If you’re an investor who follows US stocks, you’ve likely noticed the recent rocket-like surge in CPO optical transmission stocks.
Investors need to start getting to know this field because optical transmission is changing traditional data transmission methods and will replace existing ones in the future.
What is CPO? What are its advantage?
The idea behind CPO (Co-Packaged Optics) is to tightly integrate the optical engine and switching chip at the packaging level, shrinking the optical module that was originally located on the front panel and shortening the electrical signal traces, reducing losses and heat concentration. When the electro-optical conversion is closer to the chip core, it not only reduces the need for retiming paths but also significantly reduces space requirements and increases bandwidth density.
Huge market-driven demand
Market Mainstream
Currently, the mainstream configuration for large-scale AI data centers is 800G optical transmission modules, and it’s evolving towards 1.6T optical transmission modules. In 2024, 800G optical transmission modules will only account for about 20% of the market, but by 2026, this figure will reach 60%.
Explosive Demand
Due to the supply shortage in AI data center construction, there’s also a shortage of CPO optical transmission modules. With few suppliers to begin with, market demand is booming and is unlikely to cool down in the short term.
Potential Market Size
According to Lumentum’s forecast at OFC 2026, the AI optical communications market size will surge from $18 billion to $90 billion by 2030.
The Stances of Two AI Chip Giants
The latest developments and strategic differences between the two computing giants, Broadcom and NVIDIA, regarding next-generation transmission technologies have directly triggered a major upheaval in the optical communications sector. A deeper dive into the underlying reasons reveals a key difference in the “positioning” of their core products.
Nvidia both copper and optical
NVIDIA represents the “general-purpose GPU” approach: NVIDIA focuses on reference, general-purpose GPU architectures, and its products must cater to the most diverse customer base and various extreme application scenarios in the global market. In order to achieve the best balance between pursuing ultimate computing power and maintaining low power consumption, the physical limits of traditional copper wires have become a stumbling block. Therefore, moving fully to CPO technology has become an essential path for NVIDIA to break through physical limits.
To overcome future bottlenecks in computing power transmission, Nvidia has chosen a drastically different path—a full-scale shift towards CPO (co-packaged optics) and silicon photonics technologies. Observing Nvidia’s recent moves, it has invested a hefty $4 billion to acquire stakes in Lumentum and Coherent, two major optical communication equipment manufacturers.
This strategic move clearly indicates that Nvidia’s future development direction in scale-up (in-rack expansion) is undoubtedly a heavy bet on silicon photonics and CPO technologies. Furthermore, at ISSCC 2026, Nvidia dropped a bombshell with silicon photonics technology, showcasing 3D stacking and low-latency DWDM (high-density wavelength multiplexing) link technologies, further dissecting the next-generation CPO light source.
On March 16, 2026, NVIDIA GTC unveiled a new architecture called Feynman, featuring a Rosa CPU, Feynman Die Stacking Custom HBM, LP40 NVLink, NVLink 8 CPO, and BlueField-5 networking technology. For the transmission interface, NVIDIA adopts a dual strategy of “copper cabling and optical communication,” and will achieve full-scale mass production of the CPO through Spectrum 7 technology in the future.
On March 16, 2026 GTC, Huang said that Nvidia is also in full-speed production of Spectrum-X Ethernet switches, which are the world’s first switches to use CPO technology, which mounts optical components directly on the chip to convert electrons into photons.
At the GTC conference on March 16, 2026, Jensen Huang said: “There is a lot of discussion about whether Nvidia will move towards ‘copper wire extension’ or ‘optical extension’. Our answer is: both in parallel.”
According to the roadmap, NVIDIA will maintain a dual-track architecture of Kyber and Oberon (a Blackwell generation product primarily used for copper-wire scale-up). By the Feynman generation in 2028, Kyber systems will replace Oberon as the primary design for hyperscale scaling.
