China's control over exports of indium phosphide threatens to disrupt the rollout of AI data centres, Reuters reports, putting a strategic compound used in high-speed optical components at the centre of a technology supply dispute.

Indium phosphide (InP) is a compound semiconductor that carries data as light rather than electricity. Transceivers built on InP substrates move traffic between servers inside data centres and between facilities across fibre networks. As AI clusters push interconnect speeds higher, operators depend on InP-based photonics to keep pace with the bandwidth demands of large-scale model training and inference.

China dominates the global supply chain for the material. The country is the leading producer of refined indium, the metal from which indium phosphide substrates are made, and Chinese firms occupy key positions in the processing chain that turns raw material into wafers ready for device fabrication. That concentration gives Beijing a chokepoint comparable to its position in gallium, germanium and rare-earth processing — materials it has already subjected to export licensing regimes.

What does the export control mean for AI builders?

Reuters reports that China's hold over indium phosphide exports poses a direct threat to AI data centre deployment. The logic is structural: hyperscalers and network equipment vendors cannot quickly substitute the material, because viable alternatives at the required performance levels remain limited.

For data centre operators, the exposure sits in optical transceivers. Modern AI clusters interconnect tens of thousands of accelerators, and the link speeds those clusters require — moving to ever-faster lanes between racks and buildings — depend on indium phosphide and related compound semiconductors. A supply interruption would hit exactly the segment of the market now expanding fastest.

The risk scenario is not hypothetical. Beijing has demonstrated willingness to use mineral and materials export controls as leverage in trade disputes, and each restriction it has imposed on adjacent materials has sent buyers scrambling for stockpiles and non-Chinese suppliers.

Who is exposed?

The potential impact cuts across several layers of the AI infrastructure stack:

• Hyperscale data centre operators building out AI training and inference capacity
• Optical transceiver manufacturers that use InP-based lasers and photodetectors
• Network equipment vendors supplying the interconnect fabric of AI clusters
• Chipmakers and substrate producers outside China seeking to scale alternative supply

Western supply chains have taken initial steps to reduce dependence. Substrate makers in Japan, Europe and North America have expanded InP wafer capacity, and gallium arsenide or silicon photonics can cover parts of the performance envelope. But building out refining, crystal growth and wafer production to a scale that matches Chinese output takes years, not months.

What happens next?

Much depends on how strictly Chinese authorities apply any licensing regime to indium phosphide shipments, and whether Washington and its allies respond with further investment in domestic compound-semiconductor production. For AI infrastructure planners, the Reuters report is a signal to treat optical materials supply as a first-order planning risk rather than a procurement afterthought.

Companies building AI capacity now face a familiar calculation: pay a premium to diversify supply early, or accept the possibility that a single export decision in Beijing can slow the next phase of data centre construction.