Economic Observer reporter Zheng Chenye

On July 8th, Michael Helston, President and CEO of Lumentum, an American optical chip giant, stated at the European AI Industry Summit held in Paris that the supply and demand gap of indium phosphide has exceeded that of DRAM (Dynamic Random Access Memory) and NAND (Flash Memory). The order volume of the company's telecommunications customers has jumped from "hundreds of items" to "hundreds of millions of items", and the orders have been scheduled until 2028.

Lumentum is one of the hottest companies in the current AI computing power chain, supplying laser chips required for data center optical interconnection to customers such as NVIDIA. Michael Helston's statement has brought a semiconductor substrate material that most people have never heard of into the spotlight.

In the AI data center, the optical module responsible for transmitting data between servers has laser chips grown on indium phosphide substrates. According to the reporter's understanding, an 800G (gigabits per second) optical module requires 4-8 indium phosphide based laser chips, and the chip consumption of a 1.6T (terabit per second) optical module is nearly three times that of an 800G. Moreover, with each upgrade of the speed of the optical module, the consumption of indium phosphide does not increase proportionally, but accelerates, because higher speed means more laser chips and higher single power.

According to data from Jibang Consulting, the global shipment of 800G and above optical modules is expected to exceed 63 million units by 2026, accounting for 60% of the overall shipment volume.

In March 2026, Nvidia invested $2 billion each in Coherent (a leading American photonics technology company that operates the world's first mass-produced 6-inch indium phosphide production line) and Lumentum, signing long-term procurement agreements.

On June 16th, Nvidia CEO Huang Renxun appeared at the groundbreaking site for the expansion of Coherent's indium phosphide wafer fab in Sherman, Texas. Coherent CEO Jim Anderson stated at the groundbreaking ceremony that over 80% of the company's demand for indium phosphide comes from AI data centers.

According to publicly available information, China holds about 70% of the global production of refined indium, which is the core raw material for synthesizing indium phosphide. Indium has been included in export controls since February 2025. However, in the finished product market of indium phosphide substrates, Sumitomo Electric of Japan, AXT of the United States (its subsidiary Beijing Tongmei is one of the world's major suppliers of indium phosphide substrates), and JX Metal of Japan together account for more than 90% of the market share, while the total shipment volume of all domestic manufacturers accounts for only about 10% of the global supply.

In the A-share market, indium phosphide concept stocks have seen a doubling trend since the beginning of this year. Among them, Yunnan Germanium Industry (002428. SZ), one of the listed companies with the largest production capacity of indium phosphide substrates in China, has a market value exceeding 60 billion yuan, with a stock price increase of over 200% since the beginning of this year and a dynamic P/E ratio of over 1700 times.

Serious Gap

According to relevant announcements, the prepaid balance of A-share optical module leader Zhongji Xuchuang (300308. SZ) at the end of the first quarter of this year jumped from 134 million yuan at the end of last year to 1.488 billion yuan, an increase of 10 times. This advance payment is mainly used to lock in the upstream optical chip and substrate production capacity in advance.

Zhou Zihao, the market manager of a metal material company in Suzhou, told reporters that the delivery time for downstream customers to receive indium phosphide substrates this year generally reaches 24 to 40 weeks, and some specifications require a deposit of 30% to 50% in advance to be listed.

Indium phosphide occupies a unique position in the optical module industry chain. The core function of an optical module is to convert electrical signals into optical signals for transmission, and then restore the received optical signals into electrical signals to complete the transmission of data in optical fibers. In this process, the laser chip responsible for emitting light is the key component, and the manufacturing of the laser chip requires a layer of "foundation" - the substrate, on which all functional layers of the chip are grown.

Indium phosphide is the substrate material used to manufacture such laser chips. Its physical properties determine that the wavelengths emitted by lasers based on this material are concentrated around 1310 nanometers and 1550 nanometers, which happen to be the windows with the lowest fiber transmission loss. Currently, no other material can simultaneously meet the three conditions of high-frequency modulation, efficient emission, and low transmission loss.

A communication analyst from a large securities firm in southern China stated that in the current mainstream 800G optical modules, the laser generally adopts the EML scheme, which is an electroabsorbent modulation laser that integrates laser emission and signal modulation on one chip. An 800G optical module requires 8 EML chips.

