The bottleneck in the AI buildout is turning out to be a metal most people have never heard of. China has tightened its scrutiny of exports of indium phosphide, a compound essential to the high-speed optical chips that move data inside AI data centres, in a move that threatens to slow the very infrastructure the technology depends on.
Indium phosphide, or InP, is not a household material, but it is becoming a strategic one. As data-centre operators shift from pushing electrical signals through copper to sending light through optical fibres, a technique known as photonics, InP has become the core material with no ready substitute.
The faster the AI industry wants to move data between chips, the more it needs the compound, and China sits at the chokepoint.
That position is a matter of geology and processing. China produces around 70% of the world’s indium, and since export controls on InP took effect in early 2025, Beijing has been slow to approve the licences that let the material leave the country.
The delays, rather than an outright ban, are the lever: a permit that does not arrive is as effective as a prohibition, and harder to challenge.
The market has felt it. The price of a six-inch InP wafer has climbed from roughly $1,400 to about $5,000 since the controls began, an increase of around 250%, as buyers compete for constrained supply.
Nvidia-backed chipmaker Coherent warned of a shortage earlier this year, and AXT, the world’s second-largest InP substrate producer, has described the export permits as the most significant challenge it currently faces.
The episode fits a now-familiar pattern in the US-China technology contest. Where Washington has restricted China’s access to advanced chips and chipmaking tools, Beijing has answered by leveraging its dominance over critical materials, having already deployed controls on gallium, germanium, and rare earths.
InP is the same weapon pointed at a different part of the supply chain, the optical layer rather than the logic layer.
What makes InP potent is precisely that it targets infrastructure rather than end products. The compound goes into the transceivers and optical components that knit together the thousands of accelerators in a modern AI cluster, so a squeeze on it does not stop any single chip from working; it slows the rate at which whole data centres can be built and wired. The constraint shows up as delayed construction, not failed silicon.
It also lands as the AI industry’s appetite for compute is at its most acute, with operators racing to build capacity faster than the supply chain can support.
The same pressure visible in the scramble for chips and components now extends to a niche material that few outside the industry tracked a year ago. China’s leverage over it has turned a specialist input into a geopolitical instrument.
The deeper worry for the AI industry is precedent. If a delay in InP permits can slow data-centre construction, the same lever can be applied to any of the specialised inputs where China holds a commanding share, turning a diversified supply chain into a series of single points of failure.
That fragility is now a strategic planning problem for Western governments and operators alike, part of the wider contest over technology supremacy in which materials have become as decisive as the chips they enable.
Substitution offers little near-term relief. Building InP production capacity outside China is possible but slow, requiring years of investment in refining and wafer fabrication that the current shortage does nothing to accelerate.
In the meantime, buyers are left managing allocation, paying the higher prices, and lobbying through diplomatic channels for the permits to move, a position of dependence that the controls were designed to exploit.
The InP controls were also raised directly with Beijing; Coherent’s chief executive brought the licensing delays up during a US business delegation’s visit to China, a sign of how seriously the buyers take the threat.
Whether the permits start flowing again, and on what terms, is now part of the broader negotiation between the two governments over technology and trade. For the AI buildout, the answer determines how fast the lights can go on.
Stephan is the sports journalist for the Maple Grove Report.
