Silicon photonics chips are soaring in popularity.

“Market value increased by over 10 times in 8 years” and “Stock price doubled in just a few months, hitting a 20-year high”. These are not empty words but real scenarios playing out in a specific segment of the semiconductor market.

The phenomena mentioned above are the growth performances of Zhongji Innolight and Tower Semiconductor, both of which come from the silicon photonics track.

01 Silicon Photonics Chip Companies See Soaring Performance

Recently, Tower Semiconductor released its financial results for the third quarter. The financial report shows that the third-quarter revenue was $396 million, a 6% increase from the previous quarter. The gross profit was $93 million, higher than the $80 million in the second quarter. The net profit was $54 million. Tower expects the fourth-quarter revenue to reach a record $440 million, with a margin of error of 5%. This means a 14% year-on-year increase and an 11% quarter-on-quarter increase.

Most notably, Tower's silicon photonics business achieved revenue of approximately $52 million in the third quarter, a nearly 70% increase compared to the same period in 2024. The market demand for silicon photonics continues to surge, mainly driven by stronger-than-expected demand for 1.6T products and robust demand for 400G and 800G products.

Domestically, since the second half of 2025, the stock prices of optical module companies such as Zhongji Innolight, Eoptolink, and Accelink Technologies have also accelerated their upward trend, almost simultaneously completing a revaluation.

Zhongji Innolight attributes its performance growth mainly to the increased sales of high-end optical modules such as 800G brought about by the growth of computing infrastructure construction and related capital expenditures, as well as the continuous increase in 1.6T orders and shipments. Meanwhile, the performance reports of Eoptolink and Accelink Technologies are also eye-catching, with significant increases in both the third-quarter revenue and the cumulative revenue for the first three quarters.

On November 10, Zhongji Innolight announced that the company held the 25th meeting of the 5th Board of Directors on November 10, 2025, and approved the "Proposal on Authorizing the Company's Management to Initiate the Preparation Work Related to the Company's Overseas Issuance of Shares (H Shares) and Listing on the Hong Kong Stock Exchange."

In addition, according to Lightcounting's forecast, the global market size of optical modules is expected to grow at a CAGR of 22% from 2024 to 2029, and is expected to exceed $37 billion in 2029. In the Chinese market, the market size of optical modules was approximately 60.6 billion yuan in 2024 and is expected to reach 67 billion yuan in 2025.

Against the backdrop of the overall growth of the optical module industry, silicon photonics technology is maintaining a high-speed growth trend. Behind this trend are multiple factors working together.

02 Silicon Photonics + CPO: This Track is Booming

The core reason for the upsurge in silicon photonics technology is the explosion of AI computing power demand.

Firstly, the large-scale deployment of AI computing clusters has put unprecedented bandwidth pressure on data transmission. The collaborative operation of tens of thousands of GPUs in a single cluster requires ultra-high bandwidth to support the real-time interaction of massive amounts of data. Traditional chips rely on copper wires to transmit electrical signals, but they have low bandwidth, high power consumption, and high latency.

The principle of silicon photonics chips is to integrate micro-optical paths (waveguides, modulators, etc.) on silicon wafers, and data is transmitted in the form of light pulses. Its advantages are higher bandwidth, faster speed, and lower power consumption. This makes it a perfect fit for scenarios such as AI data centers and large-scale training.

Secondly, in addition to silicon photonics, there is another hot term: CPO. Silicon photonics is the technological foundation of Co-Packaged Optics (CPO).

The core logic of CPO is to co-package the optical engine with the computing chip (GPU/ASIC) to shorten the optoelectronic transmission distance (from centimeters to millimeters) and break through the bandwidth and power consumption bottlenecks. Silicon photonics technology can utilize mature semiconductor CMOS manufacturing facilities to achieve high-density integration of photonic devices and electronic chips, making it possible to mass-produce high-performance, low-cost CPO optical engines. Without silicon photonics, the cost of CPO would be difficult to control.

