China’s Indium Phosphide Export Controls Threatening AI Data Centres

The Invisible Bottleneck Threatening the AI Infrastructure Revolution

The semiconductor supply chain has a long history of hidden vulnerabilities surfacing at the worst possible moments. From the 2011 Thailand floods that devastated hard drive production to the 2021 chip shortage that paralysed the automotive industry, the pattern is consistent: a material or component that receives almost no attention until its absence brings entire technology ecosystems to a halt. Today, a compound semiconductor material that most people have never heard of is quietly emerging as one of the most consequential chokepoints in the global AI buildout.

Indium phosphide sits at the intersection of two of the most powerful technological and geopolitical forces of the current era: the exponential scaling of artificial intelligence infrastructure and Beijing's increasingly sophisticated approach to materials-based trade competition. Understanding how China indium phosphide export controls and AI data centres became entangled requires tracing a supply chain that runs from zinc smelters in Yunnan province to the optical interconnect fabrics inside the world's most advanced AI training clusters.

From Copper Wires to Light: Why AI Data Centres Have a Photonics Dependency

The Physics Problem That Copper Cannot Solve

Every generation of data centre architecture has eventually encountered a materials boundary. For AI infrastructure operating at hyperscale, that boundary is now defined by the physical limitations of copper-based electrical interconnects. As AI workloads have grown exponentially, the bandwidth density requirements inside server clusters have pushed copper-based transmission to its thermal and signal integrity limits.

The numbers illustrate the problem clearly. Power consumption for copper interconnects in AI-optimised data centres now represents approximately 35-40% of total facility energy usage, compared to the 15-20% range typical of conventional data centre architectures. At the modulation speeds required for next-generation AI training clusters operating at 800G and 1.6T optical module specifications, electrical signal transmission simply cannot deliver the required performance without generating untenable heat loads and signal degradation.

Photonic interconnects, which transmit data using light through optical fibres rather than electrical current through metal conductors, offer a fundamentally different performance profile:

• Bandwidth densities exceeding 1 terabit per second per square millimetre, approximately ten times the capability of advanced electrical interconnect solutions
• Power consumption reductions of 60-70% per bit transmitted compared to copper equivalents
• Superior signal integrity over longer distances within and between server racks
• Dramatically reduced thermal output in densely packed AI cluster configurations

The transition from copper to photonics is therefore not a technology preference but a physical necessity. As one infrastructure engineer summarised the situation, the bandwidth density constraints of copper interconnects are fundamentally insurmountable at the scale required for next-generation AI systems.

What Indium Phosphide Actually Does and Why Nothing Else Can Replace It

Indium phosphide is a III-V compound semiconductor material formed by combining indium and phosphorus in a crystal lattice structure. Its strategic importance stems from a combination of material properties that no commercially viable alternative currently replicates across the full range of data centre photonic applications.

The critical properties of InP include:

• A direct bandgap of 1.34 electron volts at room temperature, enabling efficient light emission and detection at wavelengths optimal for short-reach data centre transmission
• Electron mobility of approximately 5,400 cm²/V·s, more than three times the 1,400 cm²/V·s of silicon, allowing modulation speeds exceeding 200 gigabaud
• Thermal conductivity of approximately 68 W/m·K, enabling heat management in densely integrated photonic circuits
• A zincblende crystal structure providing superior epitaxial compatibility with related III-V materials used in photonic device stacks

These properties translate directly into the components that make AI data centre optical interconnects function. InP-based vertical-cavity surface-emitting lasers (VCSELs) achieve modulation bandwidths exceeding 30 GHz, compared to the 10-15 GHz range typical of silicon photonics alternatives. Electro-absorption modulators built on InP substrates enable the ultra-high-speed data encoding required at 400G and 800G transmission rates. Distributed feedback lasers fabricated on InP wafers provide the coherent light sources essential for wavelength-division multiplexed optical interconnects.

While silicon photonics has made meaningful technical advances, it cannot match InP performance for wavelengths below 1.3 micrometres, which are critical for short-reach data centre applications where the majority of AI cluster communications occur. The substrate-to-device manufacturing pipeline begins with raw indium refined from zinc smelter by-products, proceeds through crystal growth into InP boules, is then sliced and polished into wafer substrates, undergoes epitaxial deposition of device layers, and concludes in the fabrication of finished photonic integrated circuits. The substrate layer represents the most strategically sensitive point in this entire chain.

