theme, one per layer. For each one I walk through what it does, the numbers I check first, why it could work, what could go wrong, and what I'd watch next. Prices and statistics are approximate as of November 3 to 4, 2026, move daily, and come from several data providers that don't always agree, so check a live quote before relying on any of them. Notice one thing right away: all three are roughly 55% below their 52-week highs.
| Coherent | Corning | Fabrinet | |
|---|---|---|---|
| Ticker | COHR | GLW | FN |
| Layer | Lasers and modules | Fiber | Manufacturing |
| Price (approx.) | ~$195 | ~$120 | ~$346 |
| From 52-week high | -56% | -56% | -54% |
| P/E (trailing) | ~47 | ~50 | ~26 |
| Market cap (approx.) | ~$38B | ~$102B | ~$12B |
| Latest quarter revenue growth | ~+50% | +19% (optical +44%) | ~+51% |
| Analyst avg. target | ~$388 | ~$180 | ~$597 |
Coherent (COHR): the broadest light-maker
What it does. Coherent makes the lasers, the optical components, and the transceivers that sit at the very start of an optical link, and it's one of the few companies that controls so much of that chain itself, from growing indium phosphide wafers to selling finished 800G and 1.6T modules. It also makes industrial lasers and other materials products, though AI data centers now drive the story. It joined the S&P 500 in March 2026 and posted record fiscal 2026 revenue of $7.12 billion, up 22.5%.
The latest numbers. On November 4 it reported fiscal first-quarter revenue of about $2.37 billion, roughly 50% higher than a year ago, and adjusted earnings of $2.09 a share versus about $1.97 expected. It guided the next quarter to $2.55 billion to $2.75 billion of revenue and $2.28 to $2.48 of adjusted earnings per share, with gross margin of 40% to 42%, and its CEO said fiscal 2027 pro forma revenue growth should exceed 60%. Early reports on the stock's reaction were mixed, after it fell sharply going into the report.
Price
~$195
From 52-wk high
-56%
P/E (trailing)
~47
Market cap
~$38B
The bull case.
- It sits at the start of the chain. Whoever makes the lasers controls a scarce, hard-to-copy input, and the move to 1.6T and co-packaged optics needs more of them.
- Vertical integration. Making its own wafers and lasers can give better control of supply and margins than companies that buy them.
- Optionality. Its PhotonLink platform for co-packaged and near-packaged optics, and an optical circuit switch business, give it a seat at the next technology shift, not just the current one.
- Guidance is rising. Management pointed to growth above 60% for fiscal 2027 on a pro forma basis.
Watch out. The stock is about 56% below its June high of $440 and still trades near 47 times trailing earnings, so expectations remain high. Guided gross margin of 40% to 42% is well below the 50%-plus some rivals report, since Coherent sells more finished modules and lower-margin products. Executives and directors sold shares in March through May 2026, including the CFO at about $218 in early March and a director near $355 in May, though I did not find CEO selling. Analysts are split: the average target is roughly $388, but they range from $262 to $500, and JPMorgan trimmed its target to $385 from $465.
What I'd watch next. Whether gross margin climbs from the 40% to 42% range as 1.6T ramps; progress converting the co-packaged optics platform into real orders; how much of the optical circuit switch backlog (reported above $400 million) turns into revenue; and what hyperscalers say about spending when they next report.
Corning (GLW): the glass that carries the light
What it does. Corning is a 175-year-old materials company best known for Gorilla Glass on phones, but its Optical Communications business is now the engine: optical fiber, cables, and connectivity hardware, including new high-density products built for AI data centers. It makes the glass strands that carry the light between racks and buildings, and demand for fiber per data center is rising as AI clusters grow.
The latest numbers. In the third quarter, reported October 27, core sales rose 19% to about $5.08 billion and core earnings rose 36% to $0.91 a share. Optical Communications sales jumped 44% to about $2.38 billion and its net income rose 80% to $532 million, with sales of its generative AI products more than doubling from a year ago. It guided fourth-quarter core sales to $5.2 billion to $5.3 billion and core earnings to $0.96 to $1.00 a share, and said it has passed the $20 billion core sales run-rate goal it set.
