It soared 4 times a year, and the optical module followed Nvidia to drink soup

Text|Read finance and economics.

Nvidia became the first company to eat AI dividends, and in the past year or so, Nvidia's stock price has soared by 37 times. The outbreak of Nvidia has also affected investment opportunities in the domestic AI industry chain.

Domestic optical module products are supporting products of NVIDIA GPUs, and a GPU requires 6-8 optical modules. Nvidia has also led to a rise in the volume and price of optical modules: the arms race for technology companies to buy H300 has led to a huge increase in demand. At the same time, NVIDIA's need for higher-end 800G optical modules has led to an increase in the unit price of optical modules.

Relying on the strong binding with NVIDIA, optical modules have become one of the best-performing sectors in the domestic AI industry chain, and the stock prices of Zhongji Innolight and Tianfu Communication have risen by 42 times, 22 times. The slogan "you have to believe in light" when photovoltaic exploded has now become "you still have to believe in light", but it has changed from photovoltaic to optical module. However, the optical modules with a good future are not without hidden worries.

This article holds the following views:

1. Optical modules are one of the few sectors in China that can deliver on their performance in the AI wave. Most domestic technology companies are either stuck by foreign countries, or there is a large gap between technology and overseas, and their participation in the AI wave is not high. However, optical modules have a global competitive advantage and provide stable data processing capabilities for overseas AI.

2. The optical module follows NVIDIA to drink soup. The optical module is strongly bound to NVIDIA's GPU, and the combination of the GPU and the optical module can improve the speed of data transmission and provide more efficient computing power for AI training. Domestic optical module companies such as Zhongji InnoLight are also the leading merchants of NVIDIA, and their sales have skyrocketed with NVIDIA.

3. The involution of the optical module is not stopped. The gap in optical module technology is not large, once the volume is increased, the competition will be fierce, the gross profit margin will also decline, and downstream customers will force the rapid iteration of optical modules, the combination of the two characteristics, the optical module enterprise intelligence continues to invest in research and development in advance, and the speed of product iteration has also been rolled up from 4-5 years to 2 years.

01 Optical modules lead the AI industry chain Looking back on the development of AIGC for more than a year, building the upstream of AI infrastructure has become the biggest investment opportunity at present, after all, NVIDIA's stock price has soared 37x the case is out there.

Turning our attention back to China, the upstream AI industry chain involved mainly includes key technical fields such as chips, computing power, semiconductors, CPOs, and optical modules. In these fields, optical modules have become the sector with the least divergence and the most explosive stock price.

Optical modules can be simply understood as components that transmit data to IT equipment, and are the bridge for data transmission between devices. At present, due to technical limitations, IT equipment can only recognize electrical signals but not optical signals, and the main role of optical modules is to assist equipment in photoelectric signal conversion and help equipment to transmit information. In AI scenarios, the speed of data transmission directly determines the performance of AI applications. As a high-speed data transmission solution, optical modules mainly provide fast data processing capabilities for AI.

But it is this sector that does not sound high enough to be technologically high that has become the biggest beneficiary of the domestic AI wave. From 2023 to the present, the share prices of Zhongji Innolight and Tianfu Communication, the head companies of optical modules, have risen by 42 times, 22 times. In contrast, other AI upstream companies are not impressive. The share price of Inspur Information, the head company of computing power, has been **10% in the past year, and the chip sector and semiconductor sector have fallen by more than 20% in the past year.

The reason why optical transceivers stand out is largely due to the competitive advantages they have gained in the global field. Most domestic technology companies are either stuck by foreign countries, or there is a large gap between technology and overseas, and their participation in the AI wave is not high. For example, it is difficult for domestic chip companies to eat the incremental market brought by AI because of the performance gap, so their performance and stock price are not greatly affected by AI.

However, it is precisely because the technical attributes of optical modules are slightly weaker, like many electronic products, although there is a technical learning cycle, but the overall cost and response speed are still fighting, so it has gradually become a domestic industry with global competitive advantages. Among the top 10 manufacturers in global optical module sales in 2022, 7 Chinese manufacturers will be on the list. Relying on the world's leading technological advantages, domestic optical module companies have entered the leading business network of overseas manufacturers, such as Zhongji InnoLight is the first of the four major cloud merchants in North America and NVIDIA.

The global advantage also makes optical modules one of the few sectors in China that can be linked to and resonate with the global AI industry chain. Reflected in the performance, Zhongji InnoLight expects a net profit of 2 billion yuan and 2.3 billion yuan in 2023, a year-on-year increase of 634%~87.91%。Tianfu Communication expects the net profit attributable to shareholders of listed companies in 2023 to be 67.7 billion - 75.8 billion, a year-on-year increase of 68% to 88%.

So how do optical modules benefit from the AI wave?

