The copper wall is back, and two companies are standing on it

Hey {{first_name|Investor}} -

Before we begin, you can review the Photonics in Plain English guide published in April 2026 to learn more about the photonics supply chain and its role in AI infrastructure through light-based connectivity. Any free subscriber can access.

Open up an AI server and the story everyone tells is about the chips. The story that matters almost as much sits in the space between them: the wiring that lets thousands of chips talk to each other. Right now two companies, Astera Labs ($ALAB) and Credo Technology ($CRDO), make their living pushing that wiring, plain copper, as far as it can possibly go. Astera's switching products went from almost nothing to a serious business in a single year. Credo grew its sales more than 200% over its last fiscal year on the strength of smart copper cables. Watching these two is the clearest way into the theme I keep circling back to, because they are standing right on the wall this whole buildout is about to hit.

For decades, data centers wired their machines together with copper, the same kind of metal in the ethernet cable behind your router. Copper was cheap and good enough. The AI era broke that. Training a modern model means thousands of GPUs working at the same time, and at the cluster sizes the big cloud operators are building in 2026, copper runs into hard physical limits. It loses signal over distance. It runs hot. And it draws power at a scale that now shapes the cost of running a data center directly.

This is why Astera and Credo are worth watching closely. Their entire job is to squeeze more speed and more distance out of copper, with clever chips that catch a fading signal and rebuild it. They are very good at it, and the AI buildout keeps handing them more work. But every year the wall gets a little higher, and at some point the smartest copper in the world stops being enough.

When copper runs out of room, the answer is light. Optical connections carry data as pulses of light through hair-thin glass fiber, traveling farther on less energy and far less heat. Picture copper as shouting across a crowded room: fine up close, and a mess once the room gets big. Optical is a clean phone call, clear and efficient whatever the distance.

The catch is that the parts which turn electricity into precisely controlled light, and back again, are hard to build at scale. That difficulty is the whole opportunity. The shift happening through 2025 and 2026 moves the workhorse parts, the transceivers (the modules that convert electrical signals into light and back), from 800 gigabits per second to 1.6 terabits, with 3.2T already on the roadmaps. For scale, your home broadband is measured in megabits, so 1.6T is on the order of a million times faster. Industry tracker TrendForce expects shipments of the fastest transceivers to rise about 2.6 times in a single year. That is the demand curve pulling this whole theme forward.

The mistake most people make is shopping for "the one photonics stock." There isn't one. It helps to see the supply chain as four jobs, because different companies own different jobs and they carry very different risks.

Something has to make the light, which means the lasers and the exotic materials they are grown from. Something has to load data onto that light and read it back off, which means the modulators, detectors, and silicon that drive it. The pieces converge into a finished transceiver or a switch that goes into the rack. And every seam along the way has to be tested before anyone trusts it in a live network.

Hold onto that fourth job, testing, because it is the one most people skip and the one I find most interesting. I'll come back to it.

If you want to know where the real shortage is, watch who pays up front to lock in supply. The clearest signal in this whole theme came when NVIDIA invested $4 billion across two optical suppliers, Lumentum ($LITE) and Coherent ($COHR), each deal carrying a multi-year purchase commitment and rights to future manufacturing capacity. The company building the GPUs paid, in its own cash, to reserve the optical parts that feed them. When the biggest buyer in the market prepays to reserve next year's harvest, that tells you where the bottleneck sits.

That is the difference between a story and a setup. The money is already moving toward the parts that make and carry the light.

Here is the part worth slowing down for. The makers of optical parts are in an architecture race, several competing ways to package the optics, and nobody knows yet which design wins. That uncertainty makes the component bets harder. But one layer gets paid regardless of which design wins: testing.