Price
~$120
From 52-wk high
-56%
P/E (trailing)
~50
Market cap
~$102B
The bull case.
- Fiber is the least replaceable part. Every optical link, whatever laser or module sits on the ends, needs glass in the middle, and Corning is the best-known maker with decades of manufacturing know-how.
- It's not just a photonics bet. Corning's other businesses give some cushion that pure-play optical names lack.
- Demand is ahead of supply. Management says availability of its newest high-density products is constrained, and it's spending about $2 billion this year, mostly to add optical capacity.
- Earnings are growing faster than sales. Core earnings rose 36% on 19% sales growth in the quarter, which means the optical mix is lifting profits.
Watch out. A roughly $102 billion company at about 50 times trailing earnings leaves little room for disappointment, and the stock is about 56% below its 52-week high of $271.78. Corning is also still building a currently loss-making solar business and said it isn't setting a new margin target while it adds optical capacity. And it was a lead casualty in the last optical bubble, falling more than 90% from its 2000 peak. CEO Wendell Weeks sold shares repeatedly in 2026, per filings tracked by aggregators: about 137,500 shares near $155 in February, about 165,600 near $168 in April, and 100,000 near $186 in June, partly tied to option exercises. The average analyst target is about $180, with a range of $132 to $220.
What I'd watch next. Optical segment growth versus the 44% in the last quarter; whether new capacity arrives on time and at good returns; margins as solar losses and optical expansion costs weigh; and any sign that hyperscalers are pausing fiber orders.
Fabrinet (FN): the contract manufacturer behind the modules
What it does. Fabrinet doesn't design its own brands. It builds optical and electronic products for other companies, including the transceivers, lasers, and optical components that go into AI data centers, working in factories in Thailand and elsewhere. Think of it as the specialist assembler many of the big names rely on, and it makes money from volume and manufacturing skill instead of owning technology that could go out of fashion.
The latest numbers. In its fiscal fourth quarter (reported in August), revenue hit a record $1.316 billion, up 45%, and non-GAAP earnings were $4.10 a share. Data center revenue was $669 million, up 68% and about half of the total. In the quarter reported November 3, revenue rose about 51% to roughly $1.47 billion versus about $1.40 billion expected, and it guided next-quarter earnings to $4.77 to $4.92 a share. The stock jumped about 15% on the report and then gave back some of it the next day.
Price
~$346
From 52-wk high
-54%
P/E (trailing)
~26
Market cap
~$12B
The bull case.
- Technology-agnostic. If the winning module design changes, Fabrinet can often still build it, so it's less exposed to who wins the laser race.
- The cheapest multiple of the three. At around 26 times trailing and roughly 18 times forward earnings, it is far less expensive than Coherent or Corning, while growing about as fast.
- Growth is broad. Revenue rose about 36% for the full fiscal year, and data center is now about half of sales.
- Small and focused. A roughly $12 billion market cap means a good outcome moves the stock more, which cuts both ways.
Watch out. Customer concentration is the big one: four customers each made up 10% or more of fiscal 2026 revenue (Cisco 20%, Nvidia 16%, Nokia 11%, Amazon 11%), so losing or shrinking one would hurt. Contract manufacturers earn thinner margins than the companies they build for, and they can be squeezed on price. The stock is about 54% below its $749 high, and analysts disagree sharply: the average target is about $597 but ranges from $385 to $800, Barclays cut its target to $496 from $739, and B. Riley stayed neutral with a $364 target. Recent 2026 insider filings I found were mostly shares withheld to pay taxes on vesting, not discretionary sales.
What I'd watch next. Whether growth holds as customers' orders change; the mix between data center and other segments; how its largest customers' plans, especially Nvidia's, evolve; and margins as it adds capacity.
Two names I'm deliberately leaving out
Lumentum. It's one of the best-known optical names, and a few days ago its story changed. On November 2, Marvell agreed to buy it in an all-stock deal at a fixed ratio of 4.5 Marvell shares per Lumentum share, a premium of about 24% to its October 28 close, with the deal expected to close in the first half of 2028. Lumentum fell almost 15% that day, and because the price is tied to Marvell's stock, its value now moves with Marvell's shares, not just its own business. Until the deal closes, it's a bet on Marvell more than on lasers, which is why I'm skipping it.
