In 2026, the global AI computing race has entered a white-hot stage. From NVIDIA’s ambitious blueprint for million-GPU clusters to the accelerated iteration of various AI large models, the foundation that supports it all — data centers — is undergoing an unprecedented revolution in network architecture.
However, while most attention remains focused on GPU performance, one critical link in the supply chain is quietly becoming a new bottleneck for the expansion of AI clusters: optical modules, as well as the core substrate materials that enable their performance leap.
Today, starting with optical modules, we explore the “hardcore support” hidden behind the brilliance of this computing revolution.
From Electrical to Optical: AI Clusters Call for High-Speed Optical Modules
As Moore’s Law gradually slows down, horizontal scaling of AI clusters, or scale-out, has become inevitable. Traditional electrical interconnects are reaching their physical limits in bandwidth, latency, and power consumption. As a result, the shift from copper to optics has become one of the defining trends in data center construction in 2026.
Recently, NVIDIA has further advanced its deployment around Spectrum-X Ethernet and silicon photonics, accelerating co-packaged optics, or CPO, from concept validation toward industrial implementation.
Compared with traditional pluggable solutions, CPO can reduce the power consumption of 800G ports from 14–16W to 5.2–5.6W, representing a reduction of 60%–68%. In the era of AI clusters with hundreds of thousands or even millions of GPUs, this is not only an upgrade in interconnect technology, but also directly related to lower energy consumption, improved thermal management, and reduced total cost of ownership for data centers.
Market demand has been ignited accordingly. According to a research report by Huatai Securities, AI computing demand is driving the accelerated ramp-up of high-speed optical modules. Under a neutral forecast scenario, the market capacity driven by 800G/1.6T optical module expansion from 2026 to 2028 is expected to reach RMB 33.2 billion, RMB 52.5 billion, and RMB 64.0 billion, respectively.
LightCounting has also made a bold prediction: 2026 will be the first historic year in which transceivers using silicon photonic modulators account for more than 50% of total transceiver sales.
Yet as this wave continues to surge forward, a huge “invisible gap” is beginning to emerge — do we really have enough materials to manufacture all these optical modules?
The Core Bottleneck: The Substrate Challenge Behind Optical Modules
One easily overlooked fact is that whether for today’s 1.6T optical modules or future 3.2T solutions, the core of their light-emitting capability relies heavily on two major material systems: indium phosphide, or InP, and gallium arsenide, or GaAs.
Indium Phosphide: The “Foundation of Light” in the AI Computing Era
As a core substrate material for high-end 800G and 1.6T optical modules, indium phosphide has become an irreplaceable physical foundation for today’s high-speed optical chips, thanks to its ultra-high electron mobility and direct bandgap that precisely matches communication wavelength windows.
However, manufacturing InP is as difficult as “dancing on the tip of a needle.” Its production requires extreme environmental control under high temperature and high pressure. Equipment delivery cycles are long, yield ramp-up is difficult, and customer qualification periods are extended. It is a typical industry characterized by long cycles, heavy capital investment, and high technical barriers.
Driven by the explosive growth of AI computing demand, this high-threshold and slow-to-expand supply structure is evolving into a severe supply challenge.
According to Yole’s forecast, global demand for indium phosphide is expected to soar to 2.6 million to 3.0 million wafers in 2026, while global effective capacity will only increase to around 750,000 wafers, leaving a supply-demand gap of more than 70%.
Prices have also surged rapidly. The price of 2-inch optical communication-grade InP substrates has risen from USD 800 per wafer to USD 2,500 per wafer in just over a year, an increase of nearly two times. The price of 6-inch high-end substrates has climbed from USD 1,400 to above USD 5,000, representing an increase of more than 250%.
This “scarcity despite high prices” situation is expected to continue until after 2027, when new overseas capacity and domestic 6-inch substrate production are expected to gradually ease the imbalance.
Facing this supply-demand challenge, a leading domestic indium phosphide substrate manufacturer is accelerating the construction of its high-quality InP single-crystal wafer project. The company plans to significantly expand capacity on its existing foundation, aiming to occupy a more important position in the global high-speed optical communication materials supply chain.
Gallium Arsenide: A Versatile Material for Short-Reach Transmission
Unlike indium phosphide, which mainly targets long-distance high-speed transmission, gallium arsenide plays an irreplaceable role in data center short-reach interconnects, 5G RF front ends, LiDAR, and other applications due to its high electron mobility and efficient light-emitting properties.
In particular, in the field of VCSELs, or vertical-cavity surface-emitting lasers, GaAs is the indispensable substrate material. It supports the practical implementation of short-reach optical interconnect solutions such as near-packaged optics, or NPO.
Compared with the severe supply-demand gap facing indium phosphide, the global gallium arsenide market is also entering a growth channel driven by AI computing demand. It is expected to grow from USD 1.33 billion in 2025 to USD 1.48 billion in 2026.
Since the second quarter of 2026, driven by the multiple-fold growth in AI server demand for optical modules, prices across the GaAs industry chain — from substrates and epitaxial wafers to foundry services — have risen broadly, with some power amplifier products seeing price increases of more than 10%.
From Material Shortage to Manufacturing Capability: Siplus Semiconductor’s Precision Answer
In the face of this “material shortage,” global attention is not only focused on how to accelerate capacity expansion, but also on a more fundamental question:
How can these expensive substrate “raw stones” be precisely processed into qualified chip wafers?
This is exactly the field in which Siplus Semiconductor is deeply engaged.
We do not produce substrates, but we provide reliable precision processing solutions for all types of substrates.
Key Challenges of Indium Phosphide and Gallium Arsenide
As typical “soft and brittle” materials, InP and GaAs are extremely sensitive to processing damage during back-end thinning and polishing.
Any tiny scratch or stress concentration may directly lead to a yield collapse in subsequent lithography or epitaxy processes. How to achieve low-damage, high-flatness, and high-consistency processing on large-size, high-value substrates is a key factor in moving domestic substitution from “0 to 1” toward large-scale mass production.
Siplus Semiconductor’s Solution
Siplus Semiconductor provides full-process precision processing equipment covering slicing, grinding, and polishing for the rapidly growing optical chip industry.
Whether it is high-flatness control for indium phosphide wafers or efficient low-damage thinning for gallium arsenide substrates, Siplus equipment can support stable delivery at a mass-production level.
In the field of compound semiconductors, through innovative process coordination solutions, Siplus has helped multiple leading domestic substrate manufacturers overcome mass-production processing challenges for large-size products.
Notably, Siplus precision processing equipment has successfully entered the supply system of a leading domestic indium phosphide substrate manufacturer. In this customer’s widely watched InP substrate expansion project, Siplus precision processing equipment provides stable support for the scale production of high-end substrates.
Today’s semiconductor landscape is not only a celebration of computing power, but also a contest of materials.
As the digital wave surges forward, we have come to realize that long-term competitiveness is not only held in advanced 5nm lithography machines, but also embedded in each carefully processed indium phosphide and gallium arsenide wafer.
And Siplus Semiconductor is using “precision” as its pen, writing its own chapter on the trillion-scale foundation of the semiconductor industry.