How China Achieved Dual Control Over the Global InP Supply Chain

The Numbers Behind Beijing's Strategic Position

China's dominance over indium phosphide is structural rather than circumstantial, reflecting decades of industrial policy investment in zinc smelting capacity and downstream compound semiconductor processing. The scale of this dominance is visible in the data:

The concentration risk operates at two distinct but interconnected levels. At the upstream level, China's overwhelming share of global indium mining creates the foundational raw material dependency. At the midstream level, the presence of major InP substrate manufacturers operating production facilities inside China, most notably AXT's Chinese manufacturing subsidiary, creates a second independent chokepoint.

Furthermore, as China's export controls on materials like bismuth have already demonstrated, Beijing has refined its ability to apply precise upstream pressure without resorting to broader trade confrontation.

Beijing's approach represents a second-generation trade instrument: rather than restricting finished photonic components directly, upstream compound and substrate export licensing creates a slower, more deniable form of supply disruption that is structurally harder to counter through equivalent trade measures. Paul Triolo of consulting firm Albright Stonebridge Group has described this framework as a targeted materials chokepoint toolkit, used to slow or condition the export of upstream compounds and substrates that determine whether the optical module ecosystem can scale quickly enough to meet hyperscaler demand. (Reuters, 2025)

A Deliberate Escalation of an Established Playbook

Beijing's use of materials-level export controls is not an improvised response but rather the refinement of an approach that has already proven effective in other sectors. Since 2023, China's rare earth export restrictions on gallium, germanium, graphite, and rare earth processing chemicals have disrupted global automotive, semiconductor, and aviation supply chains, demonstrating the practical effectiveness of upstream materials controls as trade instruments.

The InP case represents an evolution in tactical sophistication. Where rare earth controls operated at the commodity level, InP restrictions operate at the processed substrate level, targeting a more specific and harder-to-substitute product. This precision makes the controls more effective because:

1. The restricted material has fewer immediate alternatives than broad rare earth categories
2. Qualification requirements for new substrates create mandatory multi-month transition timelines regardless of alternative supply availability
3. The affected technology (photonic interconnects) has no practical substitute in current-generation AI infrastructure
4. The diplomatic costs of retaliation are asymmetric, as the US has no equivalent upstream materials leverage of similar specificity

The Cascading Supply Chain Impact: From Substrate Shortage to Delayed AI Clusters

Step-by-Step: How Export Licence Delays Propagate Through the Ecosystem

The mechanism by which Chinese export licence delays translate into AI data centre deployment constraints is not immediate but follows a predictable cascade across the photonics value chain:

1. Export licence processing delays at Chinese InP substrate manufacturers, including AXT's Chinese operations, create order backlogs
2. Substrate shortages propagate to epitaxy houses and optical chip foundries dependent on Chinese-sourced wafers
3. Wafer price inflation accelerates as constrained supply meets growing AI-driven demand, with 6-inch InP wafers reaching approximately $5,000 per unit
4. Component production constraints emerge at transceiver manufacturers and laser module producers unable to source adequate substrate volumes
5. Optical module supply tightens across the photonics ecosystem, with leading manufacturers reporting sold-out order books
6. Data centre interconnect deployment slows as optical module lead times extend from weeks to months
7. AI cluster buildout timelines lengthen and capital expenditure efficiency declines as photonic component costs inflate

The Companies Bearing the Greatest Exposure

The supply chain stress is not evenly distributed. Certain nodes in the photonics ecosystem face disproportionate exposure, and China's control over indium phosphide exports is increasingly recognised as a systemic risk rather than a company-specific issue.

