Indium Squeeze Threatens America’s AI Plans

China is no longer trying to win the AI race only by building more chips; it is now quietly reshaping the physical supply chains that feed America’s data centers, using export controls on a single compound semiconductor—indium phosphide—to exert leverage on the infrastructure bottlenecks at the heart of modern AI.

Key Points

Beijing has placed indium phosphide, a critical photonics material for AI data centers, under a formal export-licensing regime, driving sharp price increases and tightening supply.
This is a discretionary control system, not a blanket embargo: shipments still move, but under slower, more uncertain licensing that creates optional friction in the AI hardware supply chain.
China produces the majority of the world’s indium, giving it structural leverage over indium phosphide and related materials even when controls are narrowly written.
The squeeze operates on two fronts: direct licensing pressure on InP wafers and intensifying scrutiny of upstream indium metal, with downstream effects on optical interconnects that limit AI data-center build-out.
The evidence clearly shows cost and timing disruption; what remains unproven is whether Beijing is using this leverage to target U.S. AI capacity specifically rather than global buyers more broadly.

From Silicon to Photonics: Why Indium Phosphide Suddenly Matters

To understand why one export-control decision in Beijing now shows up in the budgets of American cloud providers, you have to start with a quiet shift in data-center physics. For the last two decades, hyperscale computing largely meant pushing more silicon CMOS chips into racks and fighting the resulting heat with ever more aggressive cooling. That model breaks down at AI scales. When clusters link tens of thousands of accelerators, the bottleneck is not the compute die but the interconnect—the physical infrastructure moving trillions of bits per second between servers.

Optical links solve that bottleneck, and they do it using photonics—chips that manipulate light rather than electrons. Here indium phosphide (InP) is pivotal. As a compound semiconductor, InP allows lasers and high-speed modulators to operate at telecom wavelengths and bandwidths that silicon alone cannot reliably achieve. Reuters reporting on the current controls is blunt: InP “plays a crucial role in photonics for next-gen AI data centers,” and China’s restrictions “have posed significant challenges in the quest to create the fastest and most energy-efficient components for AI data centers.”

The supply chain runs through a handful of specialized firms. Substrate producers such as AXT and Sumitomo manufacture the crystalline InP wafers that serve as the foundation for photonic devices; epitaxy houses deposit precise compound layers; chipmakers integrate these into optical transceivers and co-packaged optics that sit alongside GPUs and custom AI accelerators. This chain has substantial geographic concentration in China, where AXT and its Tongmei subsidiary operate key substrate plants. When that concentration meets export licensing, leverage emerges.

China’s February 2025 Move: Formal Controls, Optional Friction

On 4 February 2025, China’s Ministry of Commerce and the General Administration of Customs issued Announcement No. 10, adding a cluster of indium-related items to the country’s dual-use export control list. The controlled items are specific and technical: indium phosphide substrates (3C004.a), trimethylindium (3C004.b), triethylindium (3C004.c), and associated production technology and process data (3E004). Refined indium metal is not listed in this announcement; the target is the compound semiconductor and its immediate precursors, not every form of the element.

Crucially, this is a licensing regime. Exporters must apply to the competent commercial authority under the State Council for a license before shipping any covered item abroad, and the controls took effect immediately. There is no language in the available record indicating a formal embargo or categorical ban. That distinction is not semantic. A blanket prohibition closes trade outright; a licensing system gives the state an adjustable valve—approvals can be slowed, scrutinized, or denied case by case, creating uncertainty and delay without overtly stopping global commerce.

Reuters’ June 2026 examination of the regime found exactly this pattern. Since February 2025, China has “restricted export licenses for indium phosphide,” with buyers reporting longer lead times, tense documentation reviews, and a notable rise in costs. AXT itself disclosed that its Q4 2025 revenue was constrained “due mainly to fewer-than-expected export control permits for indium phosphide being issued by China’s Ministry of Commerce,” and that the government now requires a permit for every customer order. This is discretionary administration in practice: some permits arrive, but not all, and not always on schedule.

