Author: Peng Cheng
As the demand for 800G/1.6T optical modules surges amid the wave of AI computing infrastructure buildout, indium phosphide (InP), a compound semiconductor used to manufacture the core substrates of optical chips, is transitioning from a niche material in specialized fields to a strategic material for the entire digital economy. InP is currently the only semiconductor that can simultaneously meet four major conditions: a direct bandgap (high electro-optical conversion efficiency), precise wavelength matching (1310/1550nm, the golden window with the lowest fiber loss), ultra-high electron mobility (supporting signals above 100GHz), and natural lattice matching with epitaxial materials (allowing the integration of lasers, modulators, and detectors on the same substrate).
This makes InP difficult to replace in optical communications. This compound semiconductor, once considered niche, is now moving from behind the scenes to the forefront. From prices doubling to capacity surging, from NVIDIA prepaying billions of dollars to lock in capacity to domestic enterprises breaking through the 6-inch full-chain domestic production, the InP industry is accelerating its capacity expansion.
01. Supply Falls Short of Demand, Prices Surge InP is widely used in DFB lasers, EML lasers, and photodetectors, serving as a critical raw material for 800G/1.6T and even next-generation 3.2T optical modules. Data shows that by 2026, the global demand for InP substrates is expected to reach 2.6 million to 3 million pieces, while the effective compliant capacity is only about 750,000 pieces, resulting in a supply-demand gap exceeding 70%.
This imbalance is directly reflected in prices. As of April 2026, the price of 2-inch telecom-grade InP substrates has skyrocketed from $800 per piece in early 2025 to $2,300-$2,500 per piece, an increase of nearly 200%; the price of 6-inch high-end substrates has surged from $1,400 per piece to $5,000 per piece, an increase of over 250%. The fundamental reason for the price surge is the long capacity expansion cycle in the industrial chain. From the construction of crystal growth furnaces to customer certification, the entire expansion cycle takes 18 to 24 months. Coupled with the reliance on overseas imports for core equipment, capacity release falls far behind the steep rise in the demand curve.
In addition to demand, the price increase of InP substrates is also related to upstream raw materials. The core raw material of InP is the rare metal indium. The latest data from China Silver Net (as of July 6) shows that the price of metal indium has reached 5,560 RMB/kg, doubling since early 2025 and hitting a nearly ten-year high. Indium rarely forms independent deposits in nature; the vast majority of indium is extracted as a byproduct during the smelting of other metals, making its supply elasticity naturally limited. Shenwan Hongyuan estimates that in 2027, the InP sector will drive a 6.77% increase in indium demand. While the proportion seems low, it is enough to trigger severe price fluctuations. The cost curve of InP substrates has been firmly locked at a high level, leaving limited room for price declines.
More critically, the global InP supply chain is beginning to fracture. In January 2026, China's Ministry of Commerce issued an announcement imposing a comprehensive ban on the export of dual-use items (including InP, indium, gallium, and germanium) to Japanese military users and for military end-uses, while civilian exports required strict licensing and end-user reviews. Market feedback indicates that the rejection rate for applications by Japanese and American enterprises for Chinese-made InP substrates has exceeded 80%. Meanwhile, the U.S. Department of Commerce launched anti-dumping and countervailing duty investigations into active anode materials from China as early as January 2025. Although separate tariffs have not yet been directly imposed on InP, the compounding effect of export control policies is evident. Under the framework of the Critical Raw Materials Act, the European Union has introduced amendments to reduce over-reliance on a single country (especially China) and incorporate recycled content requirements into mandatory standards. This means that in the future, using Chinese-produced indium will not only face higher compliance costs and export control uncertainties but may also be excluded from some high-end supply chains. All these factors are impacting the global supply and capacity expansion pace of InP.
02. Downstream Giants Begin to Lock in Capacity As InP supply becomes the bottleneck for the entire AI computing infrastructure, downstream giants are breaking traditional supply chain boundaries and directly "infusing" upstream operations. As early as March 2026, NVIDIA announced it would invest $2 billion each in Coherent and another photonics manufacturer, accompanied by long-term large-scale procurement agreements to lock in stable capacity for InP optical chips over the next few years. Lumentum's CEO disclosed that EML laser production has increased eightfold over the past three years, yet shipments remain 25%–30% below market demand. In June 2026, Jensen Huang personally attended the groundbreaking ceremony for Coherent's industry-first 6-inch InP wafer fab expansion project. NVIDIA's intention is very clear: in the AI arms race, upstream InP capacity has become a hard constraint for optical interconnects; without locking in capacity, it cannot guarantee the delivery of its own AI servers. This "direct investment by giants" model is reshaping traditional supply chain relationships, transforming InP from a general material into a strategically bound resource. It also gives downstream players the confidence for large-scale capacity expansion.