At the 1.6T stage, the technological roadmap of optical modules begins to differentiate, and silicon optical solutions are expected to occupy more than 60% of the market share. The approach of silicon photonics is to transfer the signal modulation function to silicon-based chips for lower power consumption and cost. Moreover, a 1.6T silicon optical module no longer uses EML chips, but instead uses 2 to 4 CW lasers (continuous wave laser sources), which only provide stable light source output and modulation is completed by silicon optical chips.

Zhou Zihao stated that the substrate of CW lasers is also indium phosphide, and the silicon optical scheme only replaces the material in the modulation stage, without reducing the demand for the emission stage.

The reporter also learned during the interview that the industry is currently focusing on a more cutting-edge modulation material, thin film lithium niobate (TFLN), which may be applied in future 3.2T optical modules. However, thin film lithium niobate can only solve the modulation problem, and the light source still cannot do without indium phosphide.

No matter how the modulation technology route of optical modules evolves, from 800G, 1.6T to 3.2T, each generation of products requires indium phosphide. The higher the speed, the higher the requirements for the number and power of laser chips, and the consumption of indium phosphide also increases accordingly.

The aforementioned analyst believes that there is no alternative solution for indium phosphide in the field of data center optical interconnects for at least the next 5 years.

According to market research firm Yole, the global demand for indium phosphide substrates is expected to reach 2.6 million to 3 million pieces (equivalent to 2 inches) by 2026, but the effective production capacity is only about 750000 pieces, with a gap of over 70%. Due to the fact that the area of a 4-inch substrate is approximately equal to four 2-inch substrates, the 4-inch caliber is commonly used within the industry chain to measure actual production capacity and demand.

Zhou Zihao stated that equivalent to a 4-inch diameter, the global demand for indium phosphide substrates is expected to approach 2 million pieces by 2026, climb to 2.5 million to 2.8 million pieces by 2027, and may further increase to 4 million to 5 million pieces by 2028.

At present, more than 90% of the global production capacity of indium phosphide substrates is concentrated in the hands of three companies: Sumitomo Electric of Japan, AXT of the United States, and JX Metal of Japan. Faced with the rapid expansion of demand, the three leading companies in the industry are all intensifying their efforts to expand production.

JX Metal from Japan is a leading global supplier of semiconductor sputtering targets, with indium phosphide substrates being its key business expansion in recent years. In June of this year, the company announced a cumulative investment of 120 billion yen (approximately 5.8 billion yuan) by the fiscal year 2030 to increase indium phosphide production capacity by 7 to 10 times. This is also the largest single investment in the history of the company's semiconductor materials business.

AXT completed a public offering on NASDAQ in April this year, issuing over 8.56 million shares at a price of $64.25 per share, raising approximately $632.5 million for expansion. The management of the company stated during a conference call in May this year that the backlog of orders for indium phosphide has exceeded $100 million.

According to publicly available information, Sumitomo Electric of Japan also plans to invest approximately 18 billion yen to expand its indium phosphide substrate production capacity to 3.1 times the level of the 2024 fiscal year by the end of the 2028 fiscal year (i.e. March 2029).

However, the expansion pace of indium phosphide is different from that of the chip manufacturing industry, where funding and factory construction are just the starting points.

Zhou Zihao stated that the growth of indium phosphide single crystals requires high temperature and high pressure conditions of over 1000 degrees Celsius and about 27 atmospheres. A furnace of polycrystalline materials weighs only a dozen kilograms, and the crystallized single crystals undergo multiple processes such as cutting, thinning, grinding, and polishing, each of which generates losses. More importantly, in the customer certification process, substrate manufacturers must send samples to downstream laser manufacturers for device verification and reliability testing. The entire process, from equipment debugging to downstream certification, usually takes one to two years.

Jim Anderson stated during a recent earnings call that the supply-demand imbalance of indium phosphide will continue at least throughout 2026 and 2027.

Behind the tight supply lies a deeper resource constraint: Indium phosphide is synthesized from high-purity indium and high-purity phosphorus, with the cost of indium accounting for the majority, while the supply elasticity of indium itself is almost zero.

During the interview, the reporter learned that indium is a typical associated metal that is entirely attached to the smelting process of lead, zinc, and tin mines. There is no independent indium mine in the world, and its production depends entirely on the mining scale of the main mine. It is impossible to expand production alone due to the rise in indium prices.

Public data shows that China's refined indium production accounts for about 70% of the world's total, but the purification process of high-purity indium is still a bottleneck. 7N grade products (purity 99.99999%) used for indium phosphide substrates need to undergo multiple rounds of impurity removal and zone melting purification. Only a few domestic enterprises such as Zhuzhou Keneng and Beijing Tongmei have stable mass production and shipping capabilities.