Price Shock and Supply Tightening: What the Evidence Actually Shows

The most measurable consequence so far is price. The average cost of a 6-inch InP wafer has climbed from roughly $1,400 before the controls to about $5,000 afterward—a 250 percent increase, according to Reuters and corroborated by independent technology outlets. That is not a marginal adjustment; it is a structural shock to the cost base of high-speed photonics. For AI data centers that demand thousands of optical transceivers, this translates directly into higher capital expenditure and tougher trade-offs between bandwidth and budget.

Supply has tightened as well. Digitimes reporting describes China’s export restrictions as triggering a shortage of InP substrates and creating a bottleneck in optical communications materials. AXT, the second-largest InP substrate producer globally, has openly discussed a substantial order backlog extending toward the end of the decade and the need to secure export permits for every shipment, introducing unpredictable delays. Sumitomo, the other dominant substrate supplier, faces the same licensing environment when exporting from China-linked production.

At the raw-material level, Reuters and social-media reproductions of its reporting indicate that China produces about 70 percent of the world’s indium, a byproduct of zinc refining that is critical for making indium phosphide. That upstream dominance magnifies the impact of any constraint on the downstream compound, because alternative refining capacity cannot quickly compensate. Buyers have reported customs officials demanding detailed end-user information and stretching clearance times from same day to several days, framing the process as tense but not yet prohibitive.

It is this combination—formal licensing of InP, heightened scrutiny of indium metal, price escalation, and permit delays—that constitutes the “two-front squeeze” on AI infrastructure. The controls operate where photonic demand is surging fastest, and they do so through mechanisms that amplify uncertainty rather than simply cutting off supply.

Strategy, Not Accident: Export Controls as a Trade Tool

China’s indium phosphide controls do not exist in isolation; they fit a decade-long pattern of what might be called strategic export management. Policymakers have repeatedly used licensing and list-based controls on rare earths, gallium, germanium, and other specialty inputs as levers in broader trade disputes and technology rivalries. Reuters explicitly frames the InP decision as part of the same toolkit, an extension of methods already tested in earlier resource conflicts.

The available Chinese rationale, while not fully documented in the public record, points to dual-use concerns. Semiconductor Today notes that the government imposed trade restrictions on gallium arsenide in August 2023 and indium phosphide in February 2025 “aiming to restrict the export of materials used for military applications” and requiring an export permit for every customer order. That is a familiar justification in export-control law: advanced optoelectronic materials serve both civilian AI and defense systems, including communications, sensors, and directed-energy platforms.

From a supply-chain perspective, however, the effect does not depend on whether policymakers emphasize national security or bargaining leverage. A discretionary licensing regime that covers key photonics inputs gives Beijing the ability to tune friction in response to geopolitical conditions. Approvals can move quickly when relations are stable, then slow or become more demanding when tensions rise or when other states tighten their own controls on chipmaking equipment.

This is precisely how buyers now experience the system. A European purchaser told Reuters that customs had, for the first time, demanded full end-user information, including location; a significant North American buyer said approvals that previously cleared in a day now took several days as documentation was re-examined. Neither reported outright denials or stopped shipments; Reuters, after canvassing the market, found no evidence of cargo being blocked. What they described instead was optional friction—the kind that forces firms to carry more inventory, renegotiate delivery schedules, and reconsider concentration risk.

Is the U.S. Being Targeted, or Is This Global Pressure?

The central analytical question is whether China’s InP licensing regime is aimed specifically at constraining U.S. AI build-out or whether it reflects a broader, destination-neutral policy that happens to bite wherever advanced data centers are being built. On this point, the evidence is mixed and requires careful separation of documented facts from inference.

On the factual side, we know that the control list is written in technical terms without named destination carve-outs; it covers InP substrates, precursors, and technology regardless of where they are headed. We also know that the buyers Reuters quotes include both European and North American firms and that Reuters has not found shipments that were halted, only slowed. No primary-source Chinese legal text in the public package explicitly states a goal of slowing U.S. AI deployment or differentiating license standards by country.

