China restricts exports of indium phosphide, key for AI optical chips

What happened?

China has imposed new export restrictions on indium phosphide (InP), a semiconductor compound essential for manufacturing high-speed optical chips. These chips are fundamental for data transmission in AI data centers, where low-latency, high-bandwidth optical links are required to connect thousands of accelerators (GPUs, TPUs). According to The Next Web, Beijing has raised the level of scrutiny on InP exports, requiring stricter licenses and detailed end-use justifications. This measure is part of the trade and technology tensions between China and the West, which have intensified since 2018. China controls approximately 80% of global refined indium production and a significant share of InP wafer manufacturing. The restriction comes at a time of peak demand: investments in AI infrastructure have soared, with companies like NVIDIA, Microsoft, and Google competing to secure computing capacity. According to IDC data, spending on AI data centers will grow 30% annually until 2027, reaching $500 billion. Dependence on China for this critical material echoes the 2010 rare earth crisis, when Beijing restricted exports, causing a 500% price surge and a global race to diversify sources.

Why is it important?

Indium phosphide is the material of choice for photonic devices operating in the 1.3-1.6 µm wavelength range, where optical fibers have minimal attenuation. Optical transceiver modules—such as 800G and 1.6T—rely on InP-based chips for signal modulation and detection. Without these components, data centers cannot scale server interconnects, limiting the performance of distributed AI systems. InP is particularly critical for fast-tuning lasers and Mach-Zehnder modulators used in 400 Gbps and higher links. According to a Yole Group report, the InP photonics market will grow from $2.5 billion in 2023 to $5 billion in 2028, driven by AI demand. China's restriction threatens to strangle that growth. Unlike silicon, which is abundant, indium is a byproduct of zinc and lead mining, with global annual production of only 1,500 tons. China, with its vast reserves and refining capacity, dominates the supply chain. This geographic concentration is a systemic risk for the global tech industry, comparable to dependence on Taiwan for advanced chips.

Consequences for the industry

China's move could have effects comparable to the 2020-2023 chip shortage, but focused on a critical niche. Manufacturers of optical equipment (such as Lumentum, Coherent, II-VI) and hyperscalers (Amazon, Google, Microsoft) could face delays in delivering high-speed modules. In the short term, prices of InP and optical chips could rise between 20% and 40%, according to Bernstein analysts. For example, the cost of an 800G transceiver could increase from $1,200 to $1,700. This would directly impact data center operators' margins, already under pressure from rising energy consumption. In the long term, the restriction accelerates the search for alternatives, such as gallium phosphide (GaP) or silicon photonics, although these technologies are not yet mature for the speeds required in AI. Silicon photonics, for instance, struggles to efficiently generate light at optimal wavelengths. Companies like Intel and Cisco have invested in silicon photonics, but its performance in 1.6T modules is still inferior to InP. Another alternative is lithium niobate, but its manufacturing process is complex. Transitioning to these technologies could take 3 to 5 years, leaving a critical supply gap.

Reaction of global players

Companies like NVIDIA and AMD are already diversifying their supply chains, investing in foundries outside China, such as TSMC (Taiwan) or GlobalFoundries (USA). However, InP wafer production requires specialized processes that few companies master, such as Sumitomo Electric (Japan) or IQE (UK). Sumitomo Electric produces 4-inch InP wafers, while IQE offers epitaxy services. But their combined capacity is insufficient to meet global demand. The European Union and the United States could include InP on their lists of critical materials and encourage local production through subsidies. The US CHIPS Act already allocates $52 billion for semiconductors, but does not specifically address photonic materials. The EU, for its part, has launched the European Alliance for Processors and Semiconductor Technologies, but InP is not a priority. Meanwhile, China seeks to use its material advantage as a geopolitical tool, recalling its control over rare earths and other inputs. In 2023, China already restricted exports of gallium and germanium, affecting compound semiconductor production. These coordinated actions suggest a deliberate strategy to pressure the West in the technology war.

What should readers know?

For IT and business professionals who rely on cloud and AI, this news implies that infrastructure costs could rise and deployment timelines lengthen. Startups building AI models should consider hardware availability in their projections. Investors in semiconductors and photonics should monitor stocks of companies exposed to InP, such as Lumentum (down 8% in the past week) and Coherent. At a macro level, the restriction reinforces the trend of technological decoupling between China and the West, with implications for national security and economic competitiveness. Governments should consider creating strategic reserves of indium and promoting recycling of this metal, which currently is only recovered at 5%. Additionally, research into alternatives must be prioritized. For the average reader, this means that AI services like ChatGPT or Bard could become more expensive or slower if data centers cannot expand. In summary, China's InP restriction is a reminder that the technology supply chain remains fragile and that dependence on a single country for critical materials is a strategic risk that must be urgently mitigated.