Indium Phosphide has a huge supply gap.

The demand for optical interconnection in AI data centers is growing rapidly. As Co-packaged Optics (CPO) gradually moves toward commercial application, a previously less noticed supply chain is facing new tests: the production capacity of Indium Phosphide (InP) lasers.

Judging from the CPO switches that have been shipped so far, the external light sources still mainly use lasers based on indium phosphide materials. Lumentum and Coherent are major suppliers in this field, and both companies are accelerating the construction of production capacity for high-power lasers. However, the problem is that the demand for traditional pluggable optical modules is also growing. CPO does not develop independently after replacing the original market, but adds a new driving force on top of the existing demand.

Many suppliers have clearly stated that the current indium phosphide production capacity cannot meet the demand. Combined with the capacity expansion plans of various companies, this tight situation may last until 2028.

CPO shipments begin to rise, and laser demand changes accordingly

The core idea of CPO is to integrate the optical engine and the switching chip more closely, shorten the transmission distance of high-speed electrical signals, so as to reduce power consumption and improve the overall performance of high-speed interconnection. However, optical signals still need to be provided by lasers. Therefore, even if the integration method of the optical engine and the switching chip changes, the light source remains an indispensable part of the entire system.

In Nvidia's CPO switch solution, the laser is placed outside the switch package, and the external light source module provides optical signals for the system. This design facilitates on-site replacement of the light source, and also separately raises the issues of laser reliability and maintenance.

Nvidia's solution uses an external light source module containing 8 lasers. According to its introduction, compared with the traditional design, this architecture reduces the number of lasers by four times.

However, the reduction in the number of lasers does not mean lower requirements for lasers. On the contrary, a single laser needs to provide higher output power to meet the needs of multiple optical channels.

For example, Lumentum's laser products for CPO and NPO cover power levels of 120mW, 150mW and 400mW, and its ultra-high-power CPO laser can reach a maximum output power of 350mW at 50°C. Broadcom's launched CPO external laser module uses 8 CWDM DFB lasers, each with an output power of 100mW, and the total output power of a single module reaches 800mW.

AOI (Applied Optoelectronics) also pointed out that the power of its CPO laser reaches 300mW, while the laser used in 1.6T pluggable optical modules is about 100mW. Higher power requirements mean larger chip sizes, and manufacturing yield is also under pressure.

In addition, CPO has stricter reliability requirements. When one laser needs to provide light sources for multiple channels, once a failure occurs, the scope of impact will expand. Especially in high-value switch systems, users do not want to frequently disassemble and replace equipment. Therefore, high-power lasers not only need to provide sufficient optical power, but also must have stable long-term operation capabilities.

This also means that what CPO brings is not a simple increase in the number of lasers, but an increased demand for high-power, high-reliability lasers.

Pluggable optical modules are still growing, and competition for production capacity becomes more intense

If CPO has already started to seize laser production capacity, the continuous growth of the traditional pluggable optical module market has further exacerbated supply pressure.

Cignal AI predicts that the shipments of 1.6T optical modules will exceed 10 million units in 2026. LightCounting previously predicted that from 2025 to 2030, the shipments of optical transceivers with a rate of 800G and above will triple.

This means that in the next few years, the demand for high-speed optical interconnection in data centers will not slow down due to the development of CPO. Pluggable optical modules will still undertake a large number of interconnection tasks, and CPO will gradually enter new application scenarios. The two will coexist for a fairly long period of time.

And the supply chains behind these two types of products have obvious overlaps.

Taking Coherent as an example, its 6-inch indium phosphide production lines located in Texas, USA and Sweden simultaneously produce products such as electro-absorption modulated lasers (EML), continuous wave lasers (CW Laser) and photodetectors. Lasers required by traditional optical modules and CPO both need to compete for limited manufacturing resources.

Therefore, the growth of CPO does not create a completely independent demand outside the original optical module market. At least at the current stage, it is more like adding a new demand for high-power lasers on the basis of the continuous growth of pluggable optical module demand.

Laser chip manufacturing becomes the tightest link

From indium phosphide materials to the final external light source module, the entire supply chain can be roughly divided into several links: substrate, epitaxy, laser chip and light source module. At present, all links are expanding production, but the pressure is different.