AXT (world's second-largest InP substrate producer and major supplier to Coherent) has described InP export permits as the most significant challenge it currently faces. The company manufactures most of its InP substrates in China, and its Chinese subsidiary received its first export permits only in June 2024, leaving a substantial order backlog that continued to constrain supply into 2025. (Reuters, 2025)

Lumentum has been reported as sold out through 2028 despite quadrupling its production output, illustrating that capacity expansion alone cannot offset the substrate supply constraint. (Reuters, 2025)

Taiwanese optical manufacturers including VPEC and LandMark Optoelectronics have faced direct disruptions from AXT permit delays, demonstrating that the impact extends beyond US companies to the broader Asian photonics supply chain. (Reuters, 2025)

Coherent, backed by Nvidia's $2 billion investment, identified the shortage directly during a May 2025 earnings call. The severity of the issue prompted the company's chief executive to join a US business delegation travelling to China to raise the export licence delay issue with relevant authorities. (Reuters, 2025)

The Qualification Cycle Problem: Why Switching Suppliers Is Not Simple

A critical factor that amplifies the vulnerability is the semiconductor industry's mandatory supplier qualification process. Even when alternative InP substrate sources exist in principle, photonics chipmakers cannot simply begin purchasing from new suppliers. The qualification process typically involves:

• Extensive material characterisation testing against application-specific specifications
• Integration testing with existing epitaxial processes and device fabrication flows
• Reliability and performance validation across temperature and humidity ranges
• Multi-lot statistical analysis to confirm process reproducibility
• Internal sign-off procedures and potentially customer notification obligations

Qualification timelines in photonic substrate supply chains range from several months to over a year depending on the complexity of the downstream application. For AI data centre transceiver applications, where performance specifications are tightly controlled, qualification cycles at the longer end of this range are typical. This creates what can be described as a temporal vulnerability window: even if non-Chinese supply were theoretically adequate in volume, the time required to qualify new substrate sources means the supply gap cannot be closed rapidly regardless of commercial intent.

Alternative Supply: Why the Non-Chinese Ecosystem Cannot Fill the Gap Quickly

Mapping the Non-Chinese InP Substrate Landscape

The two largest non-Chinese InP substrate producers both face structural constraints that limit their ability to absorb demand displaced from China:

Sumitomo's situation is particularly revealing. Despite being a major global producer, a person familiar with China's photonic chip industry has confirmed that Sumitomo consumes much of its InP substrate output internally, meaning the broader global market remains structurally undersupplied regardless of Sumitomo's absolute production volumes. (Reuters, 2025)

The timeline challenge for new capacity is equally constraining. New InP substrate manufacturing facilities typically require two to three years from investment decision to commercial production. This timeline reflects the capital intensity of crystal growth infrastructure, the complexity of achieving consistent crystal quality at commercial yields, and the time required for new facilities to complete customer qualification processes.

The Chinese Domestic Expansion Paradox

A less widely understood dimension of the supply challenge involves the Chinese domestic producers who are actively scaling InP capacity. Yunnan Germanium's 189 million yuan (~$28 million) investment targeting 450,000 single InP wafers annually and its reported 74% surge in InP wafer shipments represent significant capacity additions, while Guangdong Xiandao's investment in 40 tons of annual InP crystal output through subsidiary Guangdong Xianrui adds further to the domestic Chinese supply base. (Reuters, 2025)

A source at a major Chinese InP manufacturer has indicated that domestic producers are focused on the Chinese domestic market for the near term, with limited evidence that the Chinese government would favour domestic players over companies such as AXT when it comes to export approvals. (Reuters, 2025)

This creates an ironic supply paradox: Chinese domestic InP production is growing rapidly, but that growth is primarily directed inward. Meanwhile, Western photonics companies face qualification barriers that would prevent rapid adoption of Chinese domestic producers even if export approvals were granted. The supply gap cannot therefore be closed from either direction within the timeframes relevant to current AI cluster deployment schedules.

The Diplomatic Escalation: When Materials Become Trade Agenda Items

The elevation of InP export controls to the level of bilateral diplomatic negotiation represents a threshold crossing in the evolution of technology trade competition. The issue was raised during US-China trade talks in Seoul ahead of the May 2025 Trump-Xi summit, and the urgency was significant enough to prompt the chief executive of a major photonics manufacturer to travel as part of a presidential business delegation specifically to address export licence delays. (Reuters, 2025)

This diplomatic escalation illustrates a broader pattern. The geopolitical landscape of metals and mining has shifted decisively toward materials-level controls that force technology companies into geopolitical negotiations they would otherwise avoid, creating leverage that operates through commercial disruption rather than diplomatic confrontation. The compound semiconductor supply chain contains multiple analogous concentration risks beyond InP, including gallium arsenide, indium gallium arsenide, and gallium nitride substrates, suggesting that the current episode may be a preview of recurring strategic pressure points rather than an isolated incident.

