Here’s the thing about the AI boom that nobody talks about at dinner parties: it’s not really one race. It’s a relay race where each leg creates a new batch of winners.
First, it was GPUs. Nvidia printed money. Then it was the servers to hold those GPUs. Then cooling systems, because you can’t just stack thousands of processors without turning your data center into a pizza oven. Then power plants—suddenly nuclear energy became sexy. Then memory. Each bottleneck got solved, and each solution made someone very, very rich.
Now? The next bottleneck is forming, and it’s hiding in plain sight: the cables and chips that move data between GPUs inside data centers.
Think about it. A GPU costs tens of thousands of dollars. If it’s sitting idle waiting for data to arrive, that’s money literally burning. As hyperscalers scale from thousands of GPUs to hundreds of thousands, the internal plumbing becomes the constraint. The data has to move *fast*, or the whole thing grinds to a halt.
**The Great Copper vs. Optics Showdown**
Here’s where it gets interesting. There are two ways to move data: copper cables and fiber optics.
Copper is the incumbent. It’s cheap, fast, and power-efficient for short distances. But there’s a physics problem: at today’s speeds (800G), copper cables only work reliably for about three meters. That’s basically one rack. As speeds push toward 1.6T, copper’s range shrinks even more.
Optical fiber doesn’t have distance limits, but it’s expensive, power-hungry, and adds latency because you have to convert electrical signals to light and back again. It’s overkill for connecting GPUs in the same rack, but it’s the only practical option for connecting clusters across a data center.
There’s also a middle ground: active electrical cables (basically copper with smart chips embedded). They extend copper’s range to 7-10 meters while using less power than optics. It’s copper’s last stand before physics wins.
**The Timeline That Matters**
Here’s the kicker: this isn’t a binary choice. It’s a sequence.
Copper dominates *right now* for short-distance connections. Optics handles the longer distances. But in a few years—probably 2027-2029—co-packaged optics (CPO) technology will mature. This integrates photonics directly onto chips, eliminating the power and latency penalties. When that happens, optics wins everything.
Nvidia just dropped $4 billion betting on this timeline, splitting the money between Lumentum and Coherent. That’s not casual. That’s a supply chain lockup.
**The Stocks to Watch**
For the next 2-3 years (copper phase): Credo Technology, Marvell, Broadcom, and Amphenol.
For the long term (optics phase): Lumentum, Coherent, Fabrinet, and Applied Optoelectronics.
If you want one name that threads both phases without timing risk? Marvell. It plays in copper today and has optical silicon for tomorrow.
**The Bottom Line**
The AI infrastructure boom isn’t ending. It’s just moving to the next constraint. The hyperscalers have identified it. The supply chain is tight. The winners are already visible.
Get positioned before the market figures it out. History shows the biggest gains go to investors who move before consensus catches up.