Feynman will feature the new LP40 and introduce the new Rosa CPU, BlueField-5, and the next-generation SuperNIC CX10. As for Kyber racks, there will be both copper interconnect and CPO scale-up versions. He emphasized, “For the first time, we will be scaling systems simultaneously using both copper and CPO.”
Broadcom still supports Copper Cables
Broadcom primarily provides customers with highly customized ASIC computing solutions. Because of its extremely high ASIC architecture flexibility, even when using traditional copper wire technology for short-distance transmission in the architecture design, it can still precisely meet the specific customer’s computing performance requirements while also taking into account strict cost considerations.
In this debate over technology routes, Broadcom has emerged as a leading figure in the “copper-supporting” camp. Broadcom CEO Hock Tan released a clear signal during the Q1 2026 earnings call on March 4, 2026, emphasizing that as its customer base continues to expand, customers can continue using Direct Attach Copper (DAC) technology through 200G SerDes. Tan further pointed out that this technology will advance to 400G SerDes by 2028, meaning that its XPU customers have a very high probability of continuing to use copper cables in the future.
The information revealed
Regardless of their specific strategies, both Broadcom and Nvidia, these two AI computing giants, have adopted the following approach:
- In the short term, even in foreseeable future, they will continue to support data transmission using copper cables.
- In the long term, both have invested resources and launched optical communication technologies and related products.
The Future Trend is CPO
Broadcom’s Strategy
As speeds approach 448G, the transmission of electrical signals over traditional copper wires has reached its physical limits. “Fiber optics replacing copper” is no longer a prediction, but an inevitable architectural choice.
Based on cost, compatibility, and other considerations, Broadcom will continue to support copper cables for the time being. However, Broadcom understands that the future trend is CPO, and therefore invested heavily in CPO research and development many years ago. Broadcom began related R&D projects around 2019-2020, and its first-generation, industry-leading CPO product was launched in 2021.
SerDes is the Core Technology
SerDes will be the starting point for CPO. The essence of CPO is to convert electrical signals, which would otherwise be transmitted long distances at the board level, into optical signals as early as possible. The closer the optical engine is to the chip, the smaller the margin left for the electrical link. This places almost stringent requirements on SerDes’ jitter, linearity, and bit error rate.
CPO (Content Processing) is far more than simply replacing electricity with light; it’s a complex systems engineering project encompassing laser light source integration (such as Broadcom’s external ELS solution), thermal design, packaging coupling, and yield. The stronger the SerDes capability, the shorter the electrical path, and the more controllable the system margin. This ability to “strengthen the weak” is the underlying logic behind Broadcom and Marvell’s absolute leadership in the CPO field.
448G is currently the standard speed
From an industry chain perspective, both the implementation of 448G and the commercialization of CPO are forcing the entire interconnect ecosystem to upgrade accordingly. Verifying 448G PAM4 signals is far more difficult than the previous generation, imposing stricter requirements on signal integrity, bit error rate, and channel modeling.
Therefore, test and measurement vendors have begun to deploy complete 448G verification systems. For example, manufacturers such as Keysight, Anritsu, Tektronix, and Teledyne LeCroy have launched end-to-end testing solutions for 448G PAM4, covering key indicators such as transmitter SNDR, RLM, channel S-parameters, and bit error rate. Anritsu has also collaborated with Foxconn Interconnect to complete signal integrity verification for 448G Twinax cables, further improving the testing ecosystem for high-speed interconnects.
From a broader perspective, 448G is not only a SerDes speed upgrade, but also the next-generation technological foundation for the entire AI interconnect system. As model sizes continue to expand and GPU cluster sizes continue to grow, only by continuously improving SerDes speeds and optimizing interconnect architectures can the bandwidth requirements of future AI computing systems be supported.
Who are the key players in the supply chain?
The most fundamental layer: Data transmission
Ayar Labs is the most well-known manufacturer in this sector, and we will discuss this company later. Celestial AI is Ayar Labs’ biggest competitor.
