AI competition opens a supercycle for rare metals: Are tin, indium, and hafnium facing a historic revaluation?

```

AI hardware investment is expanding from “buying chips” to a longer value chain: server systems, high-speed network components, data center power supply, and high-heat-density cooling facilities are all starting to consume more basic raw materials. For minor metals, the key change is not just a new concept, but that downstream consumption is entering measurable stages.

Dongwu Securities analyst Liu Yiting stated in a June 20 research report: “Global AI capital expenditures are entering a phase of nonlinear acceleration.” Capital deployment is gradually expanding from single chip products to cover servers, high-speed networks, power infrastructure, and cooling facilities, “bringing upstream basic materials a demand dividend.”

Among them, tin, indium, and hafnium correspond to three bottlenecks in AI hardware upgrades: tin is used for PCB electroplating and SMT soldering; indium, in the form of indium phosphide, is used for high-speed optical communication; hafnium, as a high-K gate dielectric material, supports continued micro-shrinking of advanced processes. According to calculations, tin consumption in PCBs is expected to increase by 49,000 tons from 2026 to 2030; AI data center demand for indium in indium phosphide may rise from 19 tons in 2025 to 419 tons in 2030; global hafnium demand could grow from 100 tons in 2024 to 142 tons in 2030.

The common point for the three metals lies on the supply side: tin is affected by resource depletion, Indonesian policy, Myanmar’s production resumption falling short of expectations, and changes in trade flows; indium is constrained by zinc mining and smelting operations; hafnium is hindered by zirconium-hafnium separation, environmental protection, economics, and geopolitical disturbances. The core logic behind price hikes is rising demand stacked on unstable supply.

Money for AI hardware is no longer spent only on GPUs

Capital expenditure is a leading indicator for this chain. In 2026, capital expenditures by Microsoft, Google, Amazon, and Meta will collectively peak at $725 billion. From January to September 2025, investment related to AI in the US contributed 39% to actual GDP growth, higher than the 36% during the Internet bubble of 2000.

Hardware upgrades are concentrating in four directions: computing power density, memory bandwidth, interconnect speed, and energy efficiency. Chips are only one link in the chain. AI server PCB layers have increased from the traditional 8–24 layers to 28–46 layers, with some projects using 56 layers. High-speed optical modules are evolving from 800G to 1.6T and 3.2T, and internal interconnect bottlenecks in data centers are becoming more pronounced. Advanced process nodes are pushing forward, and traditional silicon dioxide gate dielectrics are approaching physical limits.

Minor metals have come into view not because of rarity alone, but because they are stuck right at these upgrade points.

Incremental Tin Demand is in PCBs, While Supply Seldom Keeps Up

Tin plays the role of soldering and connection in electronics. The expansion of AI servers, high-end PCBs, and advanced packaging all increase tin consumption.

Estimates divide tin consumption into two parts: tin used for PCB electroplating and tin used for SMT surface mounting. In the PCB electroplating segment, HDI board’s tin usage is about 40.19g/sq.m, while multilayer boards use about 12.84g/sq.m; HDI’s per-area usage is more than three times that of multilayer boards. SMT packaging uses about 294.22g/sq.m. Altogether, tin usage for PCB electroplating + SMT is about 318g/sq.m.

According to Prismark, global PCB shipments will reach 663 million sq.m in 2030, with a compound annual growth rate (CAGR) of about 6.7% during 2026–2030. This means global PCB tin usage will rise from 163,000 tons in 2026 to 212,000 tons in 2030, an increase of 49,000 tons over four years, with a CAGR of 6.9%. Assuming a global tin consumption of 380,000 tons in 2025, the growth elasticity from PCBs alone is 12.3%.

The issue is supply.

Global proven tin reserves total about 6 million tons, with a static reserve-production ratio of 20.7 years, lower than copper, nickel, cobalt and other industrial metals. Between 2015 and 2025, tin prices soared, but global mine output only grew from 289,000 tons to 290,000 tons, essentially zero growth in a decade. China’s production actually dropped from 110,000 tons to 71,000 tons, a CAGR of -4.3%.

Indonesia is an important variable. In 2025, it accounts for 21% of global mine supply, but recent years have seen frequent policy shifts—including mine approvals, illegal mining control, progressive royalties, and minimum benchmark prices—with huge export swings. Myanmar was also a major supplier, with 17% of global production in 2018, but output dropped to 12,000 tons by 2025 due to resource depletion and mining bans. Even if the Wa region resumes production in the second half of 2025, China’s import of tin ore from Myanmar by April 2026 will only recover to about 1,300 metal tons, still below the pre-ban level of ~2,200 tons/month.

South American trade flows are also changing. Peru, Brazil and Bolivia will together produce 76,000 tons in 2025, accounting for 26% of the global total. For Peru, the main tin ingot export destination is the US, with Bolivia shipping mostly to the Netherlands, UK, and US. The US is accelerating its tin supply chain layout, which may further absorb South American raw materials.

Overall, Dongwu Securities believes tin metal will face strong demand growth and disruptive supply shocks in the next 3–4 years, with significant upward price drivers. On one hand, global AI capital expenditures are accelerating, and the expansion of PCBs and other hardware will drive solid incremental demand for tin. On the other, global tin supply is highly concentrated and unstable, with many influencing factors.

