UK epitaxy maker IQE swung to core profit of GBP6 million (about US$8.1 million) in the first half of 2026 on AI data-center demand, while warning that China's control of indium phosphide (InP) could become a major industry risk.
The real signal here is not one supplier's return to profit — it is where the money is landing in the AI stack. As clusters scale, the bottleneck shifts from compute to the fabric connecting it. 800G and 1.6T optical interconnects rely on InP-based lasers and photonic components, and the epitaxy layer that grows those materials is a thin, specialized market. IQE's turnaround is a proxy for how much of the data-center capex wave is quietly flowing into photonics rather than logic.
The warning matters more than the profit. InP depends on indium, and China dominates both raw indium output and refining capacity. After Beijing's 2023 curbs on gallium and germanium, this is the logical next pressure point — a material that sits directly in the critical path of every hyperscale optical link. A single-country chokepoint on a low-volume, high-leverage input is exactly the kind of asymmetry that turns a supply question into a strategic one. Hyperscalers optimizing for GPU allocation may be underpricing the risk one layer down.
For Japan, this cuts two ways. The country's compound-semiconductor and photonics suppliers are among the few credible non-Chinese sources of InP substrates and optical components, which positions Japanese material makers as beneficiaries if buyers diversify away from Chinese supply. That is a genuine opening for Japan's materials sector, distinct from the logic-chip race it has largely ceded.
The flip side is exposure. Japanese optical-transceiver and module vendors that consume InP wafers face the same input concentration, and any Chinese tightening would hit their margins and lead times directly. For SIers and enterprise buyers procuring AI infrastructure in Japan, the practical takeaway is that data-center capacity constraints increasingly originate in obscure material chokepoints, not just GPU allocation. Procurement and cloud-cost planning should treat optical-component lead times as a scheduling variable, and domestic material sovereignty in InP deserves a place on the policy agenda alongside advanced-node fabrication.
