Economic Observer reporter Zheng Chenye

Since the beginning of July, the A-share market has experienced a significant decline, especially in the AI oriented technology sector, where the decline has been particularly intense.

In this context, the semi annual performance forecasts of four optical communication companies have been intensively implemented: on July 21st, optical chip company Yuanjie Technology (688498. SH) announced that it would achieve a net profit attributable to shareholders of 600 million yuan to 650 million yuan in the first half of the year, an increase of 1196.91% year-on-year to 1304.98%; On July 19th, Xinyisheng (300502. SZ) announced that it would achieve a net profit attributable to its parent company of 7 billion to 8 billion yuan in the first half of the year, an increase of 77.56% to 102.93% year-on-year; On July 18th, Tianfu Communication (300394.SZ) announced that it would achieve a net profit attributable to its parent company of 1.124 billion yuan to 1.304 billion yuan in the first half of the year, a year-on-year increase of 25% to 45%; On July 14th, Dongshan Precision (002384. SZ) announced that it would achieve a net profit attributable to its parent company of RMB 2.9 billion to RMB 3 billion in the first half of the year, a year-on-year increase of 282.58% to 295.78%.

The profit levels of all four companies have reached historic highs.

On July 21st, the A-share market experienced a strong rebound, with the CSI Communications Index surging 6.5%, the CSI Computing Power Index also surging 8.7%, and the Sci Tech Innovation 50 Index surging 10.73%.

At 4 pm that afternoon, Yuan Yonggang, Chairman of Dongshan Precision, held an hour long online communication meeting with multiple market investment institutions, responding to market questions about AI computing power demand, production capacity planning, and material supply one by one. In the previous week, Xinyisheng and Zhongji Xuchuang (300308. SZ) also held relevant exchange meetings.

The reporter noticed that there are significant differences between the industry supply and demand and business conditions introduced by the management of the three companies at the above-mentioned exchange meeting and the widely circulated "short essays" on social media platforms.

I have not received any notification of "cutting orders"

An optical module is a standardized hardware product that encapsulates multiple chips and optical components together, installed on switches and servers in data centers, and uses optical signals instead of electrical signals to complete high-speed data transmission.

Optical modules play an important role in the AI computing power industry chain. In AI training and inference clusters, thousands of GPU chips require high-speed interconnection to perform collaborative computing, and optical modules are the core components that complete this interconnection.

During the interview, the reporter learned that the mainstream product deployed in data centers is the 800G optical module (which can transmit 800 gigabits of data per second), and the next generation product 1.6T (which can transmit 1.6 trillion bits per second) is entering a phase of mass production. The value of a single 1.6T optical module is twice that of 800G.

Yuan Yonggang briefly introduced the quarterly shipment volume of Dongshan Precision's optical module business at the exchange meeting on July 21st - only counting AI related optical modules of 800G and above, Dongshan Precision shipped 300000 units in the first quarter, nearly 1 million units in the second quarter, less than 2 million units expected in the third quarter, and the target for the fourth quarter is 4 to 5 million units; Among them, the stocking target for 1.6T optical modules in the fourth quarter exceeded 1 million units, and the target for the first quarter of 2027 is over 6 million units.

Yuan Yonggang also stated that the total production capacity of Dongshan Precision's optical modules is planned to be 35 million units by 2027, and will double on this basis by 2028. At present, the monthly production capacity of the Taiwan base is 1 million units, and the first phase of the Thailand base has a monthly production capacity of 2 million units. Subsequent expansion will prioritize the layout in Thailand. Recently, several North American core cloud service providers (CSPs) are intensively verifying the qualifications of the Thai factory.

According to public information, Dongshan Precision is a manufacturing company that started with consumer electronics PCBs (printed circuit boards) and only entered the optical module market after completing the acquisition of Solsks Optoelectronics in 2025. In the first quarter of 2026, the revenue from the optical module sector accounted for 16% of the company's overall revenue, with a profit contribution of over 50%.

Yuan Yonggang said that the company has reached cooperation agreements with most overseas core cloud vendors, and when submitting production capacity plans to customers, their feedback is "very urgent". In addition, the number of overseas cloud providers is also expanding, and in addition to traditional cloud providers, the order volume of emerging customers such as xAI is already very large.

At the exchange meeting, some investors even directly asked if the company's profit in 2027 could reach 40 billion yuan, while Dongshan Precision's net profit attributable to the parent company for the whole year of 2025 was only 1.386 billion yuan. Yuan Yonggang replied, "Anything is possible, as long as others can do it, we won't be bad either

He also responded to the two most discussed risks in the market during the exchange meeting.

