Economic Observer reporter Zheng Chenye

As one of the most talked about concepts in the A-share market since 2026, the stock price of "Yi Zhongtian" is currently undergoing a deep adjustment.

Yi Zhongtian "refers to Xin Yisheng (300502. SZ), Zhongji Xuchuang (300308. SZ/03308. HK), and Tianfu Communication (300394.SZ). The first two produce optical modules (devices that convert electrical and optical signals in AI data centers), while Tianfu Communication supplies optical devices to optical module manufacturers.

On November 6th, the three stocks of "Yi Zhongtian" experienced intraday declines of over 12%, with a cumulative decline of up to 65% from their June stock price peak. On November 3rd, all three stocks had experienced a decline of over 9%.

The reasons for this round of stock price adjustment are complex, including the excessive increase in the previous period, the technical solution dispute brought about by Nvidia's promotion of CPO (co packaged optics, which may replace some optical modules), and the expectation of a slowdown in the growth rate of capital expenditure by cloud vendors.

On November 2nd, American chip company Marvel announced its acquisition of American optical device manufacturer Lumentum for approximately $115 billion. Both of these companies are major suppliers to optical module manufacturers, which has raised concerns in the market about increased concentration of upstream supply, and subsequently, there has been a trend of "overseas customers cutting orders" in the market.

On the evening of November 3rd, Wang Jun, Vice President and Secretary of the Board of Directors of Zhongji Xuchuang, denied rumors of customers cutting orders on a large scale during an investor conference call, stating that "the vast majority of customers have not lowered their needs and orders". Wang Jun also stated that the acquisition of Maiweier Technology "has no impact on our entire supply structure and future competitive landscape".

However, according to the third quarter report of the optical module company, there has been a slowdown in performance - New Yisheng's revenue and net profit in the third quarter of 2026 decreased by 12.98% and 19.07% respectively compared to the previous quarter, and Tianfu Communication's net profit was also lower than market expectations.

There are objective reasons for the slowdown in growth rate. The floods in Thailand in late September slowed down the shipment of New Yisheng's Thailand factory, but the impact of material shortages seems to be greater. Wang Cheng, Vice General Manager and Secretary of the Board of Directors of Xinyisheng, stated at the company's performance exchange meeting that raw materials have a certain impact on the company's output and delivery in the third quarter of 2026.

How many materials can be obtained determines how many orders can be delivered. The shortage of materials is more challenging for optical module manufacturers than a drop in stock prices.

The core material of optical modules is chips - optical chips are responsible for photoelectric conversion, while electrical chips such as DSP (Digital Signal Processing Chip) are responsible for processing high-speed electrical signals. According to the information obtained by the reporter, both types of chips are currently under pressure, and the laser chip responsible for emitting light has a larger gap in the optical chip.

There are two main ways for optical module manufacturers to cope with upstream material shortages. One is to lock the materials upstream. For example, as of the end of September 2026, the prepayments of Zhongji Xuchuang and Xinyisheng increased by 1068.43% and 7695.01% respectively compared to the beginning of the year, mainly for prepayments of materials to upstream. Secondly, it is to adopt more silicon optical solutions and use fewer laser chips - in traditional solutions, laser chips are responsible for both emitting light and loading data onto optical signals, with one chip required for each transmission channel; The silicon optical solution separates the function of loading data and hands it over to the silicon optical chip. The laser chip is only responsible for emitting light, and one chip can be divided into multiple channels, thus reducing the usage.

The so-called silicon light refers to the use of semiconductor technology to manufacture optical devices on silicon-based materials, and silicon optical chips are also a type of optical chip. These types of chips need to be manufactured by wafer foundries, and the wafers used by foundries to process silicon optical chips are silicon optical wafers.

As a result, with more optical module manufacturers adopting silicon optical solutions, after DSP and lasers, silicon optical wafers have become a new bottleneck in the upstream of optical modules, and foundry capacity is scarce. Public information shows that the production capacity of the two major foundries, Israel's High Tower Semiconductor and the United States' Gexin, has been scheduled for after 2027.

Switching to Silicon Light Solution

Optical module manufacturers are turning to silicon optical solutions, starting from the insufficient use of laser chips such as EML (Electro Absorption Modulation Laser) in traditional solutions.

EML integrates luminescence and modulation (loading data onto optical signals) onto an indium phosphide chip. In July 2026, Lumentum President and CEO Helston publicly stated that Lumentum, along with its competitor Coherent, cannot meet the laser needs of customers such as Nvidia.

The silicon photonics scheme separates luminescence and modulation. Silicon materials themselves are difficult to emit light efficiently, and indium phosphide laser chips are still used as light sources. However, continuous wave (CW) lasers are used instead, which only provide stable light. After splitting, they can be used for multiple channels, and modulation is completed on the silicon optical chip. For example, a 1.6T optical module has 8 channels, which requires 8 EMLs in the EML scheme, while in the silicon optical scheme, 1 CW laser with 2 channels requires 4 EMLs.