Light fails differently than electricity. An electrical signal mostly passes or fails a check. Light degrades as loss and noise across a range, so an optical part carries far more points that have to be measured, and more of them appear as speeds climb. A single transceiver that fails inside a running cluster can idle thousands of GPUs, which is real money burning. The companies that build the test instruments collect more per wafer and per deployment as speeds rise, whichever architecture ends up on top. The makers fight over the design. The testers get paid either way.

The name I'd anchor that layer with is Viavi Solutions ($VIAV). It makes the instruments that certify optical gear works, at the factory, at installation, and in the live network, so it gets paid whoever wins the architecture race. I hold it and have disclosed it before, and it's the cleanest expression of the whole idea: own the layer that gets paid whichever design comes out on top.

That insight, bet the bottlenecks and let the architecture race sort itself out, is the spine of how I'm thinking about this theme. The rest of the names at each layer, with the reasoning and the risks behind them, are what I'll walk through this Sunday.

One risk deserves more of your attention than any single company, and it lives at the very bottom of the stack. The optical lasers are grown on a special wafer made from a metal called indium, and both that metal and the finished wafer are concentrated in China, which controls roughly 70% of the world's supply and now requires export licenses to ship it. Since those controls came in last year, the price of one of these wafers has more than tripled. And because that control is political, it tracks the broader U.S.-China relationship, which means it can tighten as easily as it eases.

This matters because it sits under everything. Whichever optical design wins, it runs on the same material, so the bottleneck binds either way. And scarcity like this flows in one direction: the companies that have to buy the material pay more and can build only as fast as supply allows, while whoever controls the material collects. That can slow how fast the whole theme grows and squeeze the margins of the companies buying the material, even ones that look sold out.

The encouraging part for the buyers is that the strongest players are building their own supply rather than depending on China, and that is a big part of what separates a durable name here from a fragile one. I'll name the companies that own this layer, and the ones best insulated from it, in Sunday's full breakdown. It's the single risk I'd want anyone new to this theme to understand first.

If picking individual names feels like a lot, you can get exposure to this theme through funds rather than single stocks. A few broad semiconductor and AI-infrastructure ETFs hold many of the component and networking-silicon companies that ride the optical buildout, alongside the GPU makers driving the demand. There are also narrower funds focused on networking and connectivity.

Two things to understand before you go that route. First, a broad chip or AI ETF gives you the theme diluted, since photonics will be one slice of a fund that also holds plenty of unrelated semiconductors. Second, the purest photonics names, especially the smaller testing and materials companies, often sit at tiny weights or aren't in these funds at all, so a fund can underweight exactly the part of the chain I find most interesting. It's a tradeoff between owning the whole theme loosely and owning a few parts of it deliberately, and that choice is yours to make.

There's now a more targeted option taking shape. Roundhill, the shop behind the memory-focused DRAM fund, has filed for a dedicated photonics and optics ETF under the ticker LYTE, built to hold the optical-interconnect, laser, and photonic-chip names directly. As I write this it has been filed but isn't trading yet, and its holdings aren't public, so treat it as one to watch rather than buy: confirm it's actually listed and trading, then read its holdings and weights once they're out before deciding whether it gets you the exposure you want.

As always, look under the hood at a fund's actual holdings and weights before deciding.

Sunday I'll open up the specific companies at each of the four jobs, the two I'd anchor on, and the ones I'd keep on a watchlist rather than buy. Same map, real names.

My first book, Make Your Own Alpha, is on its way. It's the full version of the approach behind these issues: how to build your own portfolio deliberately enough to make work optional, without a finance background or a windfall to start. I'll share a launch date once it's set.

If you want first word on that date, plus the special launch bundle that includes the e-book and companion spreadsheet workbook, get on the launch list.

Stay disciplined - Koh

Disclosure: I own positions in $LITE, $COHR, and $VIAV.

Disclaimer: Nothing in this newsletter constitutes investment advice or a recommendation to buy or sell any security. Numbers and observations are as of publication. I may hold positions in companies discussed above. Always do your own research and consult a licensed financial advisor before making investment decisions.