Nvidia launched the Quantum-X Photonics series of CPO switches at the GTC conference, which achieved close-range interconnection between the optical engine and the chip through a silicon interposer. Compared with traditional methods, the energy efficiency has increased by 3.5 times, the signal integrity has increased by 63 times, the large-scale networking reliability has increased by 10 times, and the deployment speed has increased by 1.3 times.

The currently popular supernodes are the core carriers for the implementation of the value of silicon photonics and CPO technologies.

Supernode technology (SuperPod) is a technological architecture that integrates a large number of GPU chips into a unified supercomputing unit through high-bandwidth, low-latency interconnection technology. It breaks through the limitations of the traditional single-server architecture and extends the concept of computing units to the cabinet and even the cross-cabinet level.

Currently, there are two common supernode solutions in the industry: one is to deploy more GPUs by increasing the power consumption of a single cabinet, but there is a bottleneck in increasing the GPU density of a single cabinet due to the power consumption ceiling of data center cabinets. The other is to use optical interconnection technology to build supernodes by increasing the number of cabinets, breaking through the physical limitations of supernodes under traditional interconnection methods. As mentioned above, compared with copper cables, the long-distance transmission advantage of optical cables can decouple delivery from cabinets.

Therefore, the industry is evolving towards optical products with higher integration.

Firstly, for pluggable optical modules, the problem is that the optoelectronic conversion chips are often far from the GPUs, and signal integrity, loss, and latency will all become challenges. Supernodes need to increase the single-channel interconnection bandwidth, as well as the density and number of channels, which has given rise to a new optoelectronic integration technology - near-packaged optics. After moving the optoelectronic conversion chips from the switch to the GPU board, the transmission distance is shortened from 1 meter to 10 centimeters, the interconnection density is increased by 2 - 3 times, and the DSP chip can be removed, greatly reducing the communication latency between GPUs. Near-packaged optics is an interconnection solution that has been mass-produced.

The core technology in the next stage is CPO. It can be said that CPO and silicon photonics chips are the two core technological pillars driving the next-generation data center optical interconnection.

In the future, silicon photonics technology will gradually expand into more application fields. Automotive lidar is one of the incremental blue oceans.

Traditional lidar systems are usually manufactured using discrete mechanical and optical components, which makes it difficult to verify the reliability of the solution and also results in high costs. If thousands of optical and electronic components are combined onto a single chip through semiconductor manufacturing processes, this problem can be well solved.

Liu Jingwei, the founder of Guoke Optoelectronics, once said that silicon photonics technology is the best way to implement lidar. Silicon photonics technology can integrate (or mostly integrate) the lidar system composed of complex optical devices onto a single silicon photonics chip and process it using CMOS technology, achieving excellent performance at a very low cost. In recent years, many optical communication device manufacturers have begun to turn their next application market to the lidar track and integrate the core transceiver functions of lidar into a single chip.

In addition, silicon photonics technology is showing great potential in many aspects such as biosensing and quantum technology.

03 Who is Competing in the Silicon Photonics Chip Field?

Looking at the global development of silicon photonics, the United States was the first to develop silicon photonics and is currently the most advanced country in this field.

As early as 1969, S.E. Miller of Bell Labs first proposed the concept of integrated optics. However, due to the high loss of InP waveguides and the backward process, it was difficult to achieve large-scale integration, and this technology did not make a splash at that time. Later, Intel carried forward this technology. In 2010, after Intel developed the first 50Gb/s ultra-short-range silicon-based integrated optical transceiver chip, silicon photonics chips began to enter the industrialization stage. Subsequently, a group of traditional integrated circuit and optoelectronic giants in Europe and the United States quickly entered the silicon photonics field through mergers and acquisitions to seize the high ground. Currently, Intel is also the company with the most comprehensive layout in the silicon photonics field.

China really started large-scale research on silicon photonics around 2010. Before that, most of the research was academic. The late start has led to China lagging behind the United States in the productization process of silicon photonics. As time entered 2017, China's silicon photonics industry witnessed rapid development.