Industry and Investor Implications: What the InP Constraint Means for AI Infrastructure

Short, Medium, and Long-Term Impact Assessment

Important disclaimer: The following impact assessment represents analysis based on publicly available information and industry reporting as of mid-2025. The pace of diplomatic resolution, capacity expansion, and qualification cycle completion introduces substantial uncertainty into all projections. This content is strictly informational and does not constitute investment advice.

What Infrastructure Investors and Hyperscalers Need to Understand

Several strategic implications emerge from the current InP supply situation that are not yet reflected in mainstream infrastructure investment frameworks.

Photonic component procurement now requires multi-year forward planning. The combination of substrate supply constraints, qualification cycle requirements, and capacity expansion timelines means that photonic component availability has shifted from a manageable operational variable to a critical path item in AI cluster deployment scheduling. Facilities currently in planning need to account for these constraints explicitly.

The data centre optical transceiver market is entering a high-growth phase with constrained supply. Industry forecasts indicate the market will grow from approximately $2.5 billion in 2024 to over $8.5 billion by 2027, a compound annual growth rate exceeding 50%. Against this demand trajectory, substrate supply constraints could support sustained pricing power for non-Chinese InP substrate producers and the manufacturers who control their own substrate supply.

InP joins rare earths and advanced logic chips as a strategic materials category. For AI infrastructure investors, the photonics supply chain layer has been demonstrably reframed as a strategic vulnerability requiring the same risk management frameworks previously applied to rare earth elements and advanced semiconductor nodes.

Vertical integration is emerging as a competitive advantage. Manufacturers that have invested in own-source InP substrate capacity, such as Coherent's Texas facility expansion, are positioned to avoid the external market price inflation and availability constraints faced by competitors dependent on third-party substrate supply.

Nvidia's $2 billion investment signals into each of Coherent and Lumentum in March 2025 represent not merely financial positions but strategic supply chain anchoring, reflecting hyperscaler recognition that photonics component security requires equity-level commitment rather than purchasing relationships alone.

The Broader Pattern: Compound Semiconductors as the New Critical Materials Frontier

The InP episode provides a template for understanding where materials-level trade competition is heading. Beijing's approach has evolved from broad commodity controls toward precision targeting of specific compounds that gate entire technology ecosystems. The characteristics that make a material vulnerable to this approach are clearly visible in the InP case:

• High geographic concentration of upstream production in China
• Limited commercially viable substitutes in near-term applications
• Long qualification cycles that prevent rapid supply diversification
• Critical role in a technology ecosystem experiencing exponential demand growth
• Relatively low volumes compared to commodity materials, making Western strategic stockpiling costly but achievable

Several other compound semiconductor materials exhibit similar vulnerability profiles. Gallium arsenide substrates used in RF semiconductor manufacturing, indium gallium arsenide materials used in infrared sensing, and germanium substrates used in high-efficiency solar cells all display concentration patterns that echo the InP situation. In addition, the impact on bismuth export control pricing offers a useful parallel for understanding how quickly upstream controls can translate into downstream price inflation.

Consequently, for policymakers, infrastructure investors, and AI ecosystem participants, the central lesson of China indium phosphide export controls and AI data centres is that the materials layer of the technology stack now requires the same level of strategic attention previously reserved for the chip design and manufacturing layers. Analysts at the Congressional Research Service have similarly noted the growing strategic significance of compound semiconductor materials to US technology competitiveness. The photons travelling through AI data centre optical interconnects are only as secure as the substrate supply chain that makes their transmission possible.

This article is intended for informational purposes only and does not constitute financial, investment, or trading advice. Statements about supply chain conditions, pricing, and company positions reflect publicly available information as of mid-2025 and are subject to change. Readers should conduct independent research before making any investment or commercial decisions based on information contained herein.

Want to Stay Ahead of the Next Major ASX Resource Discovery Before the Market Catches On?

Discovery Alert's proprietary Discovery IQ model scans ASX announcements in real time, instantly identifying high-potential mineral discoveries across more than 30 commodities — including the critical and compound semiconductor materials reshaping global supply chains — and delivering actionable insights directly to subscribers. Explore historic discoveries and the returns they generated, then begin your 14-day free trial at Discovery Alert to position yourself ahead of the broader market.