Telecom
Ciena focuses heavily on optical transmission network technology—a technology that uses light to transmit data through fiber optic cables, typically in routers and switches with speeds ranging from 400G to over 1.6Tbps, particularly in the telecommunications industry.
In 2024, Nokia acquired Infinera, an optical network equipment vendor, for $2.3 billion, finally completing its full-chain layout from data center network (DCN) to data center interconnect (DCI), completing the final piece of the puzzle for the AI era.
Leveraging its high-bandwidth and low-latency optical communication technology, Nokia became key to solving the efficiency bottleneck of GPU data center clusters, attracting Nvidia to invest $1 billion in Nokia, acquiring a 2.9% stake. Nine of the world’s top ten cloud providers use Nokia’s technology, and the largest cloud provider’s quarterly investment exceeds the annual budget of the largest telecom company.
Laser Components
Coherent (ticker: COHR) and Lumentum (ticker: LITE) are major pure-play CPO players. Lumentum is a supplier of optical and commercial laser components, while Coherent is a major silicon photonics co-packaged optics (CPO) manufacturer.
Vertical integrator
Applied Optoelectronics (ticker: AAOI), a vertically integrated fiber optic network product supplier founded by Taiwanese Americans, is headquartered in the United States. Its production and R&D bases are located in Taiwan, and it also has factories in China. It is also a mid-sized manufacturer providing vertically integrated services from laser chip design to optical module packaging, boasting the largest production capacity of 800G and 1.6T optical transceiver modules in the US.
Another company, TeraHop, which is not publicly listed, mainly develops silicon photonic transmitters and receivers, using integrated photonic chips to convert electrical signals into optical signals; typical products are the TeraHop 800G OSFP112 optical transceiver and the TeraHop 1.6T OSFP224 optical module.
Fiber Optic Connectivity
Amphenol (ticker: APH) is a major manufacturer of electronic and fiber optic connectors, cables, and interconnect systems. Corning (ticker: GLW) is a major manufacturer of optical fibers and fiber optic cables.
Connectors and Components
The main manufacturers in this segment include Samtec and Molex.
Network Communications Gears
Arista’s latest XPO (eXternal Pluggable Optics) products are pluggable optical transceiver modules for high-speed (800G and above) Ethernet networks, targeting this emerging field. Of course, Arista’s biggest competitor, Cisco, also has corresponding product solutions: Cisco recently launched 800G linear pluggable optical modules (LPO) and 1.6T optical modules for artificial intelligence networks; in addition, Cisco is also exploring CPO (Optical Module Packaging), which integrates optical modules directly into the switch ASIC, instead of using pluggable modules.
Chip desigers
In addition, Nvidia and Broadcom (ticker: AVGO) are also major market players. Marvell (ticker: MRVL) is also actively participating in this emerging field to avoid falling behind its competitors in the new era of optical transmission warfare.
Capital Market Performance
Stock performance
The stock prices of all the companies mentioned in this article, except for Broadcom which has an enormous market capitalization, have recently skyrocketed. They are listed here, and interested investors can check them out to see if what I’m saying is true.
Nvidia Invests in Lumentum and Coherent
On March 1, 2026, Nvidia announced it would invest $2 billion each in the two largest CPO optical transmission companies, Coherent and Lumentum.
This further intensifies competition in the market. Each company will receive $2 billion in investment. Nvidia stated that both investments include “billions of dollars in procurement commitments and future access to” certain laser components.
Nvidia signed a multi-year strategic agreement with Lumentum Holdings to accelerate innovation in advanced optical technologies, including R&D, to enable the design of next-generation artificial intelligence infrastructure and systems. The non-exclusive agreement includes billions of dollars in procurement commitments from Nvidia and future capacity access to advanced laser components.
In addition, Nvidia will invest $2 billion in Lumentum to support R&D, future capacity, and operations as Lumentum establishes its U.S. manufacturing capabilities at a new wafer fab.
For Coherent, the agreement also covers other optical networking products.
