Optical chips, collective capacity expansion
It has to be said that the demand for optical chips is extremely high.
In recent days, a series of actions including capacity expansion, long - term agreements, investments, and supply chain bindings have emerged intensively in the global optical chip industry chain: Coherent is expanding its 6 - inch InP compound semiconductor production line in Sherman, Texas; Nokia is expanding the advanced testing and packaging capacity of photonic chips in Allentown, Pennsylvania, USA; Japan's JX Advanced Metals plans to invest up to 120 billion yen to increase the production capacity of InP substrates by 7 - 10 times; IQE has reached a multi - year InP epitaxial wafer supply agreement with Tower Semiconductor; Domestically, Solstice Optoelectronics, a subsidiary of Dongshan Precision, also announced a project to expand optical chips and high - speed optical modules in Changzhou, with a total investment of up to $1.2 billion.
A capacity race around the optical interconnection capabilities of AI data centers has already begun.
The Big Picture of Global Optical Chip Enterprises' Capacity Expansion
First, let's look at the capacity expansion actions in the United States.
On June 16, Coherent announced that it had signed a letter of intent to receive up to $50 million in direct funding from the U.S. Department of Commerce under the CHIPS and Science Act to expand its world - leading 6 - inch indium phosphide (InP) semiconductor manufacturing plant in Sherman, Texas. The day after the announcement, Coherent held a groundbreaking ceremony for the expansion at its Sherman, Texas factory. Coherent emphasized that this base has the world's first and currently largest - scale 6 - inch InP manufacturing platform. After the expansion is completed, the manufacturing space of the factory will double, and the wafer production capacity will increase fourfold.
Notably, Huang Renxun, the founder and CEO of NVIDIA, personally attended Coherent's ceremony and was on stage with Coherent's new CEO, Jim Anderson. NVIDIA had previously announced a strategic investment of $2 billion in Coherent to secure the future production capacity of its most advanced lasers, optical engines, and optical modules. Huang Renxun said in his speech on site: "AI runs on computing power, but large - scale implementation is stuck at the connection. And the Sherman factory is the place to build these 'connection neural tissues'."
Image source: techpowerup
Nvidia has incorporated "light" into the AI infrastructure supply chain with capital. As early as March this year, Nvidia announced investments of $2 billion in Coherent and Lumentum respectively, along with multi - year procurement commitments and future production capacity/access rights, for the expansion of advanced lasers, optical networking products, R & D, and U.S. manufacturing capabilities.
Lumentum is also an important part of the U.S. optical chip capacity expansion picture. In March, Lumentum announced that it would build a new advanced laser manufacturing factory in Greensboro, North Carolina, USA. The factory covers an area of about 240,000 square feet and focuses on producing indium phosphide (InP) optical devices for large - scale global AI data centers. In May, AIXTRON announced that it had received orders for multiple G10 - AsP MOCVD systems from Lumentum. Lumentum's stock price has risen by 769% in the past year.
Also on June 16, Nokia announced that it would expand the advanced testing and packaging capabilities of photonic chips in Allentown, Pennsylvania, USA, that is, to further package photonic chips into optical modules for AI and communication infrastructure. Nokia said that this base is one of the few facilities in the United States with such capabilities. After the expansion, the production capacity will increase up to 10 times the current level, and it is expected to have commercially available production capacity by the end of the third quarter of 2026.
Nokia is supplementing the photonic chip packaging, testing, and modularization capabilities, while Coherent is supplementing the front - end manufacturing capabilities of InP photonic devices. Nvidia's previous investments in Coherent and Lumentum are equivalent to locking in funds, orders, and production capacity in advance for the core suppliers of lasers and optical networking. The United States is incorporating the optical interconnection of AI data centers into its domestic semiconductor manufacturing system.
Japan is supplementing the upstream material field, which is also an area where the Japanese semiconductor industry has long excelled.
On June 16, Japan's JX Advanced Metals, one of the world's two major InP substrate suppliers, announced that it plans to invest up to 120 billion yen in the next four years to expand the production capacity of InP substrates. Adding to the previously announced related investments, the total investment scale of the company's InP production capacity construction will reach about 150 billion yen. These investments will increase the company's production capacity by 7 to 10 times.
JX Advanced Metals has been producing indium phosphide substrates since the 1980s. In the 2025 fiscal year, the company invested 25 billion yen to increase the production capacity of this material. According to a report by India's Strait Research, it is estimated that by 2034, the global indium phosphide wafer market will reach $507.21 million, almost three times that of 2025. Currently, JX Advanced Metals and its competitor Sumitomo Electric each account for about 40% of the market.
Europe also has several key actions.