From the perspective of the global industrial chain progress, US companies represented by Intel and Cisco account for most of the shipments of silicon photonics chips and modules and have become the leaders in the industry. Domestic manufacturers mainly include Zhongji Innolight, Eoptolink, Xilinx Photonics, Huagong Technology, Accelink Technologies, Brocade Communications, Huawei, and Hengtong Optic-Electric.

In the ranking of the top 10 global optical module manufacturers in 2024, Chinese manufacturers performed particularly prominently.

Zhongji Innolight ranked first for the second consecutive year with a revenue growth rate of 114% and a revenue scale of $3.3 billion, further widening the gap with its long-term competitor Coherent.

Eoptolink jumped from the 7th place in 2023 to the 3rd place in 2024 with a revenue growth rate of 175% and a revenue scale of $1.2 billion.

In addition to Zhongji Innolight and Eoptolink, Huawei ranked 4th, Accelink Technologies ranked 6th, Hisense Broadband and Huagong Zhengyuan ranked 7th and 9th respectively. This pattern fully demonstrates the leading position of Chinese manufacturers in the global optical module market.

In Q2 2025, the global sales of optical modules increased by 10% quarter-on-quarter, mainly contributed by 800G Ethernet optical modules, and 1.6T products also made contributions. As cloud computing giants such as Amazon and Meta transition to higher-speed optical modules, the sales of 400G Ethernet optical modules will decline.

It is not difficult to find that the deployment speed of 800G has significantly accelerated, driving the growth of the optical module market to a certain extent. Coherent also previously stated that the deployment speed of 800G is significantly faster than that of 400G. Meanwhile, 800G products are still high-margin products, and the high growth of net profit is traceable.

Lightcounting predicts that the global shipments of 800G optical modules will exceed 5 million in 2025, and the proportion of the LPO solution is expected to exceed 40%. In 2023, this figure was less than 500,000. 800G optical modules, especially products using LPO technology, are in a stage of rapid development and are playing an increasingly crucial role in the era of the explosion of AI computing power, continuously promoting the reshaping and transformation of the data center optical interconnection pattern.

After 800G optical modules, as AI server clusters put forward higher requirements for interconnection rates, Nvidia has chosen to switch to 1.6T optical modules in the GB300 server and also provides an option to upgrade to 1.6T optical modules in the GB200. Therefore, 1.6T optical modules have begun to take the stage.

Regarding the future development forecast, the 1.6T rate will be entered in the next 1 - 2 years. It is expected that by 2029, the optical module rate for AI applications will reach 3.2T, and 3.2T will move towards large-scale application in 2030.

Accelink Technologies and Zhongji Innolight have both completed the verification of their 1.6T products.

In the second quarter of this year, Zhongji Innolight began small-scale shipments of 1.6T optical modules, and it is expected that continuous large-scale or mass shipments will be achieved in the second half of the year. At the 2025 OFC Optical Expo, a subsidiary of Zhongji Innolight demonstrated 1.6T-DR8 and 2xFR4 OSFP optical module products equipped with a new 3nm DSP chip, as well as 1.6T-2xLR4 optical modules based on TFLN MZM technology.

Accelink Technologies has also launched 1.6T DR8 series and 2FR4/4FR2 series products based on the OSFP-XD package. The 1.6T OSFP 2VR4 optical module based on VCSEL technology is planned to enter the mass production stage in the fourth quarter of this year, and the 1.6T LRO optical module has received orders.

Recently, Nomura released a research report stating that as the shipments of 1.6T optical modules accelerate and the penetration rate of silicon photonics (SiPh) technology increases, Zhongji Innolight is expected to become the core beneficiary of the industry's upward cycle. The investment bank has significantly raised its performance expectations for the company from 2026 to 2027, with the revenue expectation increased by 44% - 53% and the net profit expectation increased by 64% - 75%. Based on the expected earnings per share of 20.41 yuan in 2026, Nomura gives a price-to-earnings ratio valuation of 30 times. Nomura is particularly optimistic about Zhongji Innolight's leading position in the high-speed optical module market and expects the company to occupy 30% - 35% and 40% - 45% of the 800G and 1.6T markets respectively.

This article is from the WeChat official account "Semiconductor Industry Insights", author: Feng Ning. Republished by