Firstly, regarding the rumor of a top North American cloud vendor cutting orders, Yuan Yonggang stated that the vendor's demand has "hardly changed" and the investment payback period for North American computing centers has been shortened to 6 months. At this rate of return, there will be no situation of cutting orders.

Secondly, there are concerns about the delay of the Broadcom Tomahawk 6 chip (the core processing chip of the next generation data center Ethernet switch, and the 1.6T optical module needs to be plugged into the switch equipped with this chip to be used), which may lead to the delay of the 1.6T switch. Regarding this, Yuan Yonggang stated that there were indeed signs of delay observed in the first half of this year, but recent tracking from two weeks to a month shows that the landing of 1.6T is accelerating, and demand in 2027 is expected to exceed previous market expectations.

In his opinion, 1.6T optical modules will be the mainstream products for a long time in the future, and it is expected that most of the market demand will still be 1.6T after 2028. The DSP chip (digital signal processing chip, responsible for signal modulation and demodulation in optical modules) required for 3.2T optical modules has not yet been launched, and there is still a long time before large-scale commercial use.

The reporter noticed that during the New Yisheng investor exchange conference held on the evening of July 19th, the company's Vice General Manager and Secretary of the Board, Wang Cheng, also conveyed similar information.

Wang Cheng stated that based on the order situation of New Yisheng, the pace is growing quarter by quarter, and orders are still growing rapidly in the second half of the year. Moreover, orders for 1.6T next year will also be "very strong", and no orders have been lowered due to reasons such as switch chips. He also mentioned that the product structure of New Yisheng is undergoing changes this year, with 800G being the absolute mainstay and 1.6T increasing in volume from the second quarter. At the same time, the shipment proportion of silicon photonics solution (a technology route that integrates photonic devices on silicon-based chips using semiconductor processes, with lower cost and power consumption than traditional discrete solutions) is also increasing quarter by quarter.

Regarding the industry competition landscape, Wang Cheng believes that in the current industry landscape, it is quite difficult for new entrants to break through core customers and achieve product volume expansion. Both supply chain and mass production capabilities pose challenges for new entrants.

Wang Cheng also stated that both the Chengdu and Thailand bases of New Yisheng are accelerating production expansion in the first half of this year, with rapid personnel growth, which is "perfectly matched" with the increase in production capacity and order demand.

It is worth mentioning that as early as July 12th, Wang Jun, Vice President of Zhongji Xuchuang, directly denied market rumors such as "second quarter performance falling short of expectations" and "intentionally suppressing stock prices" during a conference call.

A securities analyst who has long been interested in the field of optical communication told Economic Observer reporters that in his opinion, the value of optical communication in the total investment of AI computing power is still increasing, evolving from the current level of about 5% to 10% or even higher.

Upstream optical chips are still in short supply

Although there is no problem on the demand side, the profit distribution logic within the optical module industry chain is more complex. The four companies that have already disclosed their performance forecasts are distributed in different links of the optical module industry chain, and their sources of profit are also different.

The core components of an optical module include optical chips (key devices responsible for converting optoelectronic signals and determining the upper limit of the module's speed), DSP chips (responsible for signal modulation and demodulation, designed by companies such as Broadcom and Marvel, and manufactured using advanced processes), FAUs (fiber array units, components that precisely arrange and fix multiple optical fibers for optical signal coupling), and high-speed PCBs.

The reporter learned in the interview that in the cost composition of optical modules, optical components (including optical chips, optical emitting and receiving components) account for about 73% of the total cost, and the optical chips themselves account for about 80% of the cost of optical components. There are currently two mainstream technology routes for optical chips: one is EML (Electro Absorption Modulation Laser), which can directly modulate optical signals on the chip itself, with mature technology but high cost; Another approach is the silicon optical scheme, which combines a CW laser (continuous wave laser, as an external light source) with a PIC chip (a chip that integrates photonic devices on a silicon-based substrate) to achieve the same function. With higher integration and stronger mass production capability, the silicon optical scheme is becoming the mainstream direction of high-speed optical modules.

On this link, Yuanjie Technology produces the most upstream optical chips, including CW lasers and EML lasers; Tianfu Communication produces optical devices for intermediate links such as FAU and optical engine packaging components; Xinyisheng and Dongshan Precision produce complete machine optical modules that are directly supplied to cloud vendors, encapsulating the above components together.

During the interview, several industry insiders told reporters that upstream optical chips are currently in short supply. Industry insiders emphasized to reporters that expanding the production of optical chips is more difficult than the outside world imagines, and improving yield is also quite challenging. Not all manufacturers who announce expansion can truly form effective production capacity.

At present, the global high-speed EML production capacity of over 100G has long been concentrated in the hands of a few manufacturers in the United States and Japan, who adopt the IDM mode (complete chip design and manufacturing independently), with long expansion cycles and slow capacity release.

