China’s control over indium phosphide exports threatens AI data centre rollout

China’s export controls on indium phosphide, a critical material for AI photonics, are creating a major supply bottleneck for chipmakers including Coherent, Lumentum and AXT. The average price of a 6-inch InP wafer has reportedly surged 250% to $5,000, while U.S. firms are scrambling for permits, non-Chinese supply and new capacity that can take 2-3 years to add. The issue underscores a broader trade weaponization risk that could slow AI data center buildouts and pressure photonics supply chains.

Analysis

The key market takeaway is not just input scarcity, but a forced re-pricing of the AI optical stack’s fragility. InP is an upstream choke point with long qualification cycles, so the immediate beneficiaries are not the obvious end-product names but the few suppliers with non-China substrate access, inventory buffers, and customer lock-in. That argues for relative outperformance of the better-supplied optical names over the next 1-3 quarters, while China-exposed substrate makers may keep gaining domestic share even if export approvals eventually loosen.

The second-order effect is margin compression rather than outright unit destruction. Hyperscalers will likely keep building, but the cost of accelerating photonics adoption rises enough to slow some deployments, push out lower-priority projects, and preserve demand for incumbent copper-based architectures longer than consensus expects. That creates a subtle negative for the most AI-capex-sensitive semiconductor names if photonics bottlenecks delay rack density improvements and force higher power budgets into 2026 planning cycles.

The most asymmetric risk is policy reversal: even a partial easing of Chinese permits could trigger a sharp relief rally in the substrate chain because the market is pricing scarcity, not just inconvenience. But that is likely a months-long process, not days, given the need for requalification and the lack of near-term substitution capacity. Near term, the trade should favor names with pricing power and diversified sourcing; the losers are the most China-dependent input buyers with the least inventory cover and the weakest ability to pass through cost inflation.

Key Decisions for Investors

• Go long LITE vs short AXTI for a 1-3 month relative-value trade: LITE has more pricing power and better insulation from substrate disruption, while AXTI faces the highest policy/permit overhang and customer pushback.
• Add to COHR only on pullbacks and hedge with short MRVL against an AI-optics delay scenario: COHR has operating leverage to photonics demand, but MRVL is more exposed to delayed photonics adoption and capex timing risk over the next 2 quarters.
• Buy AXTI downside via put spreads 2-4 months out: the stock screens as the most headline-sensitive to permit uncertainty, and any export-license disappointment should re-rate the multiple faster than fundamentals can recover.
• Use a basket long LITE / COHR / short a broader AI-capex proxy if the market starts pricing a photonics bottleneck as a delayed AI rollout: this isolates supply-chain alpha without taking full beta to the AI complex.
• If China signals broader license approvals, take profits quickly on short AXTI and long LITE hedges: the move could reverse violently because the market is positioning for scarcity, not stable normalization.