Riding a surge in demand for optical communications driven by artificial intelligence (AI), Taiwan's compound semiconductor makers—including Visual Photonics Epitaxy, IntelliEPI, WIN Semiconductors, GCS Holdings, and Advanced Wireless Semiconductor Company—are poised for a new phase of growth. However, that expansion is colliding with a critical constraint: a shortage of indium phosphide, or Indium Phosphide, a material essential to high-speed data transmission.
Industry executives say the scarcity of InP substrates—combined with booming demand from AI data centers—has deepened a supply-demand imbalance, creating a bottleneck at a time when next-generation computing requires ever-faster connectivity.
At its 2026 developer conference, Nvidia outlined a future in which data center interconnects will rely on a combination of copper cabling, optical communication, and co-packaged optics (CPO). The company's message was clear: the industry is moving toward a "copper-and-optics in parallel" architecture. Today's dominant optical modules operate at 800G speeds, but the roadmap points toward 1.6-terabit systems.
As chip performance accelerates, transmission speed—not compute power—is emerging as the primary constraint in AI infrastructure. InP, prized for its high electron mobility, suitability for high-frequency applications, and thermal stability, has become all but indispensable.
Yet supply is struggling to keep pace. The simultaneous upgrade of legacy data centers and the rapid construction of new AI facilities have driven demand for optical components to soar. Manufacturers have been unable to ramp capacity quickly enough, and the shortage of InP substrates remains unresolved.
Globally, supply is concentrated among a handful of players, including Japan's Sumitomo Electric Industries, US-based AXT, which has production in China, and Germany's Freiberger Compound Materials. While additional suppliers in France and Japan are expanding output, the barriers to entry remain high: manufacturing equipment must withstand pressures up to 3 times atmospheric pressure, and Japan requires certification processes that can take up to 18 months.
Yung-Chung Kao, Chairman of IntelliEPI, said export controls from China exacerbated substrate shortages in the second half of 2025, trimming the company's InP epitaxy output. To cope, some customers have begun supplying their own substrates—a share that rose to nearly 20 percent by late 2025 and could exceed 30 percent this year.
Optical modules, which consist of transmitters and receivers, are also evolving rapidly. IntelliEPI expects 200G PIN photodiodes to drive growth in 2026, alongside development of 200G vertical-cavity surface-emitting lasers (VCSELs) and quantum-dot laser epitaxial wafers. Optical communications products now account for more than half of its revenue, surpassing electronics.
Meanwhile, GCS Holdings is advancing on both the transmitter and receiver fronts, with plans to begin mass production of 200G photodiodes and continuous-wave lasers in 2026. A 70-milliwatt CW laser has already completed certification, and volume production for a major customer's 200G photodiode is expected to begin in the second quarter.
WIN Semiconductors, for its part, has indicated that products such as photodiodes, continuous-wave distributed feedback lasers, and electro-absorption modulated lasers are in the validation stage. Given differing certification timelines, receiver components could begin contributing to revenue as early as 2026.
For now, however, the industry's ambitions remain tethered to a stubborn reality: without sufficient InP supply, the race to build faster AI systems may be limited not by processing power, but by the speed at which data can move.
Article translated by Elaine Chen and edited by Jack Wu
