Economic Observer reporter Zheng Chenye

On Monday, July 13th, the A-share market fell across the board, with the Shanghai and Shenzhen 300 Index falling 1.79% and the Sci Tech Innovation Composite Index falling 4.36%. In such a market atmosphere, as the top weighted stock in the Shanghai and Shenzhen 300, Zhongji Xuchuang (300308. SZ)'s stock price "Xiaohong" closed up 1.28%, with a total market value of 1235.7 billion yuan.

But on the previous trading day, July 10th (Friday), Zhongji Xuchuang's volume plummeted by 8.45%, with a turnover of 49.028 billion yuan. This is related to two "short essays" circulating in the market: firstly, the company's second quarter performance fell significantly short of expectations; Secondly, the company intentionally suppressed the stock price in order to allow participants in the ongoing H-share issuance to obtain lower prices.

July 12th is a Sunday. At 8 pm that evening, Zhongji Xushao held a conference call with over 100 domestic and foreign institutions. Wang Jun, Vice President and Secretary of the Board of Directors, was entrusted by the management to preside over the meeting. Wang Jun denied the above market rumors one by one at the meeting, stating that the company has not yet released a performance forecast and semi annual report, but is "very confident" in the operating situation of the first half of 2026, especially the second quarter performance. The company also does not intentionally suppress stock prices, delay revenue recognition, or deliberately adjust performance.

This rumor doesn't understand our moral bottom line, "Wang Jun said. Late that night, the investor relations activity record of this meeting was posted online for disclosure.

Has the demand changed?

As the most dazzling star company in this round of A-share bull market, the stock price of this leading manufacturer of optical modules skyrocketed from 65.30 yuan to 1416.88 yuan from April 9, 2025 to June 22, 2026 based on the exercise price, with a staggering increase of 20.7 times!

But for Zhongji Xuchuang, there are many controversies surrounding the company's popularity. Since the beginning of this year, there have been frequent "short essays" circulating in the market. According to the Economic Observer, in the past month and a half, Zhongji Xuchuang has responded to market rumors at least four times over the weekend - on May 30th, it declared that the chairman's speech circulating online was fabricated; On June 14th, there was a currency exchange explosion on Interactive Easy Denial; On July 5th, denied that upstream materials were blocked and responded to negative speculation about Corning Glass Bridge technology; The phone call on July 12th was the most formal of the four responses.

In fact, the experience of Zhongji Xuchuang can be seen as a microcosm of the heated debate in the entire optical module industry: this track has been rising for three consecutive years, and the market value of leading companies has surged from tens of billions to trillions, amplifying the controversy. Even a slight fluctuation can cause dramatic fluctuations in the pricing of the secondary market.

The phone conference on July 12th lasted for about an hour, and the questions from participating institutions focused on three directions: demand, price, and materials. These are also the most intense and highly discussed issues in the market over the past month.

In terms of demand, there have been rumors in the market that the demand for 1.6T optical modules has been "cut". During the conference call, Wang Jun recounted the process of rumor fermentation: In March of this year, Zhongji Xuchuang had given expectations for the industry's demand for 1.6T optical modules; Recently, the company has been confirming the latest procurement plans with customers one by one, and during the process, it was found that some customers have lowered their purchase quantities of 1.6T optical modules due to their own reasons.

But this is incomplete data, and if we use it as a reference, we will feel that the demand of the entire industry has weakened, "said Wang Jun. Some customers have decreased, while others have increased. For example, a group of newly added cloud vendors and AI big model companies have increased their demand for 1.6T optical modules, some of which are first-time purchases of this product. New customers have already passed certification and obtained the 2027 market share. In this way, the industry's total demand remains within the expected range given in March, with a change in customer structure.

The change in customer structure is an industry signal that this phone call will reveal.

The traditional major customers of optical modules are mainly North American large-scale cloud vendors, whose procurement cycles and capital expenditure plans directly determine the industry's prosperity. According to public information, Zhongji Xuchuang's overseas revenue will account for over 90% in 2025, with the top five clients all being American companies, accounting for approximately 76% of total sales.

During the conference call, Wang Jun mentioned that both cloud vendors and AI big model companies have self-developed chips and purchased GPUs, with different chip sources. However, in the end, they all need to use optical modules to complete the connection. The downstream customer pool of optical modules is expanding, and demand sources are becoming more dispersed.

In addition, from the perspective of product generation, the demand growth for 800G products is more significant than that of 1.6T products.

Wang Jun said at the meeting that the company's current orders have covered the entire year of 2026, and some customers have provided quarterly shipping guidelines for 2027. The procurement of domestic data centers is concentrated on 800G and 400G products, and the demand for both generations of products is increasing. Some domestic customers are currently bidding, and the demand will be further clarified in the second half of the year. He also mentioned that some customers provide shipping guidelines that are even precise to the month, which is extremely rare in the manufacturing industry.

An industry analyst from a large securities firm in southern China also told Economic Observer reporters that he recently investigated domestic optical module, optical chip, and fiber optic enterprises, and did not receive any information about order cancellations or demand guidance for repairs.

In terms of upstream materials, the tension of upstream materials such as optical chips and electrical chips is also easing, and Wang Jun expects it to gradually improve from the second half of this year to the first half of next year. Optical modules have an industry characteristic different from semiconductors - production equipment lead times are measured in months, expansion cycles are much shorter than wafer fabs, and once material bottlenecks are opened, the release speed of production capacity will be very fast. According to relevant announcements, as of the end of March, Zhongji Xuchuang's advance payment was 1.488 billion yuan, compared to 134 million yuan at the end of the previous year. This change in numbers reflects the company's early locking efforts in upstream materials.

