The contest over artificial intelligence capacity is increasingly being fought in the plumbing of the system—materials, licensing regimes, and supply chains—where China now exerts leverage on a critical chokepoint: indium phosphide for high‑end data‑center optics. Key Points Since February 2025, China has subjected indium phosphide (InP) and related precursors to a formal dual-use export licensing regime, immediately tightening a key input to AI data-center photonics. The controls have not stopped exports outright, but they have raised wafer prices roughly 250% and introduced delay, uncertainty, and destination scrutiny into global supply chains. Indium phosphide sits at the heart of high-speed optical interconnects that move data between GPUs; constraints at this layer can weigh on the pace and cost of AI data-center build-out. Evidence so far shows a broad, discretionary Chinese regime affecting multiple regions, not a documented, U.S-specific embargo—yet it still functions as strategic leverage in a wider tech rivalry. U.S. and allied responses range from material stockpiling and diversification to exploring substitutes such as organic electro‑optic polymers and thin-film niobate, but these alternatives remain pre‑scale. From GPU Scarcity to Materials Leverage: How Indium Phosphide Became Strategic For most observers, bottlenecks in AI infrastructure begin and end with GPUs. That focus misses the physics of how modern data centers actually run. Once you cluster tens of thousands of accelerators, the limiting factor quickly becomes how fast and efficiently you can move data between them. That is the domain of photonics—optical transceivers, modulators, and switches operating at tens or hundreds of gigabits per second over short-reach fiber inside racks and halls. https://www.youtube.com/watch?v=T5llNguBapY Indium phosphide is one of the workhorse compound semiconductors for this layer. It underpins lasers and high‑frequency devices that exceed what silicon can reliably do in the relevant wavelength and modulation regimes. Reuters reporting notes that InP plays “a crucial role in photonics for next-gen AI data centers,” and that AXT and Sumitomo together account for almost 80 percent of global InP substrate manufacturing. When you disturb this supply, you don’t just marginally affect a niche component; you tug directly on the throughput of AI models at scale. China’s February 2025 Export Controls: What Actually Changed The critical policy turn came on 4 February 2025, when China’s Ministry of Commerce and the General Administration of Customs issued Announcement No. 10, adding specific indium-related items to the country’s dual‑use export control list. Indium phosphide substrates (coded 3C004.a), trimethylindium (3C004.b), triethylindium (3C004.c), and associated production technology (3E004) all became controlled items. Exporters were required to obtain licenses from the competent commercial authority of the State Council for each shipment, with the regime taking effect immediately. Several aspects of this design matter. First, refined indium metal itself was not listed; the focus is on compound substrates and precursors used in advanced electronics and optics. Second, this is not a codified embargo. It is a discretionary licensing regime—China can approve, delay, or deny shipments on a case‑by‑case basis, with substantial latitude and limited transparency. That structure is characteristic of Beijing’s broader use of export controls as a flexible tool of statecraft rather than a blunt cutoff. Optional Friction, Real Consequences: Price Spikes and Supply Delays Even without a declared ban, practical effects have been pronounced. Since the controls took effect, the average cost of a six‑inch InP wafer has risen from roughly $1,400 to about $5,000—an increase of around 250 percent, driven by constrained supply and buyer competition. That scale of escalation is unusual in a mature materials market and indicates that licensing delays and uncertainty are materially tightening availability. Corporate disclosures corroborate this picture. AXT, the second‑largest InP substrate producer with manufacturing in China, reported that its Q4 2025 revenue came in below guidance “due mainly to fewer‑than‑expected export control permits for indium phosphide” issued by the Ministry of Commerce. The company did not receive its first permit for exporting InP until late June 2025, months after the controls took effect, underscoring how licensing discretion can ripple through production and delivery schedules. Reuters’ follow‑up reporting on indium exports more broadly adds granularity: customs officials have begun demanding end‑user information, including locations, and slowing approvals, with at least one European buyer facing paperwork demands not seen before. Critically, Reuters stated it had not identified shipments that were outright halted, which supports the interpretation that China is using scrutiny and