The first is the substrate. Sumitomo Electric plans to expand its indium phosphide substrate production capacity to about three times the original level between 2024 and 2028. This plan is calculated by wafer area and is not limited to the data center market.

JX Advanced Metals is also increasing investment. The company said in June 2026 that customers continue to require a significant increase in supply, and the actual demand is significantly higher than previously expected. The company plans to invest up to 120 billion yen in the next four years to increase production capacity to 7 to 10 times the level of fiscal year 2025, while requiring customers to adjust product prices.

The capacity expansion plans of the two material suppliers reflect the market's expectation of the growth of indium phosphide demand in the next few years. However, the expansion of substrate production capacity cannot be immediately converted into laser supply, and subsequent links such as epitaxy, chip manufacturing and testing are still required.

The situation AXT faces is more complicated. The company's management said in July 2026 that customer demand still exceeds its capacity expansion speed, and the amount of backlogged orders has exceeded 100 million US dollars. At the same time, its indium phosphide materials are produced in China, and exports still need to obtain licenses. Quarterly documents disclosed by the company in August show that some orders are still waiting for export licenses.

This means that AXT's supply is not only limited by production capacity, but also affected by the progress of export approval. Even if market demand is strong, orders may not necessarily be converted into actual deliveries in time.

The epitaxy link is also expanding. IQE raised its full-year revenue guidance in July 2026, expecting revenue growth of more than 30%. One of the reasons is the accelerated growth of indium phosphide demand. The company also plans to shift existing equipment to indium phosphide production in the second half of the year.

The business of Taiwan, China-based vendor LandMark Optoelectronics has also grown. Its revenue in September 2026 increased by 190% year-on-year, and the company attributed the growth to increased data center-related shipments. The company previously signed a five-year substrate supply agreement with Sumitomo Electric, and approved equipment investment of up to 3 billion New Taiwan dollars in June 2026.

From substrates to epitaxy, suppliers are increasing investment. But the real bottleneck is still concentrated in the laser chip manufacturing link.

Lumentum and Coherent expand production, but still cannot quickly fill the gap

Lumentum admitted at its earnings call in August 2026 that the shipments of its high-power lasers have significantly lagged behind demand. The company expects that the demand for ultra-high-power laser chips will begin to accelerate in the second half of 2027, and further increase after customers expand deployment in 2028.

This also means that the tight supply of lasers is not because CPO has been fully popularized, but before large-scale deployment has really arrived, the supply chain has already begun to bear pressure.

Coherent's judgment is similar. The company's management called indium phosphide production capacity the main limiting factor at present, and said that the demand for data center optical transceivers is consuming existing production capacity.

To alleviate this problem, Coherent plans to increase its internal indium phosphide output by more than 100% year-on-year by the end of September 2026, and plans to achieve another growth of more than 100% by the end of 2027. At the same time, the company is advancing the construction and ramp-up of the 6-inch indium phosphide production line, and said that the yield of this production line is already better than that of the 3-inch production line.

However, capacity expansion takes time, and the increase in production capacity does not mean that all products can immediately get sufficient supply. For Coherent, both traditional optical transceivers and CPO lasers need indium phosphide manufacturing resources, and how to allocate limited production capacity is still a practical problem.

AOI's situation can better reflect this contradiction. The company manufactures indium phosphide lasers by itself to supply its own optical transceiver products. The management said in August 2026 that major CPO customers are interested in its laser products, but the current production capacity is not enough to allow the company to participate in the first-generation large-scale deployment of these customers, so it must prioritize meeting the production demand of its own optical transceivers.

In other words, even if CPO customers are willing to purchase, suppliers may not have enough lasers for sale.

AOI plans to expand production capacity in the United States and Taiwan, China in 2026 and 2027, focusing on supporting the production of laser diodes, 800G and 1.6T optical modules. Its external laser light source product ELSFP for CPO is expected to gradually increase in volume in 2027, and advance towards the target of monthly output of about 400,000 units in 2028.

The situation of these companies shows that the current supply bottleneck cannot be solved simply by increasing orders. Laser chip manufacturing involves production line construction, process optimization and yield improvement. Even if manufacturers have formulated clear capacity expansion plans, new production capacity still needs a period of time to form stable supply.

