Beijing's new scrutiny of indium exports turns a quiet chip material into a live AI infrastructure risk, and Washington is only now trying to catch up.
You can spend billions on GPUs and still get caught by a metal most investors couldn't name six months ago. That is the point of China's tightening grip on indium, a byproduct of zinc refining that feeds into indium phosphide, the substrate used in high-speed optical chips for AI data centers.
China controls about 70% of the world's indium supply, and that gives Beijing leverage in a part of the AI buildout that rarely gets the attention given to Nvidia accelerators or cloud power contracts. Indium phosphide helps optical interconnects move data between GPU clusters fast enough for modern AI workloads. If that material gets delayed, the data center doesn't care how confident the capex plan looked on a slide.
Reuters reported this month that Chinese customs officials have begun asking foreign buyers of indium metal for end-user disclosures. One European buyer told Reuters it was asked for the first time this year to identify where its indium was going and who would use it. A major North American buyer said approvals that once happened the same day were now taking several days, and described the mood as tense. Indium metal itself still isn't on Beijing's formal export control list. That is exactly why the move matters. China doesn't need a headline ban to slow a supply chain.
This has not come from nowhere. China formally restricted gallium and germanium exports in 2023. It added indium phosphide, the processed substrate, to its export control list in February 2025. Now the checks appear to be moving further upstream, from the finished substrate toward the raw metal that makes it possible. That sequence is the story. Beijing is not improvising around the edges, it's building pressure point by pressure point.
The price signal is already ugly. Since China put controls on indium phosphide substrates last year, the average cost of a 6-inch InP wafer has jumped 250% to about $5,000. AXT, one of the few non-Chinese InP substrate suppliers, reported more than $60 million in order backlog as of the first quarter of 2026. That backlog is what export permits look like when they turn into missed delivery windows and crowded queues.
The alternative supply base is thin. AXT and Japan's Sumitomo control nearly 80% of non-Chinese substrate manufacturing, according to the figures cited in the report. For buyers, that means there isn't some deep bench of spare capacity waiting outside China. You can call another supplier, but if everyone else is calling the same two names, the problem has only moved from a customs desk to a production schedule.
Coherent has been the loudest public example of how quickly this quiet materials issue has become a boardroom problem. CEO Jim Anderson flew to Beijing in May as part of the US business delegation accompanying President Trump, specifically to raise export license delays as a bilateral trade issue. A chip company executive attaching a wafer supply problem to a presidential trip tells you how exposed this chain has become.
Coherent is not standing still. The company said in May that it's doubling InP wafer capacity at its Texas facility this year and plans to more than double it again by the end of 2027. That helps, but it doesn't solve the immediate problem. Building substrate capacity takes time, qualified equipment, trained staff and customer validation. Buyers who need optical components this year can't simply wish a new supply base into existence.
Google has also moved to lock in optical supply for its AI data centers, according to reporting from Cloud News. That is the right instinct. When a hyperscaler starts securing access to a niche material, you should stop treating that material as a footnote. The AI infrastructure race is no longer just about chips, power and land. It is also about whether the materials underneath the optical layer arrive on time.
Washington is late to a known dependency
The US Defense Logistics Agency has issued a request for proposals to stockpile up to 403 tons of indium over three years. That is a serious number, and it puts indium where it now belongs: inside the strategic materials discussion, beside rare earths and other inputs where Chinese dominance creates a defense and industrial risk. Fastmarkets has reported that the broader US critical minerals stockpiling push is tied to a $12 billion effort assembled since 2025.
Frankly, Washington is reacting after the warning signs were already visible. Indium phosphide has been a known dependency for optical networking for years. Gallium and germanium were the early alarm. Indium phosphide controls were the follow-up. Customs scrutiny on indium metal is the next move in the same contest, and it leaves the US trying to buy insurance after the pressure has already started.
For AI infrastructure investors, the practical point is simple. The billions going into optical interconnects rest on a supply chain with one obvious geographic choke point. Coherent's Texas expansion should help. A DLA stockpile should help. Neither changes the fact that real diversification of InP supply takes years and capital, not press releases. Until that capacity exists, Beijing keeps leverage over how quickly the next generation of AI data centers gets built.
Also read: Jio's network-native AI push at Reliance's 49th AGM is a bet that carriers can replace the app layer • UnitedHealth Group is betting $3 billion on AI agents that call your doctor before you do • The AI chip benchmark wars are back and this time Nvidia's rivals have real numbers to show