When discussing optical communication in the market, "silicon photonics" and "InP" are often put in opposition: it seems that after the popularization of silicon photonics, InP will be replaced. Coupled with the previous intellectual property (IP) lawsuit between IQE and Tower Semiconductor, it is easy to think so. However, the real industrial path is more complex, which can be seen from the actions of IQE and Tower.
On June 15, IQE and Tower Semiconductor reached a multi - year InP epitaxial wafer supply agreement to support the mass - production expansion of Tower's silicon photonics platform in the directions of 200Gb/channel pluggable transceivers, next - generation 400Gb/channel modulators, and optical path switching. The agreement stipulates that Tower needs to make a minimum procurement commitment in the first year, and IQE needs to make a corresponding supply commitment, and then also make a minimum procurement volume commitment. This also shows a trend: the next - generation silicon photonics platform does not completely get rid of III - V materials, but needs to integrate high - performance InP components into the mature silicon photonics platform. Silicon photonics is responsible for large - scale integration, CMOS process compatibility, and platform - based manufacturing, while InP continues to undertake key functions such as high - performance light sources, modulation, and photoelectric conversion.
According to another agreement, Tower will also provide IQE with a wide - ranging global royalty - free license for porous silicon patents. Previously, there was an intellectual property dispute between the two companies, and Tower will reach a settlement on this issue and resolve all lawsuits.
Tower pointed out in its first - quarter 2026 financial report released on May 13 this year that it is implementing an aggressive global multi - fab silicon photonics capacity expansion plan, aiming to increase the monthly output capacity of silicon photonics wafers to more than five times that at the end of 2025 by the end of 2026. Moreover, Tower announced that it has signed long - term silicon photonics supply contracts worth up to $1.3 billion for 2027 with several core major customers and directly received an advance payment of $290 million from customers in the first quarter of 2026. As the equipment in multiple factories is gradually installed, Tower's cumulative total global asset investment in silicon photonics - related processes, equipment, and packaging will reach about $920 million.
In March 2026, ST issued a news release saying that it is considering modular capacity expansion in Crolles, France, aiming to quadruple the 300mm silicon photonics production capacity by 2027 and further plan for subsequent capacity expansion in 2028. In addition, this project is also supported by the European sovereign supply chain plan. ST's PIC100 silicon photonics process platform based on the 300mm wafer line has entered the full - scale high - production stage for the world's top cloud providers, mainly used for the core chips of 800G and 1.6T optical transceivers.
On June 2, Swedish chip manufacturer Sivers Semiconductors (specializing in providing high - power multi - wavelength laser arrays) reached a deep strategic cooperation with U.S. pure - foundry giant GlobalFoundries (GF) to develop next - generation optical connection solutions specifically for AI data center infrastructure. Specifically, Sivers' advanced laser arrays will be directly integrated into GF's silicon photonics platform.
Domestically, the optical chip industry is in a state of rapid development.
According to industry statistics from Securities Times·Data Treasure, as of the first quarter of 2026, the total scale of ongoing projects of seven domestic listed core optical module enterprises has risen to 3.898 billion yuan. Compared with four years ago (the same period in 2022), this figure has increased by more than six times. Zhongyou Securities pointed out in a research report that overseas giants account for 95% of the global indium phosphide market, and the overall supply - demand gap in the indium phosphide industry is nearly 70%. It is expected that the high - growth trend will continue until 2028.
On the evening of June 16, Dongshan Precision announced that it agreed to let its wholly - owned subsidiary Solstice Optoelectronics and its subsidiaries layout an optical chip and high - speed optical module expansion project in Changzhou, with a total investment of $1.2 billion. The project funds will be self - raised by the company. Solstice is a vertically integrated enterprise with the capabilities of optical chip design, manufacturing, packaging, optical module assembly, and testing. After Dongshan Precision acquired Solstice, it has entered the core link of AI optical communication from the traditional electronic manufacturing and consumer electronics industry chains.
In terms of financial contribution, after Solstice was consolidated, its profit contribution to Dongshan Precision is significantly higher than its revenue share. In 2025 and the first quarter of 2026, Solstice's revenue share after consolidation was 3.58% and 16.02% respectively, while its profit share reached 22.69% and 52.92% respectively. This shows that the optical communication business not only has rapid growth but also strong profit elasticity. This is why Dongshan Precision is willing to invest $1.2 billion.
On June 3, Sanan Optoelectronics replied on the interactive platform that its indium phosphide (InP) epitaxial growth, chip manufacturing, and packaging and testing processes are leading in China. It has the process ability to mass - produce 6 - inch InP optical chips and said that the company's optical technology production capacity is 2,750 wafers per month, and the core epitaxial link has been expanded to nearly 6,000 wafers per month. In terms of products, Sanan Optoelectronics mentioned in its 2025 annual report that the company can provide lasers and detector chips such as CW light sources, VCSELs, EMLs, and PDs for optical modules. Among them, the optical chips for 400G and 800G optical modules have been mass - produced, and the optical chips for 1.6T optical modules have been sent to customers for verification.
