PuYin AnSheng Fund's Li Fan: Three Must-Know Terms in AI Investment—CPO, Optical Modules, Optical Communication

Ask AI · Why is CPO technology predicted to be a key investment focus point for 2027?

With an explosion in AI computing power, CPO, optical modules, optical communications—these specialized terms leave ordinary investors scratching their heads. What exactly are they? Why did they suddenly catch fire? And what do they have to do with AI investing?

Shanghai Pudong Development & Anshun Asset Management fund manager Li Fan provided a clear answer: optical communications are moving from “between servers” to “inside servers,” and the market space will expand multiple times, with 2027** possibly**** the key starting point for CPO ramp-up in volume**. This article presents his latest assessment of AI computing power, the optical communications industry chain, and the technological evolution of CPO in a one-question, one-answer format, helping investors clear the fog and grasp the real main line of AI computing-power investing.

Q: Many investors can’t tell apart these concepts—“CPO,” “optical module,” and “optical communications.” Can you help everyone clarify their relationships with the most accessible analogy?

A: “CPO (co-packaged optics)” is a next-generation packaging technology that integrates the optical engine with the switching-chip packaging, solving the power-consumption and signal issues at high data rates. You can understand it as “welding the translator and the chip together”—previously, the translator was separated from the chip by a distance, and signal transmission would suffer losses and consume more power. CPO sticks the two together, enabling faster speeds and less power consumption.

“An optical module” is the core product form of optical communications in data center applications today; it is responsible for converting optical and electrical signals. We can understand it as a “translator”—it translates electrical signals into optical signals for transmission over fiber, and then translates them back at the destination. In today’s AI data centers, thousands of optical modules are doing this “translation work.”

“Optical communications” is a broader industry direction, covering the complete chain from fiber optics to components to systems.

To put it another way: optical communications are like a “highway system,” optical modules are like “toll booths/vehicles,” and CPO is like “building the factory directly at the highway intersection”—reducing transportation distance and improving efficiency.

Q: Under the AI industry investment theme, why should investors care about these concepts? Which key segments are worth paying attention to?

A: As demand for AI computing power grows, traditional optical modules are upgrading to CPO. This upgrade process will bring a market space that expands multiple times. Whoever gains advantages in this upgrade may become the next-stage industry leader.

For example: an AI foundation model is like a super factory. GPU chips are the machines in the factory, while optical communications are the “conveyor belt” connecting these machines. The speed and efficiency of the conveyor belt determine the factory’s overall production capacity. As AI computing-power demand surges, the number of GPU chips grows exponentially, but the chips need high-speed interconnects to cooperate. If the “conveyor belt” isn’t fast enough, even more machines can’t deliver efficiency. Therefore, optical communications have become one of the most critical bottlenecks in AI computing power.

Q**: The market often discusses “CPO” and “optical modules” together—are these two different technical routes? In terms of investing, are they opposing or iterative relationships? How should ordinary investors view the investment cadence of these two?**

A: They have an “iterative and coexistence” relationship. In the current stage, conventional pluggable optical modules are still the main force for interconnecting across servers, while CPO is gradually penetrating from “within servers” interconnect into the server internals. We expect 2027 to be the key starting point for CPO penetration. From an investment cadence perspective: in 2026–2027, traditional optical modules will still benefit from high growth in AI computing-power demand (compound growth exceeding 3x); after 2027, the CPO-related industry chain will see volume ramp-up, bringing new incremental space. Each has its focus at different stages, and this is not a simple replacement relationship.

Q: We often hear the saying “fiber moves in and copper backs out,” but some also believe copper cabling still has cost advantages for short-distance interconnects. What’s your view on the “turf battle” between optical communications and** copper interconnect**** inside AI server racks?**

A: Our judgment is very clear: copper interconnects have already hit a physical ceiling. With increasing AI computing density and transmission speeds moving toward 800G/1.6T, signal attenuation and power-consumption bottlenecks for copper cables are becoming increasingly prominent. We expect optical communications to penetrate from “between servers” into “inside servers,” and the possible outcome is more than a fivefold market space expansion and about a tenfold increase in bandwidth. Of course, in real engineering, extremely short-distance interconnects within racks may still preserve copper-cabling solutions, but the penetration rate of optical communications will continue to rise.

Q**: **Besides CPO, the market also has new technical buzzwords like LPO (linear-drive pluggable optical modules) and OIO (optical input/output), which can leave investors dizzy. What problems do these technologies each address? How should investors distinguish between “theme hype” and “real industry trends” in terms of investing?

A: We need to filter for “real industry trends.” First is “returning to first principles”—assess whether the technology can solve real industrial bottlenecks. Second is “analyzing industry pain points”—what physical limits, power-consumption, and cost problems current technologies run into.

Third is “comprehensive evaluation of cost-effectiveness”—whether the new technology has sufficient advantages versus existing technology. Finally is “deep industry validation”—track industry conferences like OCP and GTC, and observe whether leading companies have laid out plans and whether there is a clear commercialization timeline. Using this standard: if a technology stays only at the conceptual level, with no follow-through from top manufacturers and no clear mass-production milestones, investors should be alert to the risk of “theme hype.”

Q: Currently, there is significant concern in the market about an AI bubble. You propose that “the AI industry has entered the dawn of monetization.” What is the basis for this judgment?** What signals does the pricing increase in**** cloud computing**** convey?**

A: The current pullback in U.S. software stocks and the Hang Seng Tech Index segment is not the destruction of industry value, but rather a disruption of traditional business models caused by the explosion of AI applications. Over the past 20 years, the cloud computing industry, which kept cutting prices, has started to raise prices. Computing power has become the core asset that can be monetized directly, highlighting the industry characteristic of “computing power equals revenue.” This indicates that the AI industry has moved from the investment phase into the monetization phase.

Risk warning:** Funds involve risk; invest with caution. Source of data: Wind, as of 2025.3.25****.**

The opinions and commentary provided in this material are for reference only and are time-sensitive. They do not constitute any advice or recommendation regarding any security mentioned. You bear all consequences arising from investing or acting based on the information related to this material. This material belongs to our company. Without written permission, no institution or individual may make any deletions or modifications to the content that run counter to the original intent. **The fund manager undertakes to manage fund assets with integrity and due diligence in accordance with prudent and rigorous principles, but does not guarantee that the fund will be profitable or guarantee any minimum returns. Investors need to understand that there may be scenarios where the principal invested in the fund is at risk. The fund’s past performance does not indicate its future performance. The performance of other funds managed by the fund manager does not constitute a guarantee of the performance of this fund. Since the operation time of Chinese funds is relatively short, it cannot reflect all stages of stock market development. Before investing in this fund, investors should carefully read legal documents such as the《fund contract》and《prospectus》. If you wish to purchase this fund, please pay attention to the relevant regulations on investors’ suitability management, conduct risk assessments in advance, and purchase fund products that match the appropriate risk level based on your own risk tolerance. The individual stocks, industries, sectors, and indices mentioned above are only for illustrative purposes and do not constitute investment advice. Past performance of the indices does not represent future returns.