Indium’s Upside is from Indium Phosphide, but Production Can’t Expand at Will

Indium’s traditional demand is mainly for ITO targets, accounting for about 70%, with downstream uses in LCDs and flat-panel displays; electronics, solder, and alloys each make up about 12%. Global refined indium consumption is predicted to be 2,316 tons in 2025, rising to 2,510 in 2026 and 2,813 in 2027.

The new variable is optical communications. Inside AI data centers, GPUs require high-speed data exchange. In mega-scale cluster models, data shuttling between chips consumes over 90% of system energy; after rate increases, copper interconnects are only effective for a few centimeters. Data transfer rates are moving from 100G/lane to 200G/lane and pushing further to 400G/lane, making optical interconnects the more realistic direction.

Indium phosphide’s strengths are clear: it is a direct-bandgap semiconductor with a bandgap energy of about 1.34eV, matching the low-loss window of 1310nm/1550nm in fiber optics; its electron mobility is more than ten times that of silicon, supporting 100GHz+ modulation. The laser chips in high-speed optical modules use indium phosphide as a core material.

Calculated, a 4-inch indium phosphide substrate actually consumes about 32.2 grams of indium per wafer. In 2025, AI data centers’ demand corresponds to about 600,000 wafers, equating to 19.3 tons of indium; by 2030, demand may reach 13 million wafers, or 419 tons of indium—a more than 22-fold increase. Relative to 2025’s global indium demand, this alone could bring over 20% growth.

The hard constraint on supply is that indium mostly comes as a by-product from lead-zinc polymetallic ore. 81.2% of global reserves are from these ores; primary indium comes mainly from zinc smelting residues. In other words, even if the price of indium rises, you can’t just open an ‘indium mine’ to expand supply quickly.

In recent years, zinc concentrate processing fees have fallen, disincentivizing smelters; refined zinc capacity utilization has dropped to its lowest point in five years, further limiting primary indium supply. At the same time, China in Feb. 2025 imposed export controls on indium phosphide, trimethylindium, triethylindium, and related technical data. Stocks are dropping too: according to Zhonglian Gold Platform, inventories fell from about 488.8 tons at the start of 2025 to 273.8 tons by Jan. 28, 2026.

As of June 11, 2026, domestic refined indium price was 4.7 million yuan/ton, up 58% from the start of the year.

Hafnium’s Value is in Advanced Nodes, With Difficulties in Separation & Economic Scaling

Hafnium’s traditional demand is in nuclear energy and high-temperature alloys. In the consumption structure, nuclear energy accounts for 45%, high-temp alloys/aerospace 35%, semiconductors/electronics 10%.

Changes on the semiconductor side stem from node shrinking. At 65nm and below, traditional silicon dioxide gates are too thin—quantum tunneling increases gate leakage current, pressuring chip power and reliability. Hafnium oxide’s dielectric constant is about 18–25, far higher than silicon dioxide’s 3.9, allowing increased physical thickness for the same equivalent oxide thickness, reducing leakage.

After Intel introduced hafnium-based high-K in 45nm nodes replacing silicon dioxide, NMOS gate leakage dropped 25x, PMOS over 1000x. As 3nm, 2nm nodes push from FinFET to GAA, high-K dielectric demand will continue to rise.

Demand-wise, global hafnium need is set to rise from 100 tons in 2024 to 142 tons in 2030. Semiconductor uses from 40 tons to 64 tons, accounting for nearly half the increase; high-temp alloys from 45 tons to 60 tons; nuclear from 15 tons to 18 tons.

Hafnium supply is even trickier than demand. It is mainly a byproduct of separating nuclear-grade sponge zirconium, with total global capacity over 10,000 tons/year, but actual annual output is 6,000–7,000 tons, corresponding to about 100 tons of sponge hafnium, mostly from the US, France, Russia, and China.

Zirconium-hafnium separation is highly difficult. Their physical/chemical properties are very close; in nature and zirconium products, hafnium is only 1%–3% of the total zirconium-hafnium amount. Existing processes use toxic solvents or high-concentration acids, raising environmental and equipment corrosion issues. Expanding production is not economical: the two US producers could theoretically double hafnium output, but each would also generate about 2,000 tons/year of hafnium-depleted zirconium, which is hard to sell and closes the expansion loop.

Geopolitical disruptions further push prices higher. After the 2022 Russia–Ukraine war, Russia's sponge hafnium export stopped, pushing international prices from $1,200–1,400/kg up to $4,500–5,000/kg. China included hafnium in dual-use item controls at the end of 2024, and exports of unwrought hafnium, hafnium scrap, and powder fell 22% YoY to 20.2 tons in 2025.

Domestic 4N-grade hafnium oxide price has also surged. Early 2022 it was about 4.5 million yuan/ton, and by June 16, 2026 had risen to 9.5 million yuan/ton, an increase of 111%.

Risk Warning and DisclaimerThe market carries risks and investments must be prudent. This article does not constitute personal investment advice, nor does it take into account an individual user's specific investment objectives, financial situation, or needs. Users should consider whether any opinions, views, or conclusions herein fit their own situation. Invest accordingly at your own risk. ```