The optical names are quietly becoming AI infrastructure compounders

I spend most of my time looking for businesses that can compound for years, not trade for weeks, and the optical networking space is starting to look like one of those setups.

The simple version: AI clusters need to move enormous amounts of data between chips, and they do it with light. Every generation needs faster transceivers, from 800G now toward 1.6T, and each step up needs more advanced lasers. The demand is not a one off. It scales with every new data center that gets built.

What makes this interesting as a long term holding rather than a trade is the supply side. The critical component, the EML laser, is genuinely hard to make at volume. Lumentum posted record revenue up 65% with a data center book-to-bill above 4x and lasers sold out through 2027. Coherent crossed into the S&P 500 on the back of data center growth and took a multi billion dollar Nvidia investment. Fabrinet, which assembles a huge share of the world's transceivers, is adding 50% capacity. These are not hype names, they are companies with real backlogs and real margins.

The risk is the same as any AI infrastructure play: if hyperscaler capex slows, the whole chain slows with it. And these stocks have already run a long way. So I am not backing up the truck at any price. But as a multi year theme, owning the companies that control a true bottleneck, with visibility into 2027, is exactly the kind of position I am happy to build patiently and hold.