02 Following Nvidia to eat meat, the optical module can deliver on performance, largely because it hugs NVIDIA's thighs. The development of AI is inseparable from the computing power support provided by NVIDIA GPUs, and the optical module and GPU have a strong binding relationship, and a GPU needs 6-8 optical modules.

The combination of GPUs and optical modules can improve the speed of data transmission and provide more efficient computing power for AI training.

The large increase in Nvidia's GPU shipments has also brought the logic of both volume and price rise for optical modules. From the perspective of volume, the demand for GPUs is continuing to exceed market expectations, and TSMC, as the supplier of AI chip wafer manufacturing, can feel the changes in the industry the most, and it expects the compound growth rate of the company's computing power chips to exceed 50% in the next three years.

The arms race of technology companies has also added fuel to the fire of GPUs, and Meta's purchase of 350,000 H100s is higher than the previous market consensus expectation of Meta's less than 300,000 H100s. Microsoft and Google are also following up with plans to increase GPU purchases. The popularity of GPUs has also converted orders for optical modules, according to the ** "Information Equality" report, since Meta purchased more GPUs, several head optical module companies have inserted orders and scheduled production orders from the second quarter of 2024.

In addition to the surge in volume, NVIDIA has also driven the shipment of high-end optical module products. At present, the most mainstream optical modules on the market are 200G, 400G and 800G propagation rates. The higher the speed of the optical module, the faster the data transmission speed. AI has created a greater demand for bandwidth on the network side, requiring higher-rate optical modules, and NVIDIA's latest H100 chip has increasingly increased the application of 800G optical modules.

This has also led to a sharp increase in the sales of 800G optical modules, and some snowball investors said that the market demand for 800G optical modules in 2023 will be between 1.2 million and 1.4 million. By 2024, NVIDIA alone will be able to support the market of 10 million 800G optical modules. Zhongji InnoLight also revealed in a recent investor research event that it has received clear guidance and confirmation of the share of 800G demand from key customers next year, which is several times higher than the demand in 2023.

As a product with a higher price, 800G optical modules have also led to the improvement of the profitability of optical module companies. Zhongji InnoLight revealed at the ** meeting in the third quarter of 2023 that 800G products have ranked first in terms of revenue contribution from Q3. The increase in the proportion of 800G optical module revenue has increased the gross profit margin of Zhongji InnoLight in the first three quarters of 2023 by 383 percentage points.

However, optical modules that have achieved both volume and price increases are not without problems.

03 Although it is the core component of optical communication, you will know that modules and modules are almost the same: buy core parts, get some other accessories, assemble them through technology, and realize complete functions. This also constitutes a hard flaw in the optical module - the technical content is not high.

It is difficult to open the product gap with low-tech optical modules, even if the technology gap between Chinese enterprises is basically 6-10 months. For example, in the 800G optical module, Zhongji InnoLight is a little faster, but Accelink, Huagong Zhengyuan, and Xinyisheng have also entered the delivery stage.

This kind of industrial characteristics determine that first, the technical gap between products is not large, once the volume is increased, the competition will be fierce, and the gross profit margin will also decline; Second, in the competition, channels and stable customer relationships are more important.

The solution to these two points is iteration, through iteration to solve the problem of declining gross profit margin is well understood, product iteration can rely on new products to achieve premium, reduce the impact of fierce competition in old products. In optical transceiver products, iteration is also the most effective way to obtain stable customer relationships.

Specifically, optical modules are mainly used in data transmission scenarios, and one of the characteristics of the data field is that the amount of data is increasing and the structure is becoming more and more abundant. As a result, optical modules have higher and higher requirements for speed, and product iterations have been wave after wave.

For example, before 2015, most of the optical modules in data centers were based on 10G and 40G speeds; With the outbreak of the mobile Internet, 25G and 100G optical modules began to be deployed in overseas advanced data centers in 2016, and were popularized on a large scale three years later. The advanced data center has begun to deploy 100G and 200G optical modules. In 2022, 200G products will begin to be mass-produced; Then, in 2023, NVIDIA stimulated the demand for 800G optical modules. In the case of this rapid increase in speed requirements, whoever develops high-speed products first will be the first to acquire customers and enjoy phased dividends.

The rapid iteration of product characteristics also determines that there is no eternal winner in the field of optical modules, and only through continuous R&D investment in early layout, with a short-term lead in time, to seize the market and earn excess returns. This also forces the continuous involution of optical module companies.

The degree of involution is being upgraded, and the previous upgrade cycle of optical module products was 4-5 years, while the next generation is from 800G to 1The 6T upgrade cycle will be shortened to two years. More involuted industries not only force enterprises to increase investment, affect profits, but also increase the risk of enterprises being overtaken.

When the AI-driven optical module track explodes, orders outweigh **; Certainty trumps imagination. However, once the demand for optical modules slows down, the weaknesses of large fluctuations in corporate profits and short moats will gradually be exposed.