We also lack quantified metrics linking the controls to specific U.S. infrastructure delays—missed commissioning dates for data centers, canceled GPU cluster deployments, or measured shortfalls in optical-transceiver output attributable directly to InP licensing rather than general demand growth. The best documented outcomes are price increases, permit delays, and company-level revenue and backlog impacts.

At the same time, the absence of country-specific language does not prove neutrality in enforcement. A licensing system can be applied unevenly without declaring it in public law; customs officers can ask more questions of some buyers than others, and approvals can be informally prioritized or deferred. Side B’s case—that effects are global and friction is routine—rests on the lack of shipment-level comparative data rather than on positive evidence that enforcement is equal. Analysts simply do not yet have granular customs records showing whether U.S.-bound InP shipments face longer delays or higher denial rates than those headed to Japan or the EU.

Given this, the most defensible assessment is that China has built a lever over a chokepoint material with global significance and is currently using it to create broad supply-chain friction rather than an overt U.S.-only blockade. The regime clearly raises the marginal cost and risk of expanding AI photonics capacity everywhere, including in the United States. Whether enforcement patterns quietly tilt against U.S.-linked orders is a matter for further empirical work, not assertion.

The Second Front: Scrutiny of Indium Metal and Future Expansion Risk

While formal controls currently focus on InP and its precursors, buyers are increasingly concerned about the upstream metal. Reuters’ June 19 piece and its derivatives report that China is “stepping up scrutiny of indium exports,” with customs demanding end-user information from a European buyer and stretching clearance times for a North American purchaser. Here again, no shipments have been blocked, but the pattern mirrors an earlier prelude to formal controls on gallium and germanium: informal tightening first, list-based restrictions later.

This is the second front of the squeeze. Indium metal is critical not only for InP but also for other optoelectronic and display technologies. If Beijing eventually moves from scrutiny to formal listing, it would extend licensing leverage further upstream, making it harder for non-Chinese firms to bypass InP bottlenecks by synthesizing or sourcing the compound elsewhere. Even without formal controls, the knowledge that 70 percent of global supply sits under a customs authority willing to ask more questions and take more time alters risk calculus for AI infrastructure planners.

That is why U.S. and allied policy responses have started to emerge: Pentagon stockpiling of indium, G7 discussions of critical-mineral alliances, and early investment in alternative photonics materials such as thin-film lithium niobate and organic electro-optic polymers. These are long-lead strategies; new refining capacity and new photonic platforms cannot be conjured in a year. But they are a recognition that InP controls are not a transient nuisance—they are a signal of a more granular choke-point toolkit that Beijing intends to keep using.

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

How Much Control Does China Really Have Over America’s AI Trajectory?

Given the structural facts—a concentrated indium supply, a formal licensing regime on InP, and visible price escalation—the temptation is to declare that China now has a chokehold on U.S. AI infrastructure. The reality is more nuanced. On the one hand, there is no evidence of a complete cutoff; shipments continue, and buyers can still obtain material, albeit more slowly and at higher prices. Alternative materials and architectures are under active development, from silicon photonics with heterogeneous integration to non-linear polymers and domain-specific accelerators that reduce bandwidth pressure.

On the other hand, none of those alternatives is yet mature enough or scaled enough to displace InP in high-end data-center optics over the next few years. AI build-out plans through roughly 2030 were largely written on the assumption of abundant InP substrates and steadily falling cost-per-bit. Export controls have inverted that premise: they inject uncertainty into delivery schedules and push costs up just as AI workload demand explodes. That does not give Beijing a switch it can flip to “turn off” U.S. AI, but it does give it leverage over the tempo and economics of expansion.

The prudent interpretation, therefore, is that China’s two-front squeeze—the licensing of indium phosphide and the intensifying scrutiny of indium metal—is best understood as a strategic constraint, not a decisive lock. It complicates American and allied AI infrastructure plans, forces diversification of materials and supply routes, and creates a bargaining chip that can be deployed in broader technology and trade negotiations. But its ultimate impact on U.S. AI capacity will be determined as much by how quickly alternative photonics ecosystems mature and how effectively policymakers manage concentration risk as by decisions in Beijing alone.