In addition, price negotiations are already underway. Wang Jun said at the meeting that the company is negotiating prices with customers for 2027, and the rumored price reduction is too exaggerated. If you have studied optical modules for more than three years, you wouldn't consider it a fiercely competitive industry, "he said.

However, Deng Hefang, an analyst at Zheshang Securities, also pointed out in a research report released on June 12th that the continuous expansion of production capacity and the influx of new participants in the optical module industry can easily lead to homogeneous competition and price decline. That is to say, the current tight supply is a fact, and there is also a possibility of price pressure after the release of production capacity. The two events correspond to different stages of the industry, and both long and short sides look at each other.

Substitution or division of labor?

Beyond supply and demand and price, there is a bigger concern in the market surrounding the optical module industry: Will the product form of pluggable optical modules be replaced by CPO within a few years?

CPO (Co encapsulated Optics) is a method of packaging the optical engine responsible for photoelectric conversion and the switching chip on the same substrate, reducing the electrical signal transmission path between the two from tens of centimeters to millimeter level, eliminating the need for independent optical modules inserted into the equipment panel. It is widely regarded in the industry as the next generation solution for optical interconnection in data centers.

On June 9th, SemiAnalysis, an independent semiconductor research firm in the United States, released a report stating that the large-scale production of CPO will be postponed until 2028 or even 2029.

The reason for the bearish report is focused on yield: a switching chip needs to package 32 optical engines, and the assembly yield of each optical engine is calculated at 95%. After 32 cycles, the overall yield of the machine is only about 19.4%, which is equivalent to only one qualified unit for every five units produced on the production line. Another issue is that the light engine is soldered to the substrate and cannot be reworked. Any defective substrate can only be scrapped.

In addition, CPO also faces a fundamental change in its operation and maintenance mode: in traditional data centers, a faulty optical module is replaced on-site, but CPO binds the optical engine and switch chip together, and components often need to be replaced as a whole when they fail.

Cloud vendors also have concerns about platform lock-in - the CPO solution requires binding to the packaging system of specific chip vendors, which will squeeze the supplier selection and bargaining space that cloud vendors have under the pluggable solution.

On the day of the release of SemiAnalysis's report, the US optical communication sector experienced a sharp decline, which was transmitted to the A-share market the next day.

A researcher from a large private equity firm in Shenzhen told the Economic Observer that the delay in large-scale production of CPO has precisely extended the demand window for pluggable optical modules. The demand has not disappeared, but has been reallocated on the timeline. Gilad Shainer, Senior Vice President of NVIDIA's Network Business, also publicly responded that CPO will coexist with pluggable optical modules in the long term.

The above-mentioned researchers told reporters that the CPO solution customized by Nvidia and TSMC has made rapid progress, with an expected shipment of 30000 to 50000 units in 2026 and an increase to 200000 to 300000 units in 2027. He also stated that it is a consensus in the current industry that multiple technological routes should be divided according to different scenarios and run in parallel for a long time. Pluggable optical modules, NPOs, and CPOs each correspond to different connection scenarios, and there is no such thing as a "one route fits all" situation.

Among these routes, NPO (Near Package Optics) is currently the fastest advancing one. NPO arranges the optical engine close to the switch chip on the same circuit board, shortening the transmission path of electrical signals to centimeter level, significantly improving power consumption and port density compared to pluggable solutions, while retaining independent and replaceable units. The data center basically does not need to change the existing operation and maintenance mode.

The above-mentioned researchers told reporters that NPO is mainly used in the Scale up layer (interconnection between GPUs inside the cabinet), which is different from the traditional optical module application scenarios of the Scale out layer (long-distance interconnection between cabinets and switches), and the two do not conflict. On Nvidia's next-generation GPU platforms GB300 (Nvidia's new AI server platform released this year) and Rubin (Nvidia's next generation GPU architecture expected to be launched by the end of 2026), the ratio of NPO to GPU is 1:1. The NPO is installed next to the GPU through a high-speed interface and is directly driven by the GPU's built-in high-speed serial transceiver to drive the optical engine.

He expects the industry wide demand for NPOs to be around 10 million by 2027. By 2028, with the increasing production of Amazon's self-developed AI chip Trainium 4, the demand for NPOs will further increase as one Trainium 4 corresponds to two NPOs.

Wang Jun also stated in the phone conference on July 12th that the progress of NPO has significantly accelerated since March this year. Some customers have opened up technical requirements for the company to customize NPO products, and some customers have also provided their needs for 2027 to 2028.

In fact, the layout of the domestic industrial chain around NPOs is accelerating overall.

In July of this year, Huawei, together with more than 20 industry chain enterprises such as China Mobile Research Institute, JD Cloud, Baidu, and China Electronics Standardization Research Institute, launched the first domestic NPO optical interconnect multi-source protocol project (MSA), which is an interoperability standard jointly developed by multiple manufacturers. This project aims to develop unified standards around mechanical specifications, electrical specifications, environmental reliability, and control operations, with a focus on unifying electrical connector interfaces. The first version of the technical specifications is planned to be released in the third quarter of 2026, and sample adaptation and full scenario testing will be completed. In the first half of 2027, it will promote the large-scale commercial use of NPOs.

In a research report released on July 10th, analysts such as Zhang Lin from Xingye Securities stated that NPO is expected to become a transitional solution for the current evolution of AI computing power clusters.

For Zhongji Xuchuang, the shipment volume of 800G by 2025 is at the global leading level, 1.6T is already in mass production and shipping, and NPO products have also entered the stage of customer customization. Wang Jun mentioned at the end of the phone conference that some customers are already communicating with the company about the demand for products with higher bandwidth than 3.2T.

This indicates that the trend of the entire industry is still accelerating, and the development of the entire industry is still healthy, "he said.

(Author Zheng Chenye)