Ciena. It makes the optical networking systems that tie data centers together. Its latest quarter had revenue of $1.67 billion, up 37%, an $8.5 billion backlog, and record margins. But it is expected to report next in early December, the stock has been volatile, and at roughly 78 times trailing earnings I'd rather see the next report first. It's on my watchlist.
Side by side
- Coherent is the most direct play on the laser and module chain, with the most technology upside and the most to prove on margins.
- Corning is the largest and steadiest, and the one with the highest valuation relative to its growth.
- Fabrinet is the cheapest and the most dependent on a few big customers.
- What they share: record or fast-growing results, rising guidance, roughly 55% drawdowns from their highs, and a heavy dependence on how much cloud companies keep spending on AI.
8. The risks, and how I'd own it
- The funding question. The October selloff began after Broadcom's CEO warned that access to capital could slow AI buildouts, and deepened on October 30 when Big Tech earnings raised fears about hyperscaler spending. Optical suppliers sell to those customers, so a slowdown shows up in their orders.
- The year 2000. The last optical boom ended with leaders falling more than 90%. This cycle looks healthier, with real backlogs and real profits, but shortages are exactly when companies build too much capacity.
- Customer concentration. A few cloud and networking giants account for most of the orders. If one pauses, everyone feels it.
- China. Chinese optical makers are strong, cheaper in some products, and also sold off hard in October.
- Technology shifts. Co-packaged optics could change which companies capture the profit, and a module maker today is not guaranteed a place there tomorrow.
- Valuation and volatility. Coherent and Corning still trade near 50 times trailing earnings after falling by half. These stocks can move 10% in a day, as they did several times in October.
- Insider selling. Corning's CEO sold roughly $68 million of stock in 2026, per filings tracked by aggregators, and several Coherent executives and directors sold. It's common with stock-based pay, but notice it.
- Consolidation. Deals like Marvell and Lumentum can reshape the group and change what a stock is really a bet on.
How I'd think about owning it:
- Size for a 50% drop, because these stocks have already had one. If a position would hurt too much at half the price, it's too big.
- Build it in pieces. Buying over several months, not all at once, helps with volatility like October's.
- Spread across layers, not three versions of the same bet, and consider that a diversified fund can give you exposure without choosing a winner.
- Watch backlog, lead times, and hyperscaler spending plans, not just headline revenue. Those turn before earnings do.
- Keep it a slice, not the whole portfolio.
The bottom line: the move from copper to light inside AI data centers looks real, and these businesses are growing fast, but a strong theme doesn't make every stock in it a good buy at every price, and the last optical boom is a reminder of how this can end. That part is up to you.
More soon,
Learn to Love Money
NOT FINANCIAL ADVICE. This post is an educational opinion piece and is not a recommendation to buy, sell, or hold any security. Figures are from company filings and third-party data providers as of November 3 to 4, 2026, come from sources that do not always agree, and change daily. Market capitalizations and trailing P/E ratios are approximate and partly my own calculation from reported prices and share counts. Analyst targets are opinions, not guarantees. Do your own research or talk to a licensed advisor before investing. This post was prepared with AI assistance; see our AI Disclosure.
Sources
- Company filings (SEC EDGAR, 8-K): Coherent, Corning, Fabrinet, Lumentum
- Coherent results: Benzinga, GuruFocus
- Corning results: Corning investor relations, GuruFocus call highlights
- Fabrinet results: Q4 FY2026 8-K, Yahoo Finance call highlights
- Marvell and Lumentum: CNBC, The Motley Fool
- Ciena: Q3 revenue and backlog (StockTitan)
- October optical selloff: 24/7 Wall St, SimpleTechTrend, South China Morning Post
- Sector background: BBAE: The 2026 photonics stocks rally
- Insider trades, analyst ratings and price targets (MarketBeat): Coherent, Corning, Fabrinet
- Analyst consensus (S&P Global via StockAnalysis): Coherent, Corning, Fabrinet
- Prices, P/E, market cap and 52-week ranges: Google Finance, Robinhood, CNBC and Yahoo Finance quotes (approximate; check a live quote).