In addition, in February 2025, the Ministry of Commerce and the General Administration of Customs issued Announcement No. 10, which simultaneously included organic metal sources and related production technologies such as indium phosphide, trimethylindium, and triethylene indium in export control.

In this context, data from Shanghai Nonferrous Metals Network shows that the spot price of refined indium in China has risen from about 2960 yuan/kg at the beginning of the year to 5400-5500 yuan/kg in mid August, an increase of over 80%. The quotation for 7N grade high-purity indium has exceeded 6000 yuan/kg.

Zhou Zihao told reporters that after the export control, some overseas optical chips and epitaxial production capacity have accelerated their transfer to China, and domestic optical chip companies are also rapidly expanding.

In June of this year, Dongshan Precision (002384. SZ) announced that its subsidiary Sols Optoelectronics will lay out an expansion project of optical chips and high-speed optical modules in Changzhou, with a total investment of 1.2 billion US dollars. Sols has vertical integration capabilities from optical chip design and manufacturing to optical module assembly. On August 11th, domestic CW laser chip leader Yuanjie Technology (688498. SH) announced plans to invest 4.268 billion yuan to build a semiconductor technology industrial park and expand its high-end laser chip production line.

The expansion of these optical chip production lines has directly driven the demand for purchasing indium phosphide substrates, objectively opening up a new window of opportunity for domestic substrate manufacturers. However, the quality requirements and certification standards for substrates in these production lines are consistent with those overseas, and it remains to be seen whether domestic manufacturers can meet this demand.

Domestic Leading Brand

In the global battle for indium phosphide, China's situation is quite unique: upstream indium resources are globally leading, with refined indium production accounting for about 70% of the world's total; However, the total shipment volume of downstream substrate products from all domestic manufacturers accounts for only about 10% of the global total; The largest production capacity and output is Xinyao Semiconductor, a subsidiary of Yunnan Germanium Industry. By 2025, it will actually produce 100100 indium phosphide chips (including various sizes, not converted to caliber).

The traditional main business of Yunnan germanium industry is germanium mining and processing of germanium series products. Germanium is mainly used in the fields of fiber preform, infrared optics, and aerospace photovoltaics. Generally speaking, upstream material companies that achieve deep processing of raw materials will stop and rarely actively extend to downstream semiconductor materials, as the two involve completely different process systems and customer certification systems.

According to public information, in December 2020, Huawei's Hubble Technology invested in Xinyao Semiconductor, holding 23.91% of the shares and becoming the second largest shareholder. Huawei HiSilicon began purchasing indium phosphide substrates from Xinyao at the same time. Huawei HiSilicon not only brought orders, but also provided a lot of technical support to Xinyao in improving substrate quality. Xinyao, Sumitomo, and Tongmei's substrates were compared and tested in their own epitaxial furnaces, and quality differences were reported batch by batch.

The reporter learned that Huawei HiSilicon will be Xinyao's largest customer by 2025, accounting for about 40% of its indium phosphide supply. Xinyao's other clients include the 13th Research Institute of China Electronics Technology Group, Huaxing Laser, Qianlu Optoelectronics and other epitaxial manufacturers and device manufacturers.

Xinyao purchases high-purity red phosphorus and high-purity indium required for the production of indium phosphide chips from external sources, with domestic manufacturers being the main suppliers. Xinyao's role is equivalent to processing incoming materials, turning indium and phosphorus into substrate chips that downstream laser manufacturers can use. The qualification rate during the conversion process directly determines its cost and competitiveness.

In 2025, the gross profit margin of Yunnan Germanium Industry's traditional main business material grade germanium products will decrease from 29% in the previous year to 4.32%. The price increase of raw materials such as indium and germanium has almost eaten up all the profit margin of this product line, but the simultaneous increase in volume and price of indium phosphide business has provided the company with a new source of profit.

In April 2026, the company announced an investment of 189 million yuan to expand its indium phosphide production capacity, adding an annual output of 300000 pieces (equivalent to 4 inches, including 6000 pieces of 6 inches), with a construction period of 18 months. After completion, the total production capacity will reach three times the current scale.