delay, not formal bans, to alter market conditions. How This Hits AI Data-Center Build-Out To translate materials policy into AI capacity, you have to walk the supply chain. InP substrates produced by firms such as AXT are sold to epitaxy specialists, who grow layered structures with precise bandgap and refractive-index properties. Chipmakers then fabricate lasers and modulators from these wafers for inclusion in optical transceivers supplied to networking vendors and hyperscale operators. When substrate prices triple and export permits become unpredictable, every downstream actor adjusts behavior: safety stock rises, contract terms grow more conservative, and some projects slip to accommodate longer lead times. Industry commentary and trade press coverage describe this as a developing bottleneck. Digitimes notes that China’s restrictions have “triggered a shortage of InP substrates, creating a bottleneck” in optical communications materials. Technology media similarly frames tightened InP scrutiny as a threat that could slow the very AI infrastructure the technology depends on. None of these sources provide a precise metric for how many megawatts of U.S. data-center capacity are delayed, but they converge on the view that photonics supply is now a constraining factor alongside GPUs. What the record does not yet offer is a quantified causal chain: a verified model showing that China’s controls have, for instance, pushed back specific U.S. AI deployments by known months or reduced aggregate installed GPU capacity by a defined percentage. That absence reflects the complexity of the supply chain—capacity additions, alternative sourcing, and demand shifts all move at once—and the novelty of the controls. It does not negate the clear signal from pricing, corporate backlog, and licensing friction. Two Fronts of Pressure: Export Licensing and Indium Scrutiny The “two-front squeeze” language captures a broader pattern emerging from the evidence. On one front, indium phosphide and its precursors are now locked behind a dual-use licensing regime that directly touches AI photonics. On the other, China has begun tightening scrutiny on exports of indium metal itself, demanding more information from buyers and hinting that the metal could be pulled into the formal control list. Reuters describes buyer concern that indium—a specialized metal essential for advanced data centers—may be incorporated into the same export-control framework that Beijing has used as a “formidable trade strategy.” Customs officials have already started to slow approvals and request end-user details from at least one European purchaser. Even though indium metal is not yet controlled under Announcement No. 10, the combination of existing InP licensing and looming broader metal scrutiny gives Beijing a layered toolkit: it can tighten either or both levers as circumstances warrant. From a strategic standpoint, this is classic optional friction. China is signaling that it can create delays and uncertainty around both the substrates and the upstream metal, without permanently cutting off supply. That signal alone can influence investment decisions, pricing power, and diplomatic dynamics, especially when paired with China’s dominant share of global indium production—about 70 percent, according to multiple industry and media accounts. Is the U.S. Specifically Targeted? What the Evidence Shows—and Doesn’t Here the record is more nuanced. The strongest reporting documents that the export controls and scrutiny apply to indium-related items regardless of destination. Reuters cites concerns from European and North American buyers alike, and its investigation explicitly notes that it has found no evidence of shipments being halted, let alone selectively blocked for U.S. customers. Moreover, the face of Announcement No. 10, as summarized by corporate and analytical sources, lists controlled items but does not identify particular countries; exporters must seek licenses for any international shipment, suggesting a formally destination-neutral regime. In practice, licensing discretion could produce de facto differentiation, but that is an inference, not yet backed by shipment-level data comparing clearance times or denial rates by country. At the same time, the backdrop is unambiguous: China introduced these controls against a history of tit‑for‑tat technology restrictions, including U.S. curbs on advanced chipmaking equipment and Chinese controls on gallium, germanium, and other strategic materials. It would be surprising if Beijing did not weigh the impact on U.S. AI capacity in its internal calculus. Yet the available public sources lack the internal policy memoranda or official statements that would convert that plausible inference into proof of intent. Analysts arguing for a bespoke anti‑U.S. AI strike are, at this stage, extrapolating from context. China’s Wider Export-Control Playbook Beijing’s treatment of InP and indium fits a now familiar pattern. Over