The invisible layer between every AI model and the hardware it runs on
It was unsexy work. The kind of thing venture capitalists ignored while pouring money into Pets.com. But those coatings ended up becoming one of the foundational technologies of the modern internet. Without them, wavelength-division multiplexing doesn't work. Without WDM, the fiber networks that carry every packet you send get maybe 1% of the capacity they actually deliver today.
JDSU went on to become one of the great boom-bust stories of the dot-com era. The stock hit $1,200 in 2000 and cratered to under $2 by 2002. The company that emerged from the ashes split, restructured, and eventually rebranded. Most people who lived through that era assume it disappeared.
It didn't. It got quieter and more essential.
Here's what has happened in the intervening 25 years. The optical coating expertise that started as a way to filter light in telecom fibers turned out to be exactly the technology needed for a range of things nobody was thinking about in 2000. Anti-counterfeiting features on banknotes. The 3D sensing arrays inside Face ID on your iPhone. The optical filters inside every LiDAR unit in every autonomous vehicle being tested today. The wavelength management inside the coherent optical transceivers that connect data center to data center at 800 gigabits per second and, next year, at 1.6 terabits.
And now the AI story. Because here is the piece that most AI investors have completely missed while they argue about Nvidia's margins and TSMC's capex.
When you train a frontier model on a cluster of 100,000 GPUs, the GPUs themselves are only part of the problem. Those chips need to talk to each other, and they need to talk fast. NVLink handles the short-range chatter inside a rack. But once you go beyond a single rack, you need optics. Real optics. Photons moving through fiber, modulated by coherent transceivers, filtered by precise wavelength-selective components. The bandwidth demands of AI training clusters are growing faster than the underlying optical infrastructure can be built. Meta, Microsoft, Amazon, and Google have all told suppliers the same thing this year: whatever you can make, we will buy.
The optical transceiver market was worth about $11 billion in 2023. Analysts now project it hits $38 billion by 2028, and honestly the analysts are behind the curve. Coherent (COHR) has tripled in two years on this thesis. Lumentum (LITE) has doubled. Both are now sitting at market caps that make the risk-reward less exciting for anyone showing up late.
But those two companies aren't the only ones with the expertise required. There is another player, older than both of them, with the deepest bench of thin-film optical coating IP in the world. It was one of the pieces that emerged from the old JDSU. Over the past five years it methodically reorganized itself to focus on exactly the kind of test, measurement, and optical component work that AI data center buildouts require. No announcements, no fanfare, just execution.
And it just formed a partnership to move into quantum security infrastructure, which is the next chapter of this same story. Because when quantum computers finally break current encryption standards (and they will, probably within this decade), the entire optical network layer has to be rebuilt with quantum key distribution. Which requires, you guessed it, extraordinarily precise optical coatings and filtering technology.
The company trades at a fraction of Coherent's valuation. It has been out of favor for years because its legacy telecom test business was in a cyclical trough. That trough is ending. The AI optical build-out is just beginning. The quantum security opportunity is a bonus scenario that most investors haven't even started to price in.
Its name is Viavi Solutions.
When AI training clusters get built at hyperscale, the GPUs get the credit but the photons do the work.
You're reading the open chapter.
The full Deep Dive — including the contrarian thesis, the chain reaction, the stocks we rate True North, and the catalysts we're tracking — is reserved for Compass members.
- The full contrarian thesis with our supporting data
- 3–5 specific stocks we rate True North on this theme
- Catalysts and dates we're watching for confirmation
- Saturday Deep Dive every week (52/year)
- Free Tuesday & Thursday e-letters included
- Full Compass Ratings on 4,000+ stocks
$9/mo · $89/yr · $499 lifetime. Cancel anytime. 7-day money-back guarantee.