Indium Phosphide Exports Become China’s New Chokepoint in AI Data Centre Supply Chain

AI data centre

Indium phosphide exports have become a strategic pressure point in the global AI data centre supply chain as China’s licensing controls delay shipments of a material essential for high-speed optical chips. The restrictions are exposing a new vulnerability in AI infrastructure: the physical materials behind silicon photonics and optical interconnects.

The issue has moved quickly from a specialist semiconductor concern to a high-level trade and industrial policy problem. Coherent, a key optical components supplier backed by Nvidia, warned in early May that indium phosphide shortages were already affecting the market. Its chief executive then joined a US business delegation to China as companies sought relief from export licence delays.

Indium phosphide exports matter because AI data centres are moving beyond copper-based interconnects. As AI workloads grow, hyperscalers need faster, lower-latency and more energy-efficient data transmission between processors, accelerators, switches and optical modules. Indium phosphide is one of the core materials enabling that shift.

The material is used in high-speed optical chips, lasers, detectors and photonic components. These devices support the optical links that move huge volumes of data across AI clusters. Without reliable indium phosphide substrates and wafers, the expansion of advanced AI data centre networks could slow.

China’s control over indium phosphide exports shows that critical materials policy is becoming more granular. Beijing no longer needs to restrict only rare earths or finished technology products. It can also influence upstream compounds, substrates and wafers that determine whether advanced semiconductor supply chains can scale.

Export Controls Expose a Hidden Bottleneck in Silicon Photonics

Silicon photonics has become a critical technology for AI infrastructure because it allows data to move through light rather than electrical signals. This reduces energy use per bit and supports the bandwidth required by large AI systems.

But silicon photonics is not only a silicon story. The most advanced optical systems often require compound semiconductor materials such as indium phosphide, gallium arsenide, gallium nitride and germanium-based compounds. Indium phosphide is especially important for lasers and high-speed optical devices.

This creates a difficult supply chain problem. AI companies, hyperscalers and chipmakers are racing to scale optical modules, but one of the key substrate materials remains highly concentrated. China is the world’s largest indium producer, accounting for about 70% of global output in 2024.

That concentration became more serious after China introduced export restrictions on indium phosphide in February 2025. Since then, licence delays have created backlogs for companies that manufacture or source InP substrates from China.

AXT, one of the world’s largest indium phosphide substrate producers and a major supplier to Coherent, said export permits were its most significant challenge. The company manufactures most of its InP substrates in China and only received its first permits last June. It still faces a large order backlog.

The effect has spread beyond individual suppliers. Coherent, Lumentum, VPEC and LandMark Optoelectronics all sit inside the optical components ecosystem that depends on reliable substrate supply. When permit delays hit upstream InP material, the impact moves through wafers, chips, optical modules and AI data centre equipment.

Prices show the severity of the shortage. Since China introduced export restrictions, the average price of a 6-inch indium phosphide wafer has surged by 250% to about $5,000. That price increase reflects both physical scarcity and the strategic premium attached to non-disrupted supply.

The supply squeeze also comes at a time of aggressive photonics investment. Nvidia announced $2bn investments each in Coherent and Lumentum in March. Marvell Technology also moved into photonics through its acquisition of Celestial AI, reflecting stronger demand for optical technology in AI computing.

These investments show where the industry is heading. AI infrastructure needs optical interconnects to manage power, latency and bandwidth. But China’s indium phosphide controls mean that materials availability could become a gating factor for deployment.

Companies are trying to respond. Coherent plans to double its InP wafer capacity at its Texas plant this year and more than double it again by the end of 2027. US photonics firms are also seeking supply from non-Chinese producers such as Sumitomo Electric Industries.

However, capacity additions are slow. New substrate plants can take two to three years to bring online. Qualification cycles are also long because optical chipmakers cannot easily switch substrate suppliers without testing performance, reliability and consistency.

This makes the shortage difficult to solve quickly. Even if new capacity is announced, it may not arrive fast enough to meet near-term AI data centre demand. Meanwhile, many non-China producers already consume part of their own output internally, reducing the amount available to the broader market.

China’s Materials Chokepoint Strategy Strengthens Domestic Producers

China’s indium phosphide export controls are creating both pressure and opportunity. They restrict global supply, but they also support domestic Chinese substrate producers that are expanding capacity.

Yunnan Germanium, Guangdong Xiandao and Zhuhai Dingtai Xinyuan are among China’s leading domestic InP substrate players. Their role is becoming more important as Beijing uses materials controls to strengthen strategic leverage across semiconductor and AI supply chains.

Yunnan Germanium has already moved to expand. The company announced a 189mn yuan investment in April to raise production capacity to 450,000 single InP wafers annually. Its shipments of InP wafers rose by 74% in 2025, showing fast domestic market growth.

Guangdong Xiandao is also expanding through its subsidiary Guangdong Xianrui. The project is expected to produce 40 t/yr of indium phosphide crystals, which are used as raw material for substrates.

These investments fit a broader pattern. China is not only defending control over upstream critical materials. It is also building downstream processing capacity in higher-value compound semiconductor materials.

However, Chinese producers may not immediately solve the global shortage. Some are still seeking export approvals, and any overseas shipments may be limited. Domestic demand remains a priority, especially as China builds its own AI, optical communications and semiconductor ecosystem.

AXT

Supplier qualification creates another barrier. Companies such as Coherent and Lumentum are unlikely to switch easily from established suppliers. Coherent relies heavily on AXT, while Lumentum sources mainly from Sumitomo and JX Advanced Metals. New suppliers must pass demanding qualification cycles before they can enter critical optical chip supply chains.

This gives China’s export controls a long-lasting effect. Even if alternative suppliers exist, the market cannot instantly redirect demand. The bottleneck is not only production volume. It is qualified, high-quality, customer-approved substrate supply.

The strategic lesson is clear. AI supply chains are not only exposed to advanced chips, GPUs and packaging capacity. They also depend on a deep materials stack that includes indium, phosphorous chemistry, InP crystals, substrates, wafers, lasers, detectors and optical modules.

This is why indium phosphide exports have become so important. AI data centre buildouts need more optical links as clusters grow larger. Copper interconnects face limits in speed, distance and energy consumption. Photonics offers a solution, but only if the materials chain can scale.

For the US and its allies, the response will likely require more than emergency licence negotiations. It will require investment in indium recovery, InP crystal growth, substrate manufacturing, wafer capacity and long-term offtake agreements. It may also require strategic stockpiles for high-purity indium and compound semiconductor substrates.

The issue also strengthens the case for recycling and secondary recovery. Indium is often produced as a by-product, making primary supply difficult to expand quickly. Recovering indium from industrial scrap, displays, semiconductors and related waste streams could become more important if export controls persist.

For AI data centre developers, the risk is timing. Demand for optical modules is accelerating now, while new ex-China capacity may not fully arrive until 2027 or later. That mismatch could raise costs, delay deployments and intensify competition for qualified photonics suppliers.

The market may therefore see a split. Companies with secured InP supply will be better positioned to support hyperscaler demand. Companies exposed to licence delays, qualification bottlenecks or spot-market wafers may face higher costs and delivery risk.

The Metalnomist Commentary

China’s control over indium phosphide exports shows that the AI race is becoming a materials race. The next bottleneck may not be only GPUs or power supply, but the compound semiconductor substrates needed to move data fast enough inside AI clusters.