What Happens to AI Infrastructure If China's Mineral Truce Ends?

Why the November dates matter

The mineral truce sits on top of several overlapping Chinese export-control regimes rather than replacing them.

China began requiring licenses for gallium and germanium exports in 2023. It tightened restrictions for the United States in December 2024, stating that gallium, germanium, antimony and certain superhard materials would in principle not be approved for export to the U.S., while imposing more stringent end-user and end-use review on graphite. The November 2025 agreement suspended that U.S.-specific restriction until November 27, 2026, but did not eliminate the broader licensing requirements. [2]

A second layer arrived in April 2025, when China placed samarium, gadolinium, terbium, dysprosium, lutetium, scandium and yttrium-related items under export controls. Those controls were not part of the October package that was later suspended. [4] In other words, the mineral relationship is already partially restricted even while the later measures remain paused.

Then came October 9, 2025. China announced controls covering holmium, erbium, thulium, europium and ytterbium-related materials, rare-earth production equipment and inputs, rare-earth technology, lithium-ion battery technology, artificial graphite anode materials and selected superhard materials. It also announced an extraterritorial rare-earth regime that could require Chinese licenses for certain foreign-made products containing controlled Chinese-origin rare-earth content or produced using specified Chinese technology. Those October measures were suspended on November 7 through November 10, 2026. [1]

The sequence creates a useful dated dependency map:

The International Energy Agency's 2026 critical-minerals assessment explains why these dates deserve more attention than their relatively small commodity markets would suggest. China accounts for more than 90% of refining for gallium, graphite and rare earths, according to the IEA, while germanium has relatively limited substitution possibilities in high-performance fiber optics. The IEA also found that export controls have produced large regional price gaps: European gallium and heavy rare-earth prices were around five times Chinese domestic prices, while germanium prices were almost three times higher. [5]

This is the central investment mechanism. A small mineral market can become a bottleneck for an enormous downstream capital cycle because the value of the mineral is tiny relative to the value of the equipment that cannot be completed without it.

The first dependency chain is optics, not GPUs

The most direct AI infrastructure exposure runs through optical networking.

Training clusters increasingly require massive amounts of high-speed communication between accelerators, switches and separate data-center buildings. Companies such as Coherent and Lumentum are scaling semiconductor lasers and other photonic components specifically for 800G, 1.6T and future AI networking architectures.

Several of the controlled minerals sit directly inside that chain.

Gallium is used in gallium arsenide semiconductor devices. Coherent describes its short-reach AI and machine-learning optical products as using a GaAs VCSEL technology platform, while its current datacenter portfolio also contains indium phosphide lasers and other compound-semiconductor devices. [6] Coherent's 2026 annual report says the company manufactures engineered materials and optoelectronic devices based on GaAs, InP, GaSb and SiC and uses rare-earth materials in some production processes. The company says it maintains buffer inventory and develops multiple sources to reduce raw-material risk. [7]

That does not establish that a particular Coherent product uses Chinese gallium. It does establish something more useful for risk analysis: one of the major suppliers of AI optical hardware operates on material platforms whose elemental inputs sit inside highly concentrated and politically controlled markets.

Lumentum provides an even clearer disclosure. Its 2026 Form 10-K specifically identifies Chinese restrictions on exports of indium, gallium, germanium and other rare-earth metals and critical minerals as a business risk. It says China's controls have limited access to certain materials, affected its substrate supply chain globally and could increase manufacturing costs, reduce margins or limit production. Lumentum also says some raw materials and components come from limited or sole sources. [8]

At the same time, Lumentum's AI data-center products are built around its indium phosphide semiconductor platform, including EMLs, continuous-wave lasers and devices for co-packaged optics. The company opened a new North Carolina manufacturing site in 2026 to expand U.S. production of InP-based devices for large AI data centers. [9]

The exposure is therefore not simply "China controls a mineral and Nvidia needs minerals." The relevant chain is more specific:

That is where the first meaningful corporate relationships begin to appear.

Coherent has expanded its optical-networking relationship with Nvidia while investing in U.S. InP manufacturing capacity for AI infrastructure. [10] Lumentum is expanding U.S. InP manufacturing for AI data centers. Both companies still operate within global upstream material chains that their own disclosures acknowledge can be affected by Chinese mineral policy.

There is another revealing connection one step further upstream. In June 2026, AXT agreed to develop and supply 6-inch InP wafer substrates to Coherent, with manufacturing capacity being expanded at AXT's Beijing facility. The following month, AXT entered a long-term InP substrate capacity agreement with Lumentum. [11]

Indium is subject to a separate Chinese licensing regime rather than the November gallium and germanium suspension discussed here. The AXT relationships therefore should not be treated as direct evidence that expiration of the November truce would stop Coherent or Lumentum production. They do show how deeply China remains embedded in the substrate layer of the AI photonics supply chain even after manufacturers build additional fabrication capacity in the United States.

