For two years, the AI hardware trade has been about raw power. Faster chips. More memory to keep them fed. And, the market has rewarded companies supplying both legs.
There is a third piece that has gotten far less attention, and it is becoming just as important: moving data between the chips.
Modern AI is no longer a single processor doing the work. It is hundreds of thousands of chips that have to operate as one system, passing enormous volumes of data back and forth constantly. As these systems scale, the connections between chips increasingly determine how well the whole thing performs. And the electrical connections the industry has relied on for decades are running into hard physical limits on how much data they can carry, how far, and at what energy cost.
The answer is light. Replacing electrical signals with optical ones is the basis of one of the most talked-about and least understood themes in the market: photonics and optics.
For decades, data inside and between computers moved as electrical signals over copper. It was cheap, well understood, and good enough. Due to its energy intensity, AI has maxed out copper as a reliable use case.
The problem is physics. As data rates climb toward 800 gigabits and 1.6 terabits per second, copper connections lose signal integrity over distance, leak power, and throw off heat. Every electrical link in a large AI cluster needs power-hungry components to clean up and retime the signal. At the scale of a modern AI factory, those inefficiencies compound into a wall: too much energy spent moving data, not enough left for computing on it.
Optical connections solve this. Light moves more data, over longer distances, with dramatically less power and heat. The catch is that converting between electrical and optical signals has historically been expensive, bulky, and reserved for long-distance links. The entire photonics investment thesis rests on one shift: optics is moving from the edges of the network to the heart of the chip.
Like memory, optics is not a single product. It is a handful of distinct businesses, and the easiest way to understand them is to follow the data on its journey: something has to create the light and translate the data into it, something has to do that translation closer and closer to the chip itself, someone has to manufacture all of it, and something physical has to carry the light from one place to another. Here is each step, and the companies that own it.
Computers think in electricity, but the fastest way to move data is with light. So at both ends of every optical link, something has to convert electricity into light and back again. That device is called a transceiver, and it is the workhorse of the network today. Every connection needs them, a large AI cluster needs an enormous number of them, and each new generation runs faster, from 400 gigabits per second to 800 and now toward 1.6 terabits.
The hardest part to build well is the laser inside that generates the light, which is a genuine specialty where manufacturing skill creates real barriers to entry.
The largest makers of these modules anchor the trade. Zhongji Innolight (300308 CH) and Eoptolink (300502 CH) are among the leading high-speed transceiver suppliers to the AI data center buildout. Coherent (COHR) is one of the dominant makers of transceivers and optical components. Lumentum (LITE) is a leader in the lasers that power them, and management has called the current period a breakout for laser demand tied directly to AI. Applied Optoelectronics (AAOI) and MACOM (MTSI) add further exposure to lasers and optical components.
Why it matters: this is the largest and most established layer of the trade, and it grows directly with the number of connections in an AI cluster.
Today, the optical components plug into the front of a switch, like a cable into a port. The next step is to move the optics directly onto the same package as the chip itself. That approach is called co-packaged optics, or CPO, and the payoff is real: far less power used, less heat, lower delay, and better reliability at the scale AI now demands.
This is where the high-speed connectivity names are moving. Marvell (MRVL) is a dominant force in the networking silicon that AI clusters depend on and is building co-packaged optics directly into its roadmap. Ciena (CIEN) and Credo Technology (CRDO) provide further exposure to the high-speed connectivity and interconnect shift that co-packaged designs represent.
Why it matters: this is the architectural shift driving the excitement in the sector, and once the industry designs optics into the chip, it does not go back.
For optics to reach the volumes AI demands, the components have to be precisely assembled and the underlying materials and process tools have to be supplied at scale. A key enabler here is silicon photonics, the ability to build optical parts using the same manufacturing processes as conventional chips, which helps bring optics toward the cost and scale of the broader semiconductor industry. This is the manufacturing base beneath the entire trade.
Fabrinet (FN) is one of the most important contract manufacturers of optical modules, physically building the products that the rest of the industry designs. MKS Inc (MKSI) supplies the process technology and materials that optical and semiconductor manufacturing depend on.