On July 24th, Xinyao signed a supply agreement for indium phosphide chips worth 570 million to 855 million yuan with a domestic customer who has been cooperating for many years. The amount is equivalent to 50% to 80% of Yunnan Germanium Industry's annual revenue in 2025, and the performance period is from August this year to December 2027. This is the largest order since the establishment of Xinyao, but the contract stipulates that the price will fluctuate by 20% based on the market price. That is to say, the protocol locks in quantity, while the price follows the market.

A person close to Xinyao in the industry chain told reporters that the growth of indium phosphide single crystals is the most difficult link to control in the entire production line, and also the biggest gap between domestic manufacturers and overseas leaders. Unlike the automated crystal pulling of silicon wafers, the growth of indium phosphide single crystals requires high temperature and high pressure conditions. Workers manually load polycrystalline materials into quartz tubes, seal them, vacuum them, and then fill them with high-pressure nitrogen gas. After that, the crystals slowly crystallize. The proficiency of operators directly affects the yield of each furnace.

In addition, the indium phosphide single crystal furnace is not a standardized equipment, and the thermal field structure and process parameters are designed by various manufacturers themselves, commissioned to third-party processing and manufacturing. The crystallized single crystal also needs to go through multiple subsequent processes such as cutting, thinning, grinding, polishing, etc., and each step will result in losses. Overall, the production qualification rate of indium phosphide substrates is much lower than that of silicon wafers, and the larger the size, the higher the difficulty. This is determined by the brittleness and growth characteristics of indium phosphide crystals themselves, and global manufacturers are facing the same challenge. If you can't make a 4-inch product, you can revert to a 3-inch product, and if you can't make a 3-inch product, you can revert to a 2-inch product. Currently, supply and demand are tight, and even small-sized products are sold, but they will definitely be eliminated in two or three years. "The above-mentioned person said that Xinyao's products are mainly 3 to 4 inches, which is also a common situation in the domestic indium phosphide substrate industry. The area of a 6-inch substrate is approximately 2.25 times that of a 4-inch substrate, which can significantly reduce chip processing costs while maintaining yield and process compatibility. However, the larger the crystal diameter, the narrower the control window for temperature gradients, stress, and dislocations during growth, making it more difficult to control warpage and surface defects.

Globally, AXT and Coherent are advancing 6-inch mass production through a three-year joint development and a $22.2885 million advance payment. The 6-inch indium phosphide substrate worldwide is still in the transition stage from trial production to mass production.

The reporter also learned that Xinyao is one of the few enterprises in China that can supply optical communication grade indium phosphide substrates in bulk, and is currently expanding its production line and frontline operation team on a large scale.

The above-mentioned industry chain insiders believe that in the past few years, with the deep support of Huawei HiSilicon, Xinyao has achieved the fastest improvement in substrate quality among domestic manufacturers. But from the perspective of industry rules, it takes several years to bridge the gap between design capacity and actual output, and equipment debugging, process improvement, and personnel training all require a process.

Zhou Zihao said that the most direct indicator to measure the process level of substrate manufacturers is the yield rate, which is the proportion of the weight of the finished product to the input raw materials in one go. In this indicator, Xinyao is currently at 7% to 8%, while Tongmei can achieve around 25%, with a gap of nearly three times. In terms of comprehensive recovery rate, Xinyao is about 30% and Tongmei is about 50%. Taking Xinyao as an example, about 70% of raw materials become waste during the production process.

The design capacity and actual output are not the same thing. It may take four to five years to truly reach full production, "said Zhou Zihao. The equipment is customized and improved in batches, and each batch is fine tuned based on the problems exposed in the previous batch. The actual capacity utilization rate in the first year is usually only about 70%, and the new production line in the processing section is expected to be put into trial production around December this year.

At the investor exchange event on July 16th, the management of Yunnan Germanium Industry stated that the company plans to produce 180000 pieces (including various sizes) of indium phosphide chips in 2026.

Even if the output can catch up, the substrate still needs to go through processes such as sample delivery, device verification, reliability testing, and small-scale import before entering the laser production line. The pure product certification process usually takes nearly a year.

Zhou Zihao said that downstream customers will put substrates from different suppliers into an epitaxial equipment for comparison. The microscopic quality differences of the substrates may not be detected in the factory inspection stage, but they will be completely exposed in the epitaxial stage.

The management of Yunnan Germanium Industry stated during the aforementioned investor exchange event that the company's indium phosphide product certification has basically covered well-known downstream manufacturers in China, and is currently actively expanding overseas customers.

