There’s a compound called indium phosphide that you’ve probably never thought about. It sits inside the optical transceivers that let AI data centers move massive amounts of data at blistering speeds. And right now, China has its hand firmly on the supply valve.
Beijing added indium phosphide, commonly referred to as InP, to its export control list on February 4, 2025. The move created a permitting bottleneck that has rippled through the global supply chain for high-speed optical components, the kind needed to wire up the hyperscale AI clusters that every major tech company is racing to build.
The chokepoint no one saw coming
Inside a modern AI data center, thousands of accelerators need to communicate with each other at extraordinary speeds. That communication happens through optical transceivers, and the highest-performance versions of those transceivers, the 800G and beyond models that next-generation AI clusters require, rely on indium phosphide substrates.
Over 60% of InP consumption goes to optical communications and AI data centers.
The problem is concentrated supply. AXT Inc., a major producer of InP substrates, operates through its subsidiary Tongmei in China. When Beijing imposed the export controls, Tongmei’s ability to ship product ground to a near-halt. The company didn’t receive its first InP export permits until around June 11, 2025, more than four months after the restrictions took effect.
That delay translated directly into revenue pain. AXT reported significant revenue shortfalls tied to permit issuance problems in Q4 2025 and has warned that the impacts will extend into 2026.
A familiar playbook with new stakes
China has previously weaponized its dominance over critical materials like gallium and germanium, imposing export restrictions that sent shockwaves through semiconductor supply chains. Indium phosphide is the latest entry in that playbook.
Restricted availability of InP could directly hinder the deployment of energy-efficient 800G-plus optical transceivers. Without those transceivers, scaling AI clusters becomes harder, slower, and more expensive.
The scramble to diversify
As of mid-2026, US officials are actively engaging with Chinese counterparts to address the licensing bottleneck and support AI infrastructure timelines.
Companies aren’t waiting for diplomacy to solve the problem. Lumentum, a major optical components maker, has been shifting production outside of China to reduce its exposure. New InP production capacity is being developed in locations including North Carolina, part of a broader push to onshore or near-shore critical semiconductor materials.
Building out substrate manufacturing isn’t like spinning up a new software service. Crystal growth facilities require specialized equipment, cleanroom environments, and months of qualification before they can produce material that meets the exacting standards of high-performance optics. The gap between deciding to diversify and actually having alternative supply flowing is measured in years, not quarters.