Our Take
Chip supply risk is real, but the story conflates export control with scarcity—China's dominance is structural, not recent policy.
Why it matters
Data centre operators and semiconductor buyers depend on indium phosphide for photonic components and RF chips that handle AI workloads at scale. Any disruption to Chinese exports reshapes capex timelines and supply contracts for the next 18 months.
Do this week
Infrastructure teams: audit your indium phosphide-dependent supplier contracts and lock multi-year commitments before Q2 2025 so you can avoid single-source dependency.
China's export stranglehold on a critical AI chip material
China controls the majority of global indium phosphide (InP) exports, a semiconductor substrate essential for high-speed integrated circuits used in AI data centre networking and optical components. Reuters reports that this export concentration now poses a supply-chain risk to data centre buildouts as demand for AI infrastructure accelerates.
Indium phosphide is not a new material, but its role in photonic integrated circuits and millimetre-wave components has grown critical as data centres scale training and inference workloads. The compound substrate enables faster signal processing than traditional silicon in specific applications, particularly in optical transceivers and radio-frequency front-ends.
The supply vulnerability stems from manufacturing concentration, not recent Chinese policy announcements. InP wafer and substrate production is clustered in a small number of regions, with China controlling a significant share of export volumes. Any disruption—regulatory, geopolitical, or capacity-driven—cascades into chip assembly delays for vendors serving hyperscalers.
Capex timelines hinge on optical and RF component availability
Data centre operators do not buy indium phosphide directly; they buy finished optical transceivers, power amplifiers, and photonic chips from vendors like Broadcom, Marvell, and others who depend on InP substrates. If Chinese export restrictions tighten or supply tightens, lead times for these components extend from months to quarters.
The risk is neither hypothetical nor distant. Hyperscalers are already in multi-year capex cycles to add GPU-accelerated capacity. Any constraint on optical networking or RF components forces either design workarounds (costlier, slower) or delayed facility activation. Secondary markets for InP wafers are thin, so substitution is not trivial.
This is not about Taiwan or geopolitical posturing alone; it is about a single-source, high-value input to chips that AI infrastructure cannot easily skip.
Lock supply agreements before policy tightens
Infrastructure procurement teams should map indium phosphide dependencies in their optical and RF supply chain now. Ask vendors: which components rely on InP substrates, which suppliers source them, and what geographic diversification exists.
For operators finalizing data centre phased build plans, prioritise locking multi-year contracts for photonic and RF components with confirmed InP supply commitments. Single-year or spot-market contracts leave you exposed to 6–12 month lead-time extensions if export constraints emerge.
For chip vendors, this is a supply-chain resilience moment. Qualifying alternative substrates or pre-positioning InP inventory is expensive but cheaper than customer defection when lead times spike.