And the amount of laser used can be further reduced. Wang Jun stated in a conference call on November 3rd that if the proportion of silicon optical solutions continues to increase and the modulation capability of silicon optical chips is further improved, the silicon optical solution "can use 1 to 4 instead of 1 to 2, so there is still room for overall cost reduction". That is to say, in this case, the amount of CW laser used in the 1.6T optical module can be further reduced by half.

In addition to reducing material usage, the total cost of the silicon photonics solution is also lower. According to the prospectus of Xihe Technology, a silicon optical chip design company, based on the calculations of optical module manufacturers, the total cost of silicon optical solutions can be reduced by nearly 20% at the same speed and distance.

In fact, Wang Xiaoli, Vice President and Chief Financial Officer of Zhongji Xuchuang, also listed the increase in silicon photonics solution penetration rate as one of the reasons for the company's gross profit margin increase during the third quarter report interpretation conference call. In the third quarter of 2026, the company's gross profit margin was 46.6%, an increase of 0.2 percentage points compared to the previous quarter.

In addition, the silicon light scheme is more energy-efficient. Xihe Technology stated in its prospectus that the power consumption and reliability of such solutions are superior to traditional solutions. This has more practical significance in the United States. A senior investor who has long been concerned about the field of optical communication told reporters that the market generally sees the Democratic Party's victory in the midterm elections as a disadvantage for optical modules, but many people do not realize that the silicon optical solution is actually energy-saving. ICTIME Chief Analyst Lin Meibing also told reporters that in situations of power shortage, customers are more sensitive to the power consumption of optical modules when selecting, which is beneficial for silicon optical solutions.

The proportion of silicon photonics solutions is also constantly increasing. As early as April 2026, Zhongji Xuchuang announced on its investor interaction platform that its 800G and 1.6T products accounted for over half of the silicon photonics solutions. A communication analyst from a large securities firm in southern China told reporters that the proportion of silicon optical solutions in 1.6T optical modules in the entire industry will be about 60% to 70% in 2026, and may reach around 80% in 2027.

As the proportion of silicon photonics solutions increases, the pressure falls on the wafer foundry process.

Silicon optical wafers are in short supply

There are not many foundries capable of producing silicon optical wafers, and it will take time for the new production capacity to be in place.

Wafer is the basic material for manufacturing chips, and foundries make circuits on the wafer and cut it to obtain bare chips. The circuits for processing electrical signals are made on ordinary chips, while devices for processing optical signals such as modulators and detectors are made on silicon optical wafers, with different processes from ordinary chips.

Lin Meibing told reporters that the manufacturing process of silicon optical chips is complex, with a high yield threshold, and there are not many foundries that can mass produce them. Currently, they are mainly concentrated in a few manufacturers such as Israeli wafer foundry High Tower Semiconductor and American wafer foundry Gexin.

During the performance meeting on November 2nd, Elwanger, CEO of Gaota Semiconductor, stated that the company's silicon optical chip foundry capacity contract with customers has been extended until 2029. Ge Xin CEO Brin also stated at the earnings conference on November 4th that the company's silicon optical chip foundry capacity has actually been sold out, with the 2027 capacity already fully booked and the 2028 capacity being gradually booked.

Due to limited production capacity, downstream customers can only lock in early. Gaota Semiconductor announced in May 2026 that it had signed contracts with its largest clients, corresponding to a silicon optical wafer foundry revenue of $1.3 billion in 2027, and received $290 million in customer prepayments for reserved production capacity.

The OEM factory is also expanding production. Elwanger stated that by the third quarter of 2027, the wafer casting capacity of high tower semiconductor silicon photonics wafers (putting wafers into production lines) will increase threefold compared to the level in the third quarter of 2026, and most of these capacities already have customers. Brin also stated that Gexin plans to expand its silicon optical chip production capacity to about three times by 2027.

A later batch of new capacity will be built after 2028, and the output target of Gaota Semiconductor's new factory in Japan will be in 2029. The new capacity of Taiwan, China's wafer foundry will not be installed until the end of 2028.

After the new factory is built, silicon optical chips cannot be transferred immediately. The market manager of an optical module manufacturer in southern China told Economic Observer reporters that if they switch factories for silicon optical chips, they need to go through another process validation, which usually takes about a year. During the interview, the reporter also learned that some silicon optical products had skipped this process in order to catch up with the schedule and directly transferred to the new factory for production, resulting in poor yield rates for some batches and affecting downstream delivery.

It is difficult to achieve the new production capacity in the short term. Lin Meibing believes that the amount of production capacity that can be obtained now depends on whether there is an early signing of contracts.