Quality catch-up

On August 7th, Yunnan Germanium Industry announced abnormal fluctuations in stock trading, showing that the company's static P/E ratio has reached 2681 times, while the average level of its non-ferrous metal industry is 26 times; The revenue of the compound semiconductor materials business (including gallium arsenide and indium phosphide, not split) in 2025 is approximately 138 million yuan, accounting for 12.93% of the company's total revenue and contributing 14.29% to gross profit.

In addition, the performance forecast of Yunnan Germanium Industry shows that its expected net profit attributable to shareholders in the first half of 2026 is 55 million to 80 million yuan, a year-on-year increase of 148% to 261%. The improvement in performance mainly comes from the increase in sales volume and average price of indium phosphide products; The gross profit margin of traditional main material grade germanium products is still only 4.32%, and the drag on overall profits has not yet been lifted.

Among domestic indium phosphide substrate enterprises, Yunnan Germanium Industry started earlier, has the largest production capacity, and is also one of the few manufacturers to receive long-term orders from top customers. From Hubble's investment in 2020 to signing a supply order worth nearly 900 million yuan this year, although there is still a gap in yield and customer certification compared to overseas leaders, its subsidiary Xinyao Semiconductor has taken 5 years to advance its indium phosphide business to the stage of bulk supply.

Public information shows that in addition to Yunnan Germanium Industry, there are also multiple domestic enterprises laying out in the indium phosphide industry chain.

In the substrate process, Pioneer Technology can provide 2-6 inch indium phosphide substrate products; Mingga Semiconductor focuses on upstream polycrystalline synthesis, with an annual production capacity of nearly 30 tons of indium phosphide according to the company's official website information. In the epitaxial process, Sanan Optoelectronics (600703. SH) is one of the companies with a deep layout in China. According to the company's response on the investor interaction platform in April this year, its indium phosphide epitaxial production capacity has been expanded to nearly 6000 pieces per month.

Overall, domestic indium phosphide substrate products are still dominated by 3-4 inch mid to low end specifications, with a generational gap compared to Sumitomo and Tongmei in high-end substrate products, but the pace of catching up is accelerating.

At the same time, there have been more companies in the A-share market pursuing the concept of indium phosphide across borders. For example, Xingye Technology (002674. SZ), which specializes in natural leather processing, announced in June this year that it plans to acquire the indium phosphide business and closed six consecutive daily limit ups. In July, it officially completed the acquisition of Qingdao Li'ang Jingdian indium phosphide business and assets for over 50 million yuan. According to the announcement, the total assets of the subject matter are 17.02 million yuan, with a net asset of 6.37 million yuan. The revenue for the first five months of 2026 was 2.34 million yuan, and the net profit was negative 1.44 million yuan. The number of orders on hand does not exceed 2 million yuan, and there has been no profit in the past three years. The products mainly consist of 2-3 inch substrates and do not involve optical module customers.

Since early August, multiple listed companies with the concept of indium phosphide have issued intensive announcements or risk warnings regarding abnormal stock trading fluctuations, pointing to similar situations: either the proportion of related business revenue is extremely small, or actual revenue has not yet been generated. Some companies even explicitly stated in their announcements that "their main business does not include indium phosphide substrate business".

Chen Qiqi, an analyst of Shanghai Steel Federation, said in an interview with the media that about 60% of the increase in indium price within the year included a certain foam component, and AI's actual demand for indium was about 5%. According to LightCounting data, the market size for indium phosphide optical chips is expected to reach approximately 6.9 billion US dollars by 2031. Overall, indium phosphide remains a relatively small category in the entire semiconductor material landscape.

A long-term investor who has been paying attention to the domestic semiconductor materials field said that the real turning point of the supply and demand balance in the indium phosphide industry chain is around 2027, when overseas new production capacity will begin to be released in a concentrated manner, and domestic 6-inch substrates may also enter the stage of small batch shipments. Prior to this, only manufacturers who can obtain certification from top downstream customers and maintain a stable yield rate of 4-inch and above products at the delivery level can truly benefit from this cycle.

According to the reporter's understanding, domestic manufacturers have not yet occupied a significant share in Nvidia's indium phosphide substrate supply system. Export controls objectively create a window period for domestic enterprises, but a complete round of customer certification takes one year.

In Zhou Zihao's view, the core task of domestic enterprises in the next two to three years is to achieve substrate quality that can pass the certification of top customers. "Resources are in our hands, and the market is also shifting to China. The key is whether the quality can keep up.

(Author Zheng Chenye)