AI's biggest problem isn't smarter software. It's power.
Each Nvidia chip burns energy to send data through copper wire. And copper is running out of room. So Nvidia is betting on light.
In the past three months, Nvidia has put at least $6.5 billion into photonics firms. Photonics moves data with light instead of electricity. Light loses far less energy on the way. That eases the power problem inside today's data centers.
The cash went to a few names. Nvidia put $500 million into Corning for optical parts. It joined a $500 million round for a startup called Ayar Labs. The rest went to Lumentum, Coherent, and Marvell. All three build the same light-based tech.
So what does the tech actually do? It uses light to move data between chips, memory, and whole data centers. Today, most of that still runs on copper. Copper is cheap and steady, but it draws a lot of power. Light draws much less.
Think of copper like shouting across a loud room. It works, but it costs energy. And you hit a limit fast. Light is more like a laser pointer. It moves fast and wastes almost no power.
We break down the AI buildout every morning in Market Briefs - five minutes a day. Join and you also get a free investing masterclass.
AI models keep getting bigger. Bigger models make chips trade huge amounts of data. Push that through copper and the power bill soars. At some point it caps how much you can grow.
That cap is what Nvidia wants to dodge. Forrester analyst Alvin Nguyen says copper would slam the firm into a wall. Funding light-based tech now keeps that wall away.
Nvidia is already moving. It has put some of this tech into the gear that links big clusters of chips. At its GTC event in March, CEO Jensen Huang said the firm is starting to scale it. He added that Nvidia needs far more of it than the world makes today.
Nvidia isn't the only buyer. Rival AMD joined the same Ayar Labs round. It also bought a photonics startup last year. The venture arms of Alphabet and Microsoft backed another optics startup in April.
The tech works in the lab. Building it at scale is the hard part. The parts must line up just right. One slip, and the whole piece gets tossed.
That's why Morningstar's Brian Colello expects copper to stay on top for now. Light should take over slowly as data needs rise. Nick Patience of the Futurum Group agrees the tech is sound. The hard part, he says, is making it at scale.
Most experts don't see light-based chips going big until 2028. Nvidia is spending now so the supply is ready. Traders have already bid up the makers. Lumentum has more than doubled this year. Coherent is up about 96%, and Marvell has roughly doubled too. The folks piling in think the shift comes sooner.
Want the AI trades worth watching early? Join 350,000+ investors reading Market Briefs and get a 45-minute investing course free when you sign up.
