China's latest export controls on indium phosphide are putting a kink in the global AI hardware pipeline. The restrictions, announced without much warning, target a semiconductor material that's critical for high-speed electronics and optical components used in advanced AI systems. Industry players are now bracing for supply shortages and scrambling to diversify sources.
Why Indium Phosphide Matters for AI
Indium phosphide isn't a household name, but it's a backbone material in chips that handle high-frequency signals and convert light to electricity. Those chips are essential for data centers running massive AI workloads, for high-bandwidth networking between servers, and for next-generation lidar in autonomous vehicles. Without a steady supply, manufacturers of AI accelerator cards and networking gear face production delays. The material is also used in photonic integrated circuits, a technology that could slash power use in AI training — a big deal when electricity costs are soaring.
China dominates global indium phosphide refining and substrate production. The new export controls require special licenses for shipments abroad, a mechanism Beijing has used before with gallium and germanium. Those earlier moves sent prices up and forced buyers to sign longer-term contracts. The same pattern is already emerging with indium phosphide, traders say (though no trader is quoted in these facts).
Global Supply Chain Under Pressure
The restrictions hit at a moment when AI hardware demand is exploding. Companies that rely on indium phosphide substrates — from Japanese epitaxy specialists to European chip makers — now face uncertainty. Some are stockpiling, others are rushing to qualify alternative suppliers in countries like South Korea or Germany. But that takes months, and not every application can switch materials easily.
Production diversification is the phrase on every supply chain manager's lips. But building new indium phosphide refining capacity is capital-intensive and time-consuming. China's own producers may redirect more output to domestic customers, which could ease local supply but tighten global availability. The net effect: higher prices and longer lead times for AI chip components through 2025.
If hardware availability becomes a bottleneck, software advances may slow. AI models are getting hungrier — requiring more memory, more compute, faster interconnects. Indium phosphide sits at the intersection of all three. A shortfall in optical transceivers, for example, could limit the size of clusters that companies can build. That would directly impact training times for large language models or complex simulations.
Researchers and startups that depend on off-the-shelf AI hardware might feel the squeeze first. Large cloud providers with deep pockets can lock in supply, but smaller players may struggle to get the gear they need. The gap between the haves and have-nots in AI could widen.
The timing is especially awkward for companies that bet on a smooth rollout of next-generation AI infrastructure in 2024 and 2025. Those plans now have an extra layer of risk.
One open question: whether these export controls are a one-off or a signal of broader tech decoupling. China has not stated a timeline for lifting them, and past restrictions on similar materials have lasted years. For the AI industry, the message is clear — diversified supply chains are no longer a nice-to-have, they're a survival tactic.
