The Great Test of Optical Communications under AI’s Demand for Computing Power: From Rushing to Buy to Rushing for Technology
Summary of Key Points
This news article reveals how the explosion in AI computing power is reshaping the optical communications industry. Essentially, because AI servers require the transmission of massive amounts of data, traditional electrical wiring is no longer sufficient, and “light” must be used as the medium of transmission. This has led to a surge in demand for optical modules—the key components responsible for converting electrical signals to optical signals. Not only are 800G products in short supply, but 1.6G products are also seeing increased production, with orders already booked through 2027.
However, behind this excitement lies a fierce technological competition: the industry is shifting from a focus on “production capacity” to a focus on “the speed of technological innovation.” Upstream suppliers of optical chips, electrical chips, and testing equipment are under significant pressure, especially since high-speed electrical chips are still dominated by overseas giants. For domestic companies, the challenge is no longer simply about having orders but about whether they can keep up with the pace of the next generation of technologies (such as 1.6T, 3.2T, NPO), or they will be left behind.
---
Detailed Analysis
1. Why has “light” become a critical resource for AI? – Electrical connections are no longer enough
In simple terms:
Imagine an AI data center as a super-busy logistics center. Previously, data was transported using ordinary trucks (electrical signals/copper wires). But with the exponential growth in the amount of data that AI needs to process, these trucks (electrical connections) are becoming too slow, consume too much energy, and have limited transmission distances.
So, people have started using “high-speed trains” (optical signals) to transport the data. Optical modules act as the “handlers” that load the data onto the high-speed trains and then unload it.
- Current situation: The most in-demand products are 800G optical modules.
- Trend: 1.6G optical modules are already in the trial phase and are being sought after by leading companies.
- Challenge: As AI clusters grow larger and data exchange becomes more frequent, optical interconnection is the only solution that can keep up with the expansion of computing power. This is why there is a severe shortage of these modules.
2. 1.6G is here, but don’t cheer too soon – The battle of different technical approaches
In simple terms:
Many think that replacing optical modules with faster ones just means using a new box, but that’s not the case. The current technological race is like an arms race, where companies are competing not only in speed but also in energy efficiency and compactness.
The article mentions several new terms: LPO (Linear Direct Drive), NPO (Near-Package Optics), and CPO (Co-Package Optics). Don’t let the letters scare you; they all aim to solve the same problem: how to integrate optical components more efficiently into chips.
- LPO: Simplifies the system by removing complex intermediate components (DSP chips) to save energy and reduce latency, suitable for applications that are sensitive to power consumption.
- NPO: Moves optical components closer to the computing or switching chips to minimize signal transmission distances.
- CPO: Goes a step further by integrating optical and switching components into a single package, achieving the highest density and lowest power consumption, but it’s also the most technically challenging.
Key point: Standardized, pluggable optical modules are still the mainstream because companies are not yet ready to adopt NPO/CPO on a large scale. However, in AI clusters, NPO and CPO will become more common for optimal performance.
3. Pressure is spreading upstream – From a shortage of modules to a shortage of chips
In simple terms:
Previously, the focus was on optical module manufacturers (such as Zhongji Xuchuang, NeoPhotonics, etc.). Now, the pressure has shifted to suppliers of the components needed for these modules.
- Optical chips: These are the “heart” of optical modules. To support 1.6G, higher-speed and more stable chips (e.g., 200G EML, thin-film lithium niobate TFLN) are required. These chips are in short supply and are still in the research or small-scale production stages.
- Electrical chips: These are the biggest bottleneck. In addition to optical chips, high-speed electrical chips (DSPs) are needed to process data. Currently, these are dominated by American giants like Broadcom and Marvell. Although domestic companies are good at manufacturing optical modules, they rely on foreign suppliers for electrical chips.
- Other components: Fibers, connectors, and polarization-maintaining fibers are also in short supply.
Conclusion: The pressure in the supply chain is no longer just about adding more production lines; it requires breaking through the technological barriers of core chips. Companies that control these chips will have more influence in the industry.
4. Testing equipment: The overlooked players and opportunities for domestic substitution
In simple terms:
Once optical modules are made, how do you know if they work properly? Testing equipment (such as oscilloscopes and error analyzers) is essential to verify their performance.
- Why important: As the speed of optical modules increases (from 800G to 1.6G), the signals become more complex and weaker, making testing much more difficult.
- Market changes: High-end testing equipment was previously dominated by foreign companies (e.g., Tektronix, Agilent). With the rise of 1.6G, the need for more versatile testing equipment has increased, giving domestic companies (e.g., Lianxun Instruments, Wanliyan) an opportunity to catch up.
- Market trends: The Chinese optical communication testing equipment market was worth about 3.3 billion yuan in 2024 and is expected to double to 6.59 billion yuan by 2029, with foreign companies holding 84% of the market and domestic companies 16%. This indicates significant potential for substitution.
- Cyclical nature: The demand for testing equipment fluctuates; companies need to be strategic and avoid overinvesting during periods of high production.
5. The ultimate challenge: From manufacturing capability to the right to define technology
In simple terms:
In the past, Chinese optical communication companies had advantages in manufacturing, cost control, and fast response times. These helped them gain a large share of the global market during the 800G era.
But with the 1.6G and future generations, the game has changed:
- New competition criteria: It’s no longer about production volume or cost; it’s about having more advanced technologies, a more autonomous supply chain, and the ability to define future standards.
- Domestic companies’ weaknesses: While they are strong in module manufacturing, they still lag behind in upstream chips (especially high-speed electrical chips) and key materials (e.g., thin-film lithium niobate).
- Future prospects: The next round of competition will focus on companies that can keep up with 3.2T and NPO technologies. Those that only focus on assembly and manufacturing may become low-end contract manufacturers, with profits squeezed by both upstream chip suppliers and downstream customers.
Implications for everyone:
If you follow investment or industry trends, don’t just focus on leading optical module companies; also pay attention to hidden champions in upstream fields such as optical chips, electrical chips, and testing equipment. Be cautious of technological transitions, as choosing the wrong path (e.g., focusing only on pluggable modules while ignoring NPO/CPO) could lead to falling behind in the next generation of competition.