Currently, many mainstream pluggable optical modules use 8-inch silicon wafers. According to the planning of contract manufacturers, the total 8-inch production capacity can basically cover this demand. However, most of the production capacity for 2027 has already been pre ordered through agreements, which means that the factories that lack production capacity are those that have not signed contracts in advance.

12 inch silicon wafers are more compact and are more commonly used for next-generation products such as NPO (Near Package Optics). Lin Meibing stated that according to the manufacturer's demand plan for 2027, the existing production capacity is far from sufficient, and the new round of expansion will focus on 12 inches. However, the construction period is long, and the supply will still be tight before the end of 2027. But in Lin Meibing's view, this round of supply shortage may not be sustainable in the long run, because there are currently more than ten manufacturers investing in silicon optical chip manufacturing, and after new entrants increase their volume, the supply will gradually increase.

In fact, newly entered contract manufacturers have begun to compete for 12 inch orders. Industry insiders have told reporters that there have been differences in quotes, with some manufacturers quoting around $10000 per piece and others quoting around $8000.

But before the new production capacity is truly released, whether it is possible to sign in for production capacity depends on the position of the relevant manufacturers in the OEM factory.

The gap between self-developed and outsourced

The position of the relevant manufacturers in the OEM factory is related to where the silicon optical chips come from.

There are two sources for silicon optical chips: one is that optical module manufacturers design them themselves and then commission contract factories to manufacture them; The second is that optical module manufacturers purchase from third-party chip companies. The aforementioned communication analyst believes that in the 800G era, optical module manufacturers were focused on delivery and yield, but in the 1.6T era, silicon optical solutions became mainstream, and chip design capabilities and OEM production capacity began to affect market share.

Top optical module manufacturers have already reached the stage of chip design. For example, in response to inquiries from the Shanghai Stock Exchange, Xihe Technology stated that optical module manufacturers such as Zhongji Xuchuang and Xinyisheng have "extended their design capabilities for silicon optical integrated chips", which are used to complement their own optical module products and are not sold to the public. Public information also shows that Paisi Technology, a subsidiary of Zhongji Xuchuang, and Alpine, acquired by Xinyisheng, are both developing their own silicon optical chips.

Optical module manufacturers with the ability to independently develop silicon optical chips will directly place orders with foundries. During the interview process, several industry insiders told reporters that top optical module manufacturers have a high share of orders for silicon optical chips from foundries and will also prioritize production capacity allocation over third-party chip companies. If optical module manufacturers purchase from third-party chip companies, the cost will increase. As a reference, the gross profit margin of third-party chip company Xihe Technology in the first half of 2026 is 59.24%.

Of course, as the proportion of top optical module manufacturers developing their own silicon optical chips increases, the pressure on third-party chip companies is also increasing. In the first half of 2026, the sales proportion of Xihe Technology's largest customer was 55.55%. The company reminds that Longtou Optical Module manufacturers already have the ability to independently develop silicon optical integrated chips. If downstream manufacturers carry out large-scale self research and change their outsourcing strategies, their customer demand and business performance may be adversely affected.

Xihe Technology is a large-scale third-party chip company that adopts a fabless model (only designing chips and outsourcing manufacturing to foundries). According to Frost&Sullivan's data, its global market share of silicon photonics integrated chips is expected to be around 13% by 2025. Public information shows that Xihe Technology's orders are growing rapidly: its revenue in the first half of 2026 was 569 million yuan, exceeding the full year revenue of 461 million yuan in 2025; As of the end of August 2026, the outstanding orders amounted to 4.551 billion yuan.

With the rapid growth of orders on hand, the demand for wafers from third-party silicon-based chip companies is also growing rapidly in 2027. During the interview, the reporter learned that the demand for wafers from third-party silicon optical chip companies has significantly exceeded the production capacity they have already signed contracts with contract manufacturers to guarantee.

The marketing manager of the aforementioned optical module manufacturer also stated that their company's silicon optical chips mainly rely on external procurement. When the OEM production capacity is tight, the chip company also has to queue up, and the time for the company to obtain the chips ranks behind the top manufacturers, resulting in longer delivery times.

In fact, domestic silicon wafer foundry platforms have already started, although the cumulative scale is not yet large. According to Frost&Sullivan's data, as of the end of April 2026, wafer foundry Yuexin Semiconductor (301660. SZ) is one of the few companies in China with the ability to mass produce 12 inch silicon optical wafers. Its OEM products cover 400G, 800G, and 1.6T silicon optical modules, with a cumulative wafer output of over 7000 pieces.

Regarding this, Lin Meibing stated that there is already a foundation for the design of silicon optical chips in China, and there are not many platforms capable of mass producing 12 inch chips in the manufacturing process. Some third-party chip companies have started to transfer their products to domestic 12 inch platforms for outsourcing, but domestic wafer foundries need time to undertake high-end products, process validation, and yield improvement.

(Author Zheng Chenye)