In the material segment, in April this year, Yunnan Germanium Industry officially launched the "High - Quality Indium Phosphide Single - Crystal Wafer Construction Project". This project plans to expand a production line with an annual output of 300,000 wafers (equivalent to 4 - inch, including 6,000 6 - inch wafers). On the basis of the existing annual output of 150,000 wafers, the total production capacity will eventually reach 450,000 wafers per year, with a construction period of 18 months. Currently, industry verification and equipment installation are being carried out as planned, and the production capacity will be gradually released with the construction progress.
The domestic optical chip industry chain is being completed from "module assembly" to the entire chain of "materials - epitaxy - chips - packaging and testing - modules".
The Growth of Optical Chips Is an Established Fact
As we all know, in the field of optical chips, CPO is the "holy grail" of the industry. However, the implementation speed of CPO has been postponed. Therefore, the industry also has a huge concern about the optical communication sector: if the future CPO (Co - Packaged Optics) fails to be implemented for a long time or weakens, will optical module companies lose their growth potential?
The latest optical report from Morgan Stanley gives a very clear refutation. Morgan Stanley points out that investors pay too much attention to the time node of "when to use CPO" and ignore the underlying constant - the demand for bandwidth growth.
Regardless of whether the market ultimately expands through pluggable optics, NPO, CPO, OBO, or a hybrid architecture, the demand for higher bandwidth should continue to drive the increase in optical engines, lasers, and related components per GPU/rack. Morgan Stanley's view is that how the architecture evolves is just a route issue, but the overall increase in the usage of optical components is certain.
What are CPO, NPO, and pluggable?
Traditional Pluggable: The optical module is like a USB flash drive and is plugged into the front panel of the switch. It is connected to the internal switching chip (ASIC) through copper wires.
NPO (Near - Packaged Optics): The optical engine is moved inside the switch, next to the switching chip, to shorten the copper wire distance.
CPO (Co - Packaged Optics): The optical chip and the switching chip (or GPU) are directly packaged on the same substrate, completely eliminating long - distance copper wires and reducing power consumption and latency to a minimum.
Currently, CPO indeed has fatal pain points such as extremely complex packaging, low yield, and the possibility of the entire motherboard being scrapped if one component fails (poor maintainability/serviceability). Therefore, the large - scale popularization of CPO will probably slow down. But even if the market does not use CPO in the short term and continues to use traditional pluggable optical modules or adopts a "copper/CPO hybrid route", the number of optical engines and lasers corresponding to each AI server and each GPU is still increasing significantly.
The controversy over CPO is not only about the packaging location but also about the light source route. The essence of CPO is to place the optical engine as close as possible to the switching chip or computing chip to shorten the transmission distance of high - speed electrical signals and reduce power consumption and bandwidth bottlenecks. However, there is no single answer for the light source in the industry at present.
Currently, there are three main routes that have received more attention: SiPh + CW Laser (Silicon Photonics + Continuous - Wave Laser), VCSEL (Vertical - Cavity Surface - Emitting Laser), and MicroLED (Micro - Light - Emitting Diode). The differences in maturity, cost, distance, and power consumption of different routes determine that CPO is likely not to be implemented in a single form but will form a co - existence of multiple solutions at different distance levels in AI data centers.
The SiPh + CW Laser, that is, the "silicon photonics chip + continuous - wave laser" solution, has the highest technological maturity, and the effective transmission distance can exceed 1 kilometer. It is more suitable for connections in data centers with high requirements for bandwidth, distance, and reliability. However, there are still problems with system - level power consumption, coupling packaging, and cost.
The advantages of VCSEL lie in high energy efficiency, low cost, and strong arraying ability, and its technological maturity is also relatively high. However, the effective distance is usually limited to within 100 meters, and it is more suitable for short - distance interconnection within or between cabinets. Therefore, the positioning of VCSEL is not to replace SiPh + CW Laser but may become a supplementary solution in short - distance, low - cost, and high - density optical interconnection scenarios.
MicroLED is more like a potential solution for the future. It has the potential for low latency, low cost, and high energy efficiency, but the effective distance is shorter, and the technological maturity is the lowest. This is a "dark horse" route that has attracted much attention in the field of optical interconnection in recent years. Silicon photonics chip startup Ayar Labs and others are actively exploring the introduction of MicroLED originally used in the display field into high - density proximal optical interconnection at the Chiplet level. It mainly uses extremely small - sized (micron - level) LED arrays as light sources, which are directly integrated on the edge or substrate of computing chips (such as GPUs and HBMs), and the MicroLEDs are directly driven by electrical signals to flash and emit light for data transmission.
It can be seen that in the future, CPO will probably