Germanium creates another route into the same networking buildout.

The U.S. Geological Survey says fiber optics are the largest U.S. end use for germanium. Germanium dioxide and germanium tetrachloride are used to manufacture fiber-optic glass for data networking and telecommunications. USGS estimated that U.S. imports of germanium metal fell 67% in 2025 following China's U.S. export ban. [12]

Corning's optical-fiber documentation explains the physical connection. Conventional optical fiber commonly uses a germania-doped silica core, although some ultra-low-loss fiber designs use different compositions. [13]

That mineral dependency now intersects directly with hyperscale AI spending. In January 2026, Meta and Corning announced a multiyear agreement worth up to $6 billion under which Corning will supply optical fiber, cable and connectivity products for Meta's U.S. data-center expansion. [14]

Again, this does not prove that the products supplied to Meta contain Chinese germanium. It identifies the relevant dependency surface. A renewed germanium restriction would first affect the availability and pricing of qualifying germanium feedstock. The transmission mechanism would then run through fiber and component manufacturers before reaching hyperscaler construction schedules.

This is why the optical layer deserves more attention than the headline GPU supply chain. The AI accelerator can be manufactured and still fail to become revenue-generating compute if the surrounding network cannot be built at the required speed.

The suspended rare-earth controls add another optical connection. China's October 2025 package included erbium and ytterbium-related materials. Coherent sells erbium-doped fiber amplifiers for data-center interconnects and produces specialty fibers doped with erbium, ytterbium and other rare earths. [15] If those controls return, the risk moves beyond conventional fiber and compound-semiconductor lasers into optical amplification and specialty-fiber manufacturing.

The relevant exposure is therefore layered rather than singular: germanium in fiber, gallium in compound semiconductors, rare earths in amplification and specialty optics, and China-linked substrate manufacturing further upstream.

Power conversion is the second direct AI bottleneck

The next important route runs through electricity rather than data.

The power required by AI racks is increasing quickly enough that semiconductor suppliers are redesigning the electrical architecture around the servers. Nvidia's emerging 800 VDC architecture is intended to support much higher rack-level power density, and wide-bandgap semiconductors are becoming increasingly important in that transition.

Gallium nitride is one of those technologies.

Infineon joined Nvidia's MGX AI Factory ecosystem in May 2026 to support next-generation power-delivery architectures. Infineon says its AI power portfolio combines silicon, silicon carbide and gallium nitride, and its 800 VDC reference designs use CoolGaN devices to achieve higher power density and efficiency. [16]

Navitas Semiconductor is pursuing the same layer of the stack. In June 2026 it announced participation in Nvidia's MGX ecosystem with an 800 VDC power-delivery board built around its GaN and SiC technologies. [17]

Neither relationship proves dependence on Chinese gallium. Semiconductor companies can source materials through multiple stages and countries, and the country in which a finished GaN wafer or device is fabricated is not necessarily the country in which the original gallium was refined.

The global material balance is nevertheless unusually concentrated. USGS reported that China accounted for 99% of global primary gallium production in 2024. [18] The IEA similarly identifies gallium as one of the minerals with the highest combination of supply concentration, strategic importance and limited alternative supply routes. [5]

That makes gallium policy relevant to an AI infrastructure layer whose importance is rising.

The critical point for investors is not that a gallium restriction would make every GaN device unavailable. It is that a renewed licensing shock could raise the cost of securing material, encourage inventory accumulation, extend qualification timelines for alternative sources and create a larger premium for non-Chinese supply precisely when data-center power architectures are moving toward greater use of GaN.

A temporary material disruption can therefore produce a capital-efficiency problem even when the hardware ultimately ships. If power-system components become more expensive or harder to qualify, a fixed hyperscaler capital budget buys fewer completed racks.

That distinction matters for interpreting AI capex. The industry can continue reporting enormous spending commitments while the composition of that spending shifts away from accelerators and toward increasingly expensive supporting infrastructure.

Graphite and battery controls create a related but somewhat slower channel.

The October 2025 package that is suspended until November 10 included controls on certain lithium-ion batteries, artificial graphite anode materials, battery-production equipment and associated technologies. [19] The IEA estimates that China controls more than 90% of graphite refining and warns that a complete disruption of battery-grade graphite trade could place more than $300 billion of downstream production outside China at risk. [5]

AI data centers increasingly require battery backup systems, uninterruptible power supplies and other energy-storage equipment to manage both reliability and rapidly changing electrical loads. The battery chain is unlikely to stop an AI cluster as quickly as a missing optical transceiver, because operators can use different backup architectures and battery chemistries. But the direction of the exposure is clear: a mineral conflict that begins in transportation batteries can migrate into the electrical infrastructure supporting AI.