The aforementioned securities analysts also believe that the supply and demand gap of high-speed EML has continued to remain at a level of 25% to 30% or even higher. The core substrate material used to manufacture EML chips, indium phosphide, has its wafer factories operating at full capacity and its production capacity has been locked in until 2028.

Yuan Yonggang also mentioned at the exchange meeting that two large Japanese companies have stopped producing 100G EMLs and have all switched to 200G EMLs. He also stated that the Sols Optoelectronics 100G EML under Dongshan Precision has been mass-produced, and the 200G EML has the capacity for mass production. It will be shipped within 2026. Yuan Yonggang stated that Sols Optoelectronics is currently the only domestic manufacturer capable of supplying 200G EMLs to North America.

According to the performance forecast, Yuanjie Technology is expected to achieve a revenue of 900 million to 950 million yuan in the first half of 2026, an increase compared to the same period last year

339.13% to 363.53%; It is expected to achieve a net profit attributable to the parent company of 600 million yuan to 650 million yuan, an increase of 1196.91% year-on-year to 1304.98%. Such high profit margins and profit growth rates are a direct manifestation of the current "shortage of optical chips".

But the profit logic of downstream whole machine optical modules is different.

For example, the gross profit margin of New Yisheng for the whole year of 2025 is 47.81%, and for the first quarter of 2026 it is 49.16%. In Wang Cheng's words during the conference call, "the profit margin can basically remain relatively stable". Yuan Yonggang's judgment is that the price of optical modules will continue to decline, but the gross profit margin of top manufacturers will not decrease as a result - economies of scale and efficiency improvements can hedge the impact of price declines.

That is to say, the profit growth of the entire optical module relies on the increase in shipment volume and the intergenerational upgrading of products.

Yuan Yonggang believes that under the same shipment volume, simply switching from 800G to 1.6T can double the revenue. The price of domestic optical modules is currently only about half of that overseas. Dongshan Precision's strategy is to prioritize investing limited materials in high gross profit orders overseas, and only conduct domestic business when prices permit.

Tianfu Communication is located in the middle link of the optical module industry chain. The company's products mainly consist of passive and active optical devices, with a long-term gross profit margin of over 50%. In the first quarter of 2026, it was 56.60%, higher than that of complete machine module manufacturers.

A senior investor who has long been interested in the field of optical communication told reporters that Tianfu Communication's products are indispensable components for the internal components of optical modules. Regardless of which technology route downstream adopts, they cannot do without it. The income stability is strong, but the growth ceiling is lower than that of the whole machine manufacturer.

Tianfu Communication also stated in its semi annual performance forecast that the shortage of certain materials has had an impact on product production, while exchange losses have pushed up financial expenses. These are the two reasons why the company's growth rate in the first half of the year fell to the lowest level in nearly three and a half years.

It is not difficult to see that there is a structural conflict of interests between the upstream and downstream of the industrial chain: when upstream optical chips and DSP chips are scarce, upstream companies enjoy product price increases and high gross profit margins, while downstream optical module companies' capacity ramp up is limited by material supply; Once the upstream production capacity expansion is implemented and the material supply is abundant, the price increase logic of upstream companies will be squeezed, while downstream optical module companies can accelerate the increase in volume.

At present, the head optical module company is increasing its efforts to lock in upstream materials. For example, Yuan Yonggang mentioned at the exchange meeting that Dongshan Precision has prepaid over 1 billion US dollars to upstream DSP chip suppliers to lock in more than 25 million chip production capacity through full issuance of letters of credit.

An optical communication researcher from a private equity firm in southern China analyzed to reporters that the overall supply of DSP chips will be limited in 2027, and upstream manufacturers will prioritize supporting top optical module enterprises, making it difficult for second - and third tier manufacturers to obtain sufficient materials.

Yuan Yonggang summarized this pattern in one sentence during the exchange meeting: the top optical module companies are not worried about the supply of silicon optical PIC chips, because the contract factories are expanding production exponentially, and some products can obtain production capacity as long as the names of cloud vendor customers are reported. But he also emphasized that the company has no plans for self-developed PIC chips and epitaxial wafers, "it's already too late", and the current top customers have specified requirements for silicon optical suppliers, so self-developed is meaningless.

Yuan Yonggang also talked about a further variable.

He believes that when the connections inside the data center cabinets are fully switched from copper cables to fiber optic cables, the usage of optical chips will increase significantly, and the demand inside the cabinets will increase tenfold. In his view, this round of switching may come around 2029, when "light is the main focus, and if copper is converted to light, there will be a huge demand for optical chips".

(Author Zheng Chenye)