Why it matters: this is the picks-and-shovels layer, the manufacturing base the entire buildout depends on.
Finally, something has to physically carry the light from one component to another, the fiber, the connectors, and the precision materials that join everything together with extraordinary accuracy. It is the least glamorous layer and one of the most essential.
This is where fiber and materials specialists lead. Corning (GLW) is one of the most important optical fiber makers in the world. Sumitomo Electric (5802 JP) and Fujikura (5803 JP) are among the largest optical fiber and cable makers globally, holding large shares of a market that demands manufacturing accuracy measured in fractions of the width of a human hair.
Why it matters: this layer tends to be steadier than the rest, earning its place through long design cycles and deep customer relationships rather than chasing spot prices.
Optics has historically been a brutally cyclical business, tied to telecom spending booms and busts. This cycle could be signaling a new regime.
The demand driver is structural. Optical networking's share of data center capital spending has risen sharply as AI clusters scale, moving optics from an afterthought to a co-equal infrastructure layer alongside compute and memory. As long as AI clusters keep growing, the interconnect problem keeps growing with them.
Additionally, the architectural shift toward co-packaged optics is a one-way door. Once the industry designs optics directly into the chip package, it does not revert to the old pluggable model. That creates a durable, multi-year transition rather than a one-time upgrade.
AI has been measured by the power of its chips and the size of its memory. But performance at scale depends just as much on how quickly those chips can move data between one another, and that is where the industry is now hitting the limits of electricity over copper.
Light is how it breaks through. The companies that make the lasers, transceivers, silicon photonics, and physical connections carrying AI's data are the layer of the trade that most investors sense but few can name.
Photonics has long been viewed as a niche corner of the chip world. In the age of AI, it is becoming essential infrastructure.
This information is provided solely as general investment education. None of the information provided should be regarded as a suggestion to engage in or refrain from any investment related course of action. Investing involves risk, loss of principal is possible.
Not an offer: This document does not constitute advice or a recommendation or offer to sell or a solicitation to deal in any security or financial product. It is provided for information purposes only and on the understanding that the recipient has sufficient knowledge and experience to be able to understand and make their own evaluation of the proposals and services described herein, any risks associated therewith and any related legal, tax, accounting or other material considerations. To the extent that the reader has any questions regarding the applicability of any specific issue discussed above to their specific portfolio or situation, prospective investors are encouraged to contact 1-855-561-5728 or consult with the professional advisor of their choosing.
Forward-looking statements: Certain information contained herein constitutes “forward-looking statements,” which can be identified by the use of forward-looking terminology such as “may,” “will,” “should,” “expect,” “anticipate,” “project,” “estimate,” “intend,” “continue,” or “believe,” or the negatives thereof or other variations thereon or comparable terminology. Due to various risks and uncertainties, actual events, results or actual performance may differ materially from those reflected or contemplated in such forward-looking statements. Nothing contained herein may be relied upon as a guarantee, promise, assurance or a representation as to the future.
Use of Third-party Information: Certain information contained herein has been obtained from third party sources and such information has not been independently verified by Roundhill Financial Inc. No representation, warranty, or undertaking, expressed or implied, is given to the accuracy or completeness of such information by Roundhill Financial Inc. or any other person. While such sources are believed to be reliable, Roundhill Financial Inc. does not assume any responsibility for the accuracy or completeness of such information. Roundhill Financial Inc. does not undertake any obligation to update the information contained herein as of any future date.
Any indices and other financial benchmarks shown are provided for illustrative purposes only, are unmanaged, reflect reinvestment of income and dividends and do not reflect the impact of advisory fees. Investors cannot invest directly in an index.
Except where otherwise indicated, the information contained in this presentation is based on matters as they exist as of the date of preparation of such material and not as of the date of distribution or any future date. Recipients should not rely on this material in making any future investment decision. The performance data quoted represents past performance. Past performance does not guarantee future results. Current performance may be lower or higher than the performance data quoted.