The suspended superhard-material controls create another second-order route. China's October package covered specified synthetic-diamond powders, crystals, wire saws, grinding wheels and related manufacturing technology. [20] These products are used in precision cutting, grinding and semiconductor manufacturing processes. Their effect would therefore show up less as a direct data-center shortage and more as additional friction in the capital equipment and wafer-processing chain.

The truce ending would stack new restrictions on top of old ones

The biggest analytical mistake is treating November as an on-off switch.

Some Chinese mineral controls remain active today. China's April 2025 restrictions on terbium, dysprosium, yttrium and other medium and heavy rare earths continue to operate under licensing requirements. Reuters reported in May 2026 that shipments of several controlled heavy rare earths remained materially below earlier levels even while Washington and Beijing were discussing an extension of the broader mineral truce. [21]

The IEA's 2026 data show that the market has not returned to its pre-control equilibrium either. Gallium, germanium and heavy rare-earth prices outside China continue to carry substantial premiums over Chinese domestic prices. [5]

Expiration of the suspended rules would therefore not introduce mineral risk into an otherwise normal supply chain. It would add another layer of restrictions to a system that is already more constrained, more expensive and more politically conditional than it was three years ago.

For AI infrastructure, that suggests three different categories of exposure.

The first is direct material exposure. Gallium, germanium, graphite and certain rare earths enter products or production processes used in optical networking, power electronics, fiber and batteries.

The second is qualification exposure. High-performance semiconductor and optical components cannot necessarily replace one supplier, substrate or material source with another immediately. A replacement may require testing, device redesign, customer qualification or process changes. The economic bottleneck can therefore persist long after substitute material technically becomes available.

The third is deployment exposure. AI infrastructure is a system. Accelerators, memory, switches, transceivers, fiber, power supplies, backup power and cooling must all arrive together. A shortage in a lower-value component can delay monetization of a much more expensive GPU cluster.

That makes the investment transmission mechanism different from a simple commodity-price trade.

A gallium price spike does not translate mechanically into a comparable increase in the selling price of an AI server. The more important question is whether the material shortage changes component lead times, manufacturing yields, qualification requirements or the number of completed racks that can be delivered against a fixed capital budget.

What the corporate map says to watch

The most exposed relationships are not necessarily the companies with the largest mineral purchases. They are the relationships where a difficult-to-replace material sits upstream of a component that is itself becoming a bottleneck.

Corning and Meta are one example. Meta has committed to a multiyear optical-infrastructure agreement worth up to $6 billion while germanium remains a controlled mineral whose largest U.S. end use is fiber optics. [12][14]

Coherent and Nvidia are another. Coherent is expanding optical semiconductor capacity for AI infrastructure and Nvidia-related networking demand while operating across GaAs, InP and rare-earth-dependent optical technologies. [6][7][10]

Infineon and Nvidia create a similar power-chain relationship. Infineon's GaN technology is being designed into the emerging 800 VDC AI power ecosystem at the same time that gallium remains one of the world's most geographically concentrated refined materials. [16][18]

Lumentum is important for a different reason: its disclosure connects the abstract mineral issue to an actual operating company. Lumentum says Chinese mineral controls have already affected material access and its global substrate supply chain, even though the company reported that the trade environment had not produced a material impact on its consolidated financial statements as of June 27, 2026. [8]

That combination is more informative than a hypothetical supply-chain map. It shows that the vulnerability exists while also preventing an exaggerated conclusion that the current controls have already caused a financial crisis.

The same discipline should apply if the truce ends.

An expired suspension would not prove an AI hardware shortage. Export controls are administered through licenses, exemptions and end-use rules, and inventories differ by company. Governments could also negotiate another extension before November.

But the companies worth monitoring first would not be limited to Nvidia, AMD, Micron or the hyperscalers. The more sensitive indicators are likely to appear in optical-component lead times, compound-semiconductor substrate orders, specialty fiber, GaN power modules, battery systems and the smaller suppliers that sit between raw material and finished rack.

Conclusion

If China's mineral truce ends, AI infrastructure is unlikely to suffer a single dramatic "GPU shortage." The more realistic risk is a series of smaller bottlenecks that accumulate around the GPU.

Optical networking appears to be the most direct first-wave exposure. Gallium intersects with GaAs devices, germanium with fiber-optic manufacturing, and suspended controls on erbium and ytterbium reach into specialty fiber and optical amplification. Coherent, Lumentum and Corning sit in different parts of that chain, while their relationships with Nvidia, Meta and other hyperscale customers connect those material markets directly to AI deployment.

Power infrastructure is the next important layer. Infineon and Navitas are pushing GaN into the higher-voltage architectures being developed for increasingly dense AI racks. A renewed gallium shock would therefore arrive just as gallium-containing power devices are becoming more strategically useful.

Graphite, battery technology and superhard materials would create slower second-order effects through backup power and semiconductor manufacturing.