On March 8, the list of four companies to be swapped in the S&P 500 index was released, including Lumentum and Coherent, effective March 23. Following this news, Lumentum closed 14.73% higher and Coherent closed 7.04% higher on the next trading day.
Nvidia, AMD, and Intel invested in Ayar Labs
It’s also worth mentioning that in 2025, NVIDIA, AMD, and Intel, the three leading chip manufacturers, all invested in Ayar Labs, a startup specializing in optical I/O. This is no coincidence, as they all recognized that optical transmission is revolutionizing the way data is transmitted in the future, and those who don’t participate will be left behind.
Ayar’s technology uses optical signals instead of traditional electrical signals, aiming to accelerate data transmission by connecting AI computing chips and storage chips. As hyperscale cloud vendors and countries invest hundreds of billions of dollars in vying for dominance in AI infrastructure, the value of this technology is increasingly evident.
The innovation of Ayar Labs’ optical I/O solution lies in its monolithic packaging design. This design allows optical modules to be directly integrated with the chip, enabling large-scale direct connections between nodes and effectively overcoming limitations in I/O density, data rate scaling, and interconnect power efficiency.
Compared to traditional pluggable optics and electrical SerDes interconnects, Ayar Labs’ optical I/O solutions offer 5-10 times higher bandwidth, 4-8 times greater energy efficiency, and reduced latency to 1/10th, maximizing the computational efficiency and performance of AI infrastructure while improving cost, latency, and power consumption, significantly enhancing AI metrics.
To meet the surging demand, Ayar Labs is expanding its production and testing capabilities, broadening its global operations, and accelerating the deployment of its co-packaged optics (CPO) solutions.
Price performance of the companies in this post
| ticker | 4/14/2026 price | 1/1/2026 to 4/14/2026 | Past 1 year performance | Past 5 year performance | P/E | Market Cap ($B) |
| NVDA | 196.51 | 5.37% | 75.14% | 1117.53% | 40.11 | 4775.193 |
| AVGO | 380.78 | 10.02% | 112.79% | 693.29% | 74.27 | 1802.606 |
| CIEN | 467.19 | 99.76% | 686.78% | 727.18% | 296.69 | 66.06 |
| MRVL | 133.83 | 57.48% | 150.99% | 174.35% | 43.58 | 117 |
| COHR | 313.42 | 69.81% | 454.33% | 283.76% | 337.77 | 61.202 |
| LITE | 852.79 | 131.37% | 1518.50% | 816.58% | 260.76 | 60.889 |
| APH | 148.72 | 10.05% | 126.16% | 334.09% | 44.49 | 182.841 |
| AAOI | 146.39 | 319.94% | 1178.52% | 1757.74% | #N/A | 11.008 |
| ANET | 154.37 | 17.81% | 110.89% | 684.40% | 56.08 | 193.972 |
| CSCO | 82.61 | 7.24% | 44.42% | 59.97% | 29.76 | 326.301 |
| GLW | 172.82 | 97.37% | 313.64% | 276.51% | 94.31 | 148.455 |
| NOK | 10.35 | 59.97% | 101.36% | 150.61% | 77.01 | 50.003 |
| CRDO | 159.52 | 10.86% | 307.15% | #N/A | 88.53 | 29.433 |
I am the author of the original text, the essence of this story was originally featured on Money Magazine, Issue of May 2026
Related articles
- “What are silicon photonics? Why it’s important? and current progress“
- “Spectrum-X and Quantum-X switches, Nvidia’s new territory“
- “Portfolio of Nvidia investment empire “
- “The reasons behind Broadcom share price consistantly outperformance“
- “Significant changes in Broadcom’s business approach“
Disclaimer
- The content of this site is the author’s personal opinions and is for reference only. I am not responsible for the correctness, opinions, and immediacy of the content and information of the article. Readers must make their own judgments.
- I shall not be liable for any damages or other legal liabilities for the direct or indirect losses caused by the readers’ direct or indirect reliance on and reference to the information on this site, or all the responsibilities arising therefrom, as a result of any investment behavior.