On the domestic front, Huawei's Hubble Technology invested in Xinyao Semiconductor, a controlling subsidiary of Yunnan Germanium Co., Ltd., in 2020, acquiring a 23.91% stake to become the second-largest shareholder. This investment not only provided financial support but also included an agreement that Xinyao Semiconductor would prioritize the supply of gallium arsenide (GaAs) and indium phosphide (InP) substrate materials to Huawei's affiliates. The cooperation focuses on core materials such as InP substrates. Xinyao Semiconductor's products have passed Huawei HiSilicon's testing and verification and are applied in fields such as 5G and data centers. In 2025, Huawei locked in an order for 80,000 InP wafers from Xinyao Semiconductor (accounting for 53% of its capacity), with an advance payment ratio of 40% (industry practice is <20%). Beyond capital backing, the deal deepened the mutual interest binding through the agreed priority supply rights.
03. Global Enterprises Begin Capacity Expansion Faced with a historic gap, major global manufacturers have launched aggressive capacity expansion plans.
Overseas, traditional giants are accelerating their deployment. AXT in the U.S. plans to expand by 200 4-inch single-crystal furnaces, with a capacity target of 50,000 pieces per month in 2026, and plans to quadruple its total capacity by the end of 2027; Sumitomo Electric plans to invest about 18 billion yen to increase InP substrate capacity to 3.1 times that of fiscal 2024 by the end of fiscal 2028; Lumentum expects that by the end of fiscal 2026, EML capacity will grow by over 50% compared to 2025, following the company's previous push of about a 40% InP capacity expansion plan; Coherent is expanding its 6-inch InP wafer capacity in Sherman, Texas, U.S., and expects to achieve its target of doubling capacity by the end of 2026 a quarter ahead of schedule, with plans to double it again by the end of 2027.
Domestic enterprises are expanding capacity just as rapidly. Yunnan Germanium Co., Ltd. (through its subsidiary Xinyao Semiconductor) is the absolute leader, with an existing capacity of 150,000 pieces per year (converted to 4-inch equivalent). In April 2026, it launched an expansion project with a total investment of 189 million RMB, adding a production line with an annual output of 300,000 pieces (equivalent to 4-inch, including 6,000 pieces of 6-inch), bringing the final total capacity to 450,000 pieces per year. GRINM Advanced Materials currently has an InP capacity of 150,000 pieces per year (covering all specifications from 2 to 6 inches). Its 6-inch products have completed technical breakthroughs and achieved small-batch supply, with yield steadily improving. The planned addition of 250,000 pieces per year in InP capacity is expected to reach full production in the second half of 2027, with a total capacity target of 400,000 pieces per year. Pioneer Microelectronics plans to invest 1.7 billion RMB in fixed assets to upgrade and expand production lines using existing sites, introducing core production equipment such as high-end crystal growth, precision polishing, and defect detection, focusing on 4-6 inch high-end GaAs and InP single-crystal substrate products. Upon completion, the project will form an annual production capacity of 3 million GaAs substrates and 3 million InP substrates, totaling 6 million high-end semiconductor substrates per year. The construction cycle is from August 2026 to August 2029. Guangdong Pingrui Jingxin's semiconductor technology industrial park has a total investment of 1.1 billion RMB. Upon completion, it is expected to form an annual production capacity of 300,000 InP single-crystal substrate pieces, with total annual sales revenue exceeding 600 million RMB. In addition, Sanan Optoelectronics' Wuhan base, featuring the country's first 6-inch InP epitaxy mass production line, has expanded its core epitaxy process capacity to 6,000 pieces per month. Xianrui Technology has launched an annual expansion project for 40 tons of InP crystals. This production line received environmental impact assessment approval on March 18, 2026 (Qinggao Approval Huan [2026] No. 3), and is only the final step away from implementation and production. Dingtai Xinyuan is actively expanding its InP substrate capacity, though the expansion and full production timeline remains uncertain. However, capacity expansion is not an overnight process. The long construction cycle of production lines, the 12-24 month lead time for core equipment such as MOCVD, and the generally 1-2 year customer certification cycle determine that the tight supply-demand situation in the industry will last at least until 2028.