The broader conclusion is that China's mineral policy creates a risk to the conversion of AI capital spending into operating compute, not merely to the production of individual chips. A hyperscaler can secure its GPUs and still face a deployment bottleneck in optics, fiber, power or energy storage.

That is the dependency investors should watch as November approaches.

Methodology and limitations

This analysis maps publicly documented Chinese export-control measures to disclosed material uses and corporate supply relationships in AI infrastructure.

A named company's use of a material does not establish that the material was sourced from China. Global mineral production shares likewise do not prove direct supplier dependence. Companies may hold inventory, use intermediaries, maintain alternative suppliers or qualify substitutes that are not publicly disclosed.

The corporate relationships cited here represent selected examples, not complete supply chains. The analysis does not estimate stock-price sensitivity, earnings impact, pass-through rates or the probability that any specific Chinese control will be reinstated.

The supplied Altsets relationships were not used because they primarily describe semiconductor equipment, memory and battery relationships that do not establish a sufficiently direct connection to the mineral controls examined here.

Sources

1. "Decision to Suspend Implementation of Announcements No. 55, 56, 57 and 58 of 2025 and MOFCOM Announcements No. 61 and 62," Ministry of Commerce of the People's Republic of China and General Administration of Customs, November 7, 2025, mofcom.gov.cn.

2. "Announcement No. 72 of 2025 Adjusting Implementation of Announcement No. 46 of 2024," Ministry of Commerce of the People's Republic of China, November 9, 2025, exportcontrol.mofcom.gov.cn.

3. "MOFCOM response on continued delay of rare-earth export controls," Ministry of Commerce of the People's Republic of China, April 9, 2026, exportcontrol.mofcom.gov.cn.

4. "Announcement No. 18 of 2025 on Export Controls for Certain Medium and Heavy Rare-Earth Related Items," Ministry of Commerce of the People's Republic of China and General Administration of Customs, April 4, 2025, mofcom.gov.cn.

5. "Global Critical Minerals Outlook 2026, Executive Summary and Outlook," International Energy Agency, July 2026, iea.org.

6. "Datacom Innovation in the AI Era," Coherent Corp., coherent.com.

7. "Coherent Corp. Annual Report on Form 10-K for the fiscal year ended June 30, 2026," U.S. Securities and Exchange Commission, 2026, sec.gov.

8. "Lumentum Holdings Inc. Annual Report on Form 10-K for the fiscal year ended June 27, 2026," U.S. Securities and Exchange Commission, 2026, sec.gov.

9. "Lumentum Announces New U.S. Manufacturing Facility to Produce Advanced Lasers for the World's Largest AI Data Centers," Lumentum Holdings, March 26, 2026, lumentum.com.

10. "Department of Commerce CHIPS Program Announces Letter of Intent with Coherent to Expand Indium Phosphide Production," National Institute of Standards and Technology, June 16, 2026, nist.gov.

11. "AXT Inc. Quarterly Report on Form 10-Q for the quarter ended June 30, 2026," U.S. Securities and Exchange Commission, 2026, sec.gov.

12. "Mineral Commodity Summaries 2026: Germanium," U.S. Geological Survey, 2026, usgs.gov.

13. "Optical Fiber Technology Innovations" and Corning optical-fiber technical documentation, Corning Incorporated, corning.com.

14. "Corning and Meta Announce Multiyear, up to $6 Billion Agreement to Accelerate U.S. Data Center Buildout," Corning Incorporated, January 27, 2026, corning.com.

15. "Amplifiers" and "Laser and Amplifier Fibers," Coherent Corp., coherent.com.

16. "Infineon Joins NVIDIA's MGX AI Factory Ecosystem to Transform Power Delivery Architecture for Next-Generation AI Server Racks," Infineon Technologies, May 29, 2026, infineon.com.

17. "Navitas Collaborates with NVIDIA MGX Ecosystem to Accelerate 800 VDC AI Infrastructure," Navitas Semiconductor, June 3, 2026, navitassemi.com.

18. "China: Mineral Production Statistics," U.S. Geological Survey National Minerals Information Center, 2026, usgs.gov.

19. "Announcement No. 58 of 2025 on Export Controls for Lithium Batteries and Artificial Graphite Anode Material Related Items," Ministry of Commerce of the People's Republic of China and General Administration of Customs, October 9, 2025, mofcom.gov.cn.

20. "Announcement No. 55 of 2025 on Export Controls for Superhard Material Related Items," Ministry of Commerce of the People's Republic of China and General Administration of Customs, October 9, 2025, mofcom.gov.cn.

21. "Trump, Xi to weigh rare earth truce extension, but China's curbs still bite," Reuters, May 13, 2026, reuters.com.

How to Cite This

For research inquiries or data access: press@altsets.com

Go Deeper