China's critical mineral controls are no longer a background trade issue for AI infrastructure. They now sit uncomfortably close to the optical networking supply chain that every large data center build depends on.
The AI hardware story is usually told through GPUs, power contracts, and the race to pour concrete fast enough. But the quieter constraint is moving data between all those chips once they arrive. That is where indium phosphide matters. It is a compound semiconductor used in high-speed lasers, photodiodes, and modulators for optical networking, the equipment that lets data center clusters behave like one enormous machine instead of a warehouse full of isolated servers.
The original version of this story overstated the export rule around indium phosphide substrates. What China announced on February 4, 2025, was export control over indium and several other strategic materials, including tungsten, tellurium, bismuth, and molybdenum. That does not mean every indium phosphide wafer is automatically blocked at the border. It does mean suppliers that depend on China-linked upstream material now have another licensing layer sitting between customer demand and physical delivery. In a market already stretched by AI buildouts, that distinction matters.
Beijing has been building this toolkit for several years. Gallium and germanium were placed under export licensing in 2023. Graphite followed. In December 2024, China went further by banning exports to the United States of gallium, germanium, antimony, and other dual-use materials, although later trade diplomacy softened parts of that regime while leaving licensing leverage in place. As AP reported when the February 2025 controls were announced, the new measures covered elements critical to modern high-tech products. The pattern is clear enough: China is not only responding to U.S. chip restrictions, it is mapping the minerals that sit underneath the hardware stack.
Why InP Matters
Indium phosphide is valuable because AI clusters are becoming bandwidth machines. A rack full of accelerators is only useful if memory, switches, transceivers, and optical links can keep feeding it data. Copper reaches its limits quickly at the speeds and distances modern AI systems require. Optical networking pushes more data with lower latency and better power efficiency, which is why the industry is moving toward faster transceivers and co-packaged optics.
That is the strategic context behind Nvidia's March 2026 photonics deals with Lumentum and Coherent. Nvidia committed $2 billion to each company through nonexclusive, multiyear agreements tied to advanced laser components and optical networking capacity. The point was not simply to make another financial investment. It was to secure access to the pieces of the network that make future AI factories work at scale.
Those deals also show how the market is reacting to supply risk. The largest buyers are not waiting for spot markets to normalize. They are using balance sheets to reserve capacity, influence supplier roadmaps, and reduce exposure to chokepoints they cannot control. That leaves smaller cloud providers, networking vendors, and AI infrastructure startups in a harder position. They may have demand, but demand is not the same thing as guaranteed access to lasers, wafers, and qualified optical components.
The Real Constraint
The danger is not that China flips one switch and the entire optical supply chain stops tomorrow. Supply chains rarely break that cleanly. The more realistic risk is slower and more frustrating: license delays, uncertain customer approvals, higher working capital needs, and procurement teams forced to carry more inventory than they planned. That kind of friction does not produce one dramatic headline, but it changes how companies plan data center expansions.
It also shifts power toward vertically integrated suppliers. Companies that control more of their laser, photonics, or module production have a better chance of navigating material shortages and licensing uncertainty. Firms that mostly assemble components sourced through exposed supply chains have less room to maneuver. In a normal cycle, that difference might show up as margin pressure. In an AI infrastructure boom, it can determine who gets to ship.
For investors and operators, the takeaway is straightforward. The AI buildout is not constrained by chips alone. It is constrained by the materials, optics, power systems, and manufacturing capacity that sit around the chips. China's indium controls are not the whole story, but they are a useful warning. The next phase of AI infrastructure will reward companies that treat supply chain control as strategy, not paperwork.
Also read: DDN Is Hunting a Strategic Investor and Its New Valuation Will Be a Referendum on AI Storage • Federal agencies now have three days to patch the most dangerous vulnerabilities and AI acceleration is the explicit reason • China sank a data center in the ocean and it outperforms anything America has built on land