the past decade, China has steadily refined a “granular toolkit of choke points” in critical inputs—rare earths, battery materials, specialty gases, and compound semiconductors—managed through licensing, quotas, and customs discretion. The goal is not always to cut off supply; often it is to introduce enough uncertainty to remind counterparties where key dependencies lie. Reuters explicitly frames China’s InP move as an extension of established export restrictions on rare earth elements, describing export controls as one of Beijing’s most powerful trade tools. When viewed together with restrictions on gallium arsenide (GaAs) imposed in August 2023 for military-related applications, and the 2025 InP controls that similarly cite dual-use concerns, a coherent strategy comes into focus: secure leverage over materials that matter for high‑end electronics, optics, and defense, and be prepared to tighten the screws as geopolitical tensions rise. How U.S. and Allied Actors Are Responding On the U.S. side, response has begun on several tracks. Defense and industrial-policy actors have reportedly moved to stockpile indium and related materials, seeking to buffer near‑term shocks. Trade diplomacy has explored critical-minerals cooperation among G7 partners, aiming to diversify away from Chinese supply and to coordinate responses to export controls. At the industry level, companies are experimenting with technological workarounds. Photonics firms and investors highlight organic electro‑optic polymers, exemplified by Lightwave Logic’s perkinamine platform, as a way to achieve high-speed modulation on silicon photonics without relying on brittle, heat‑sensitive InP chips manufactured under Chinese jurisdiction. Thin-film lithium niobate offers another candidate, promising excellent performance but facing manufacturing challenges and smaller wafer sizes. These directions are serious, but they are not ready-made replacements: most are still in validation phases, with design wins and large-scale deployment projected closer to the late 2020s. For the foreseeable future, InP remains central to high‑performance data-center optics. Implications for America’s AI Build-Out For U.S. policymakers and infrastructure builders, the lesson is straightforward and uncomfortable: AI capacity is now constrained not just by domestic policy and GPU access, but by foreign discretion over obscure materials. A licensing delay in Beijing can ripple through wafer fabrication in China, epitaxy capacity in Asia and Europe, and transceiver assembly for deployment in American server farms. The evidence does not justify alarmist claims that China has “choked off” U.S. AI; exports continue, and the controls are formally global. It does, however, support a sober assessment that China has engineered a lever it can pull to raise costs, slow timelines, and inject risk into the physical layer of AI infrastructure. That lever operates across two fronts—InP substrates and the upstream indium metal—and sits within a broader tapestry of export controls and tech rivalry. For a country intent on sustaining AI leadership, treating these materials and licensing regimes as peripheral would be a mistake. The strategic contest has moved into the supply chain, and the side that best understands and manages these invisible chokepoints will shape not just market share, but the tempo at which new AI capabilities can be built, deployed, and governed. In early May, the CEO of American chipmaker Coherent flew to China with President Trump after warning investors his company was running short of indium phosphide, a key material that turns electricity into laser light for high-speed data links in AI data centers. China produces… pic.twitter.com/pKWhNcvZit — The Epoch Times (@EpochTimes) July 26, 2026 Where the Evidence Is Thin and What Comes Next Several gaps remain for analysts hoping to move beyond inference. The full legal text and implementing guidance of Announcement No. 10, including any internal criteria for license approval or denial, are not widely available in translation. Shipment-level data comparing license processing times and outcomes by destination are similarly scarce. Corporate testimony has begun to surface in earnings calls and SEC filings, but a systematic picture of who faces what friction is still emerging. Closing those gaps would allow a more precise characterization of China’s intent and practice: whether the regime is genuinely country-neutral, whether certain destinations are quietly deprioritized, and how tightly controls are coupled to specific AI-related applications. Until then, the prudent course is to treat the existing evidence as what it is—strong on the existence and broad impact of controls, weaker on targeted intent—and to plan for a future in which materials governance is a central theater of technological competition. Sources: zerohedge.com, reuters.com, timesoftunis.com, x.com, kucoin.com, thenextweb.com, digitimes.com