The boom has also attracted cross-industry players. On June 21, 2026, Xingye Technology, which primarily deals in natural cowhide leather, announced a plan to acquire the InP substrate and semiconductor electronic materials business of Qingdao Lion Crystal Electronics for 55 million RMB in cash. The acquisition covers all assets, business teams, patents, trademarks, and proprietary technologies. Suqian Liansheng announced in June 2026 its cross-industry entry into the InP substrate track, planning to establish a joint venture. The first phase will invest 100 million RMB to build an annual production line of 120,000 pieces of 4-6 inch products, expanding to 400,000 pieces per year in the second phase.
04. Breakthroughs in Domestic InP Technology Beyond capacity expansion, the systematic breakthroughs in domestic InP technology are equally noteworthy.
Full-Chain Domestication of 6-Inch Production is the most milestone achievement. In August 2025, Jiufengshan Laboratory, in collaboration with Xinyao Semiconductor, relied on domestic MOCVD equipment and InP substrate technology to overcome the challenge of large-size epitaxial uniformity control. They successfully developed the epitaxial growth process for 6-inch InP-based PIN structure photodetectors and FP structure lasers for the first time, with key performance indicators reaching international leading levels. This achievement also marks the first time in China that a collaborative domestic application from core equipment to key materials has been realized in the preparation of large-size InP materials, providing critical support for the industrialization of optoelectronic devices.
In terms of Crystal Growth Process Innovation, domestic enterprises are upgrading from the traditional LEC (Liquid Encapsulated Czochralski) method to the VGF (Vertical Gradient Freeze) method. In the past, the mainstream preparation method for InP in China faced significant process difficulties, high dislocation density, and a tendency to produce twin crystals. Huaxin Jingdian uses the VGF method to produce InP single crystals, resulting in higher product quality and stability. Pioneer Microelectronics' independently developed VGF InP single-crystal growth technology, combined with key technologies for low-damage wafer polishing and ultra-clean surface cleaning, has successfully produced 6-inch InP substrates with low dislocation density, stable electrical performance, high flatness, and clean surfaces.
Heterogeneous Integration is also advancing simultaneously. The hybrid/heterogeneous integration of InP and silicon photonics (SiPh) is the mainstream technical direction in the optical communication field. InP is responsible for providing light sources (lasers, amplifiers), while silicon handles passive waveguides and electrical interconnects. The two achieve optoelectronic integration through wafer bonding, micro-transfer printing, or 3D hybrid integration. Commercial optical transceiver modules from Intel and Cisco adopt heterogeneous integration technology; China's Jiufengshan Laboratory and Sun Yat-sen University have also successfully achieved heterogeneous integration of InP lasers on silicon wafers, proving the feasibility of large-scale mass production.
05. Conclusion Looking back from the middle of 2026, the surge in InP is by no means a simple cyclical shortage, but a violent collision between the AI computing revolution and the semiconductor material supply chain. And just in early July, He Tingbo of Huawei updated and released Version 2 of "Time Scaling Theory for Multi-Layer Electronic Systems." He’s Law 2.0 defines τ (time constant) as a hierarchical composite variable spanning four levels: devices, circuits, chips, and systems. Its value is jointly determined by underlying hardware parameters, local architecture, and communication overhead. If Logic Folding is about taking shortcuts for signals at the circuit level and compressing wiring delays through 3D stacking at the chip level, then τ optimization at the system level points to a more harsh fact: in large AI clusters, over 80% of energy is consumed in data movement; over 70% of system costs are allocated to data storage. The direct implication is that reducing the transmission time of data between chips, between racks, and within packages is no less important than reducing the time consumed by computation itself.
This is precisely where the strategic significance of InP lies. Huawei's deployment of the Hi-ONE high-density optical interconnect node engine and the unified memory semantic bus (Lingqu Bus) at the system level aims to push cross-cabinet optical interconnect bandwidth to a single-channel 8 Tb/s and compress the SerDes transmission distance from 100 cm to 5 cm. The realization of these system-level τ compressions is entirely built upon the foundation of InP optical chips.