New players enter the market, and alternative technologies still need time

Facing the supply pressure of indium phosphide lasers, the industry is also looking for new production capacity sources and technical routes.

Swedish company Sivers Semiconductors announced in September 2026 that it will invest 30 million US dollars to expand its indium phosphide plant in Glasgow, UK, planning to increase the annual production capacity of continuous wave DFB lasers to more than 100 million units, in addition to the production capacity provided by foundry partners.

The project is expected to be put into operation in the fourth quarter of 2027, which exactly falls at the stage of accelerated growth of CPO demand. The company's management said that customers need to significantly increase laser production capacity.

This means that in addition to existing major suppliers such as Lumentum and Coherent, other manufacturers also hope to expand their market share with the help of this new round of demand growth. However, whether the new factories can be put into operation as planned and how fast the actual production capacity ramps up remain to be verified subsequently.

On the other hand, the industry is also trying to use lasers of different materials and structures.

For example, Coherent has already demonstrated a CPO solution using gallium arsenide VCSEL; Quintessent has begun to provide quantum dot laser samples for AI optical interconnection applications; micro LEDs are also regarded as one of the potential technical routes for future CPO light sources.

However, it should be noted that these solutions are still in the demonstration or sample stage. According to existing information, there is no evidence that they have entered the currently shipped CPO switches and replaced indium phosphide lasers on a large scale.

Therefore, in the short term, the industry cannot expect new technologies to quickly change the existing supply pattern. Indium phosphide will still be the main material platform in this market. Whether other technologies can truly achieve large-scale application depends on subsequent product verification, reliability performance and customer introduction progress.

Before 2028, the contradiction between supply and demand is unlikely to ease

Judging from the expansion rhythm of all links in the industrial chain, the supply pressure of indium phosphide lasers may continue for several more years.

On the one hand, CPO is moving from the stage of product verification and early procurement to large-scale deployment. Lumentum has disclosed the first batch of external light source module procurement orders, which are expected to be delivered in the second half of 2027; Coherent's high-power continuous wave lasers have also begun to ramp up, and related revenue is expected to grow starting from the fourth quarter of 2026.

On the other hand, the demand for traditional pluggable optical modules is still expanding. Products such as 800G and 1.6T continue to promote the upgrading of data center optical interconnection, and consume the production capacity of the indium phosphide industrial chain together with CPO.

The capacity expansion plans of suppliers are also worthy of attention. Sumitomo Electric's substrate expansion plan extends to 2028, and JX Advanced Metals' capacity improvement plan covers the next four years. Laser manufacturers such as Coherent, Lumentum and AI are also accelerating capacity expansion, but their descriptions of the existing supply and demand gap are not optimistic.

LightCounting predicts that by 2031, the indium phosphide laser market will reach 6.9 billion US dollars, and will remain one of the largest sub-sectors in the optical chip market.

Judging from the current industrial progress, what is worth paying attention to in the next two years is not only how much the shipment volume of CPO switches can grow, but also whether high-power lasers can be delivered on time, and whether the capacity expansion plans can be truly converted into effective production capacity.

For Lumentum and Coherent, the core task is to expand production capacity, improve yield, and meet the needs of traditional optical module and CPO customers at the same time; for AXT, in addition to capacity expansion, the progress of export licenses is also an important variable affecting delivery; Sivers hopes to catch up with the next round of market growth through the new production capacity put into operation in 2027.

The popularization of CPO is usually regarded as an important change in the optical interconnection architecture. But in this change, what determines whether the product can be smoothly ramped up may not only be the switching chip, optical engine or system design. Those seemingly insignificant substrate and laser chip manufacturing links may also determine the growth rate of the entire industrial chain.

At least at the current stage, indium phosphide lasers are still one of the most noteworthy bottlenecks in the CPO supply chain. With the simultaneous growth of CPO and pluggable optical modules, the competition for production capacity before 2028 is far from over.

This article is from the WeChat official account "Semiconductor Industry Watch" (ID: icbank), author: Editorial Department, released with authorization from 36Kr.