AI Demand Triggers Global Capacity Race for Indium Phosphide Semiconductors
A severe supply-demand gap in critical optical materials is driving billions in investment as AI data centers shift toward advanced optical interconnects.
The explosive growth of AI data centers has ignited a global "capacity arms race" for Indium Phosphide (InP) semiconductors. These materials are essential for the high-speed optical interconnects that allow AI GPU clusters to communicate, yet current production is failing to keep pace with demand.
Industry data reveals the InP sector is grappling with a supply-demand gap of nearly 70%, a deficit expected to persist with high growth through 2028. This shortage is particularly acute at the foundational level of the supply chain; international giants currently control approximately 95% of the global InP substrate market, creating a strategic bottleneck for firms attempting to scale their optical infrastructure.
The Shift to Optical Interconnects
Indium Phosphide is a compound semiconductor prized for its direct bandgap, which makes it indispensable for generating and detecting laser light in fiber-optic systems. As AI clusters scale, the industry is moving away from traditional pluggable optics toward Near-Packaged Optics (NPO) and Co-Packaged Optics (CPO). These architectures reduce signal loss and power consumption—critical hurdles in maintaining GPU performance.
While silicon photonics is increasingly used for routing light, InP remains the primary material for light emission and modulation. This has created a complementary partnership where InP and silicon are integrated heterogeneously to achieve the bandwidth necessary for next-generation AI workloads.
Massive Capital Injections
To break existing overseas monopolies and secure future capacity, companies are committing massive capital expenditures. In Japan, JX Advanced Metals plans to invest up to ¥120 billion to increase its InP substrate capacity, targeting an increase of up to 10 times current levels by FY2030.
Chinese firms are also aggressively expanding. Sanan Optoelectronics has begun mass-producing 6-inch InP optical chips, currently maintaining a capacity of 2,750 wafers per month with a target of nearly 6,000 wafers per month. Additionally, Source Photonics, a subsidiary of DSBJ, is investing US$1.2 billion in an expansion project for optical chips and high-speed optical modules in Changzhou, China.
Strategic Implications
This transition from electrical to optical interconnects represents a fundamental architectural shift. Because InP production is so highly concentrated, the ability to secure substrate and foundry capacity has become a matter of strategic importance. The current rush for localization, particularly within China, is a direct response to the vulnerability of relying on a market where a few international players hold nearly all the leverage.
What to Watch
Industry observers are now monitoring whether these massive investments can close the 70% supply gap before AI scaling hits a physical wall. While capital is flowing into substrates and epitaxy, the next critical phase will be the scaling of advanced packaging and testing to integrate these InP components into CPO architectures. The speed at which these new production lines come online will likely determine the pace of AI hardware evolution through the end of the decade.