5 stocks that can profit from the next wave of AI infrastructure
The first phase of the AI boom was dominated by chips. Demand for computing power made Nvidia $NVDA one of the most valuable companies in the world and carried other accelerator makers along with it. But as AI clusters grew, the boom hit a different limit. Hundreds of thousands of fast processors are useless if data doesn't flow between them just as fast. When they wait for the network, expensive chips spin idle.
Key points
42 times earnings on one end, 11 on the other. Both companies benefit from the same shift, as AI clusters have outgrown the speed at which data can flow between them. Yet for the same limit, the market charges five times differently.
Nvidia itself invested $2 billion in Lumentum this year and another $2 billion in Coherent, including multi-year purchase commitments. This effectively put the availability of optical components on par with memory supply.
Lumentum reported revenue growth of +109% last quarter, but a full-year accounting loss of almost $7 billion. The loss isn't a sign of trouble—the company is operationally profitable; it just can't manufacture fast enough.
HPE reports networking growth of +74.9%, but after adjusting for the Juniper acquisition only about 10% remains, and the data center segment even declined. The group's cheapest stock at 11 times earnings has a catch that's invisible in the headline number.
Arista earns an incredible 49.9% at the operating level, but you'll pay 42 times earnings per share. And Nvidia is now pushing into its most lucrative market with its own Ethernet.
As transmission speed and distance increase, traditional copper interconnects can't keep up, and their role is being taken over by optics, which are moving ever closer to the chip itself. Attention that once belonged exclusively to GPUs is therefore shifting to the companies that build the interconnects between them. That this isn't a marginal layer was confirmed by Nvidia itself when it invested a combined $4 billion in two optical suppliers this year.
Deutsche Bank recently named this shift and initiated coverage of five companies that could profit from it. They share one physical barrier, but their valuations diverge dramatically: from 11 times expected earnings to 42 times. The same bet on AI networking thus costs five times differently.
The five are Lumentum $LITE, Coherent $COHR, Arista Networks $ANET, Cisco $CSCO, and Hewlett Packard Enterprise $HPE. Each offers a different mix of growth, margin, and price, and that difference is where the entire decision of where to bet lies.
Optics aren't new; what's new is their share of the rack's cost
Optical interconnects in data centers aren't new—telecom networks have run on them for decades. What's changing is where optics sit and how much of the total system value they represent. As long as servers communicated across a corridor, copper sufficed. But AI training connects tens of thousands of accelerators that must work as a single machine, and here both distance and data volume hit copper's physical limit. The signal weakens faster as transmission speed increases, so with the move to 800 gigabits and on to 1.6 terabits per second, optics are pushing ever closer to the chip itself.
This transition doesn't just increase the number of optical modules per rack; it also changes their type and price. Alongside traditional pluggable transceivers come optical switches, external laser sources, and so-called co-packaged optics, where optics are integrated directly into the switch chip package. Each of these layers is more expensive and more complex to manufacture than the previous one, which explains why laser suppliers suddenly report year-over-year revenue growth in the tens of percent.
How strategic this shift is was shown by Nvidia itself. In early March, it invested $2 billion in Lumentum and the same amount in Coherent. Both deals aren't just capital injections: they include multi-billion-dollar purchase commitments and priority access to future production capacity. Nvidia effectively admitted that it considers optical component availability as critical as memory supply or chip packaging, areas it had previously invested in as well. Both partnerships are non-exclusive, so both companies can continue supplying competitors.
This investment, however, carries a double reading. On one hand, it confirms the limit is real and demand for optics will grow for years. On the other, much of that future is already priced into the stocks. Lumentum's shares rose roughly tenfold over twelve months, Coherent's value nearly quadrupled over the same period. That's why for each of the five companies, the key question isn't just how fast it's growing, but how much the market already charges for that growth.