Sumitomo Chemical has taken another step in supplying materials for AI optical communications to outside customers. On August 31, 2026, the company announced that it had established a mass-production system for 4-inch InP (indium phosphide) epitaxial wafers (epiwafers) at its Ibaraki Plant in Hitachi, Ibaraki Prefecture, and had begun sales. In China, Wuhan Tianyuan has acquired an InP substrate company, and JX Metals is planning a large investment in substrates. Both involve InP, but making the crystal substrate and growing the thin films for optical semiconductors on top of it are different processes. That difference shows how the start of sales may affect the optical communications supply chain.
Substrates and epiwafers: the manufacturing steps in between
In an exhibition notice Sumitomo Chemical published in January, InP epitaxy technology was presented as a research and development subject. The August announcement moves on to mass production and sales of 4-inch products, and states a goal of reaching annual sales on the order of 10 billion yen for the InP epiwafer product line by the mid-2030s. This growth target applies to this product line, not to the compound semiconductor business as a whole.
A semiconductor substrate is the base of the crystal from which devices are made. On InP, thin films such as InP and InGaAs are stacked to build the layer structure needed for light emission and detection. This thin-film growth is called epitaxial growth, and the product is the epiwafer. Making the substrate and growing layers on its surface are separate processes. After epitaxy come further steps: processing into chips, testing, and assembly into optical components.
As JX Metals' technical materials indicate, the substrate's surface condition, crystal orientation, impurities and defects all affect later epitaxial growth. Sumitomo Chemical explains that during high-temperature growth, composition and surface condition must be controlled precisely, and the process must be managed all the way to the source gases. Using technology from more than 25 years of mass-producing gallium arsenide (GaAs) and gallium nitride (GaN) epiwafers, it says it has achieved uniform crystals and film thickness, along with reproducibility in volume production. To sell externally on a continuing basis, the company must be able to produce the same quality of thin film repeatedly.
Wafer diameter of 4 inches is one condition for product selection, but productivity depends on whether the required layer structure can be made stably and whether the output is good product in downstream processes as well. Sumitomo Chemical has not disclosed monthly output, yield figures or specific layer specifications, so supply scale and cost differences from competitors cannot yet be measured. What can be confirmed is that optical semiconductor makers considering procurement now have one more supplier of InP wafers.
Why silicon photonics needs InP
Silicon photonics uses circuits on a silicon substrate to guide, split and control light. Silicon, however, is not a material that emits light efficiently. This led to the idea of combining InP-based semiconductors, which handle light emission, with silicon waveguides, which carry the light. The hybrid laser that Intel and the University of California, Santa Barbara announced in 2006 is an early example demonstrating that principle.
Still, there is more than one way to place the InP light source. Intel now publicly describes optical I/O with lasers built into the photonic integrated circuit, explaining a configuration that does not require an external light source, along with wafer-level testing and laser burn-in, which runs devices under set conditions to detect early failures. The design lets quality be checked before components are finely divided and assembled.
In CPO (Co-Packaged Optics), which places the optical circuit close to the electronic circuit, both are mounted in the same package. This is a separate design choice from whether the light source is built in or external. In the configuration NVIDIA has described, external laser light sources with independently managed temperature and replaceability are used, with eight lasers driving 32 transmit lanes. A light source that can be replaced on its own also helps with maintenance on the switch side. NVIDIA says sharing light sources can cut the number of lasers to one-quarter compared with a conventional configuration. The figures in the company's technical explanation should be read within the scope of that design example.
This light-source sharing also matters when considering material demand. The number of lasers does not necessarily rise in step with the number of optical communication ports, and InP wafer needs do not rise by that number either. Some configurations share light sources, and integration methods differ. Chip area and yield also matter. Estimating material sales from port counts alone is premature.
Funding for mass production after a 7.5 million yuan acquisition
In August, Wuhan Tianyuan acquired 60% of Zhongxun Semiconductor and entered the semiconductor substrate business. The consideration was 7.5 million yuan, and the company set out a plan to invest an additional 67 million yuan or more in equipment, facilities and working capital after the acquisition. The total comes to more than about 74.5 million yuan. After gaining control through the acquisition, it plans to put in the funds needed to install volume-production equipment and run the business.
Zhongxun Semiconductor was founded in 2022 and holds 29 valid patents. According to Wuhan Tianyuan, it has the technology needed to produce 2- to 4-inch single-crystal InP and other products. But as of August 30, it had no sales for 2026. Further investment will be decided depending on progress in customer qualification. There is a gap between being able to make substrates and completing qualification and generating continuing sales.
While its main business is environmental and green energy, the company is expanding into AI computing infrastructure and cited demand for 800G and 1.6T optical modules. Its policy is to grow the new field using operating cash flow from its main business. It is a management decision to direct funds earned in environmental businesses toward nurturing semiconductor materials, but whether those funds lead to repeat purchases by customers will depend on future verification and production progress.
Stage differences in supply, seen alongside JX Metals' large investment
Sumitomo Chemical's start of sales, the capital injection into Zhongxun Semiconductor and JX Metals' capital investment plan differ in how and when they affect supply volume.
| Company | Process | Confirmed status | What to watch |
|---|---|---|---|
| Sumitomo Chemical | InP epiwafers | Mass-production system for 4-inch products established; sales started | Customers, volumes and product mix undisclosed |
| Zhongxun Semiconductor / Wuhan Tianyuan | InP substrates | 60% acquired; additional investment planned. No 2026 sales as of Aug. 30 | Customer qualification and investment decisions |
| JX Metals | InP substrates | Investment policy of up to 120 billion yen over four years | Details of capacity expansion and actual operation |
In June, JX Metals announced a policy of expanding substrate production capacity to 7 to 10 times the fiscal 2025 level, including previously announced increases. The investment is up to 120 billion yen over the next four years, and the company has said it received requests from customers for production increases beyond its earlier expectations and intends to request price revisions as well. The company's announcement is a plan to increase substrate supply capacity, and cannot be compared on the same metrics as Sumitomo Chemical's epiwafer sales or Wuhan Tianyuan's acquisition.
JX Metals' plan affects substrate supply, which comes before epitaxial growth. Sumitomo Chemical's external sales cover the process of building the required layer structure on a substrate. Even if more substrates are made, the materials reaching optical semiconductor makers will not increase if epi processing cannot keep up. The reverse is also true. Viewing these as moves to expand capacity process by process shows which part of the supply chain each company's investment supports. However, no business relationships among these companies have been confirmed.
Conditions for turning optical communications growth into good-quality supply
Even if high-speed optical communications for AI push the materials market, the path to a finished product is not single. Substrate flatness and impurities affect epitaxial growth, the epi layers affect device performance and yield, and a chip becomes a usable component only after packaging and testing. With built-in lasers, wafer-level testing can be part of the design; with external light sources, temperature management and replaceability carry a different value.
For customers, the adoption decision depends on whether they can receive materials suited to their devices and packaging on a stable basis. The next things to watch for Sumitomo Chemical are supply volumes and adoption record after the start of sales; for new Chinese entrants, progress from customer qualification to continuing sales. If quality and supply volume connect from substrates through epitaxy to light-source packaging, plans to expand optical communications in AI data centers can become components that can actually be procured.