Copper cannot do what AI needs it to do. At 10,000-GPU scale, copper interconnects overheat, create latency, and cap bandwidth. Optical transceivers are the only solution. The AI data centre buildout is not a GPU story alone. Every Blackwell rack, every Vera Rubin deployment, every hyperscaler cluster needs optical modules to function. AAOI is up 488% this year. LITE is up 180%. COHR has doubled. The market is pricing in exactly this reality.

Here is the number that tells you this is structural and not just a trade: Goldman Sachs just raised its 2026 demand forecast for 800G transceivers to 33.5 million units, up 58% from its prior estimate. The global market hits USD 26 billion this year. EML laser supply is completely allocated. You cannot build more optical modules than your laser supply allows. Constrained supply plus exploding demand equals sustained pricing power. This is not a short squeeze. This is a supercycle.

The short-term traders will come and go — AAOI dropped 14% on May 7 before rallying 8% on May 13. That volatility is real. But the structural demand is real too. Ask yourself one question: is AI compute going to need MORE optical bandwidth next year than this year? The answer is obviously yes. That is why I am watching COHR, LITE, and FN for the next pullback entry. 📈