Chapter 2
Raw Materials & Industrial Inputs
Chapter 2: Raw Materials & Industrial Inputs
2.1 Overview
Beneath the semiconductor materials layer (Chapter 3) and the data center infrastructure layers (Chapters 13-16) sits a foundation of raw materials and industrial inputs that the entire AI buildout depends on. These are not chip-specific; they are the metals, minerals, gases, and industrial commodities consumed by every layer of the supply chain. Copper wires every data center. Rare earth elements power the magnets in every wafer stage, every cooling fan, every disk drive. Neon gas fills the lasers that pattern every advanced chip. Gallium and germanium are precursors to the compound semiconductors that enable high-frequency communication.
This chapter differs from the others in that few companies here sit at monopoly-like chokepoints in the way ASML or TSMC do. Instead, the risks are geographic. China dominates production of rare earth elements (~60-70% of mining, ~90% of processing), gallium (~98% of primary supply), germanium (~60%), polysilicon (75-85% of global output), and metallurgical silicon (75-80% by 2024, up from ~64% in 2020). In October 2025, China significantly expanded its export controls on rare earths, adding five new elements and restricting equipment for processing 12. These are not hypothetical risks; they are actively deployed geopolitical weapons.
There is one raw material chokepoint that rivals anything in the semiconductor layers: high-purity quartz from Spruce Pine, North Carolina. Two mines in this district (operated by Sibelco and The Quartz Corp) supply an estimated 70-90% of the world’s semiconductor-grade high-purity quartz. This quartz is processed into the fused silica crucibles used for Czochralski silicon crystal growth, the process by which every silicon wafer begins. Sibelco’s IOTA facility produces quartz at 99.9992% purity 11. No other known natural deposit has the geological properties to produce quartz at this grade. Synthetic silica (produced from silicon tetrachloride via flame hydrolysis) can achieve comparable purity and is used in some EUV-grade applications, but at significantly higher cost and energy intensity; scaling synthetic production to replace Spruce Pine volumes would require years of capital investment and process qualification. Hurricane Helene’s September 2024 landfall in western North Carolina temporarily disrupted operations at Spruce Pine, causing immediate concern across the semiconductor supply chain. This is a single geographic point of failure with no near-term substitutable alternative at current cost structures.
The AI buildout amplifies demand for these materials because data centers are extraordinarily material-intensive. Each megawatt of data center capacity requires approximately 27-33 tonnes of copper for wiring and cooling 34. S&P Global projects copper demand for data centers will grow from 1.1 million metric tons in 2025 to 2.5 million by 2040 5. A projected supply deficit could reach 10 million metric tons by 2040 5. Copper prices hit a record $13,200 per metric ton ($6/lb) in January 2026, and analysts project continued strength through late 2026 67.
2.2 Market Sizing & Growth
Copper: Global copper market approximately $250-280 billion (at ~$11,000/tonne and ~25 million tonnes annual production). Data center copper demand projected from 1.1M tonnes (2025) to 2.5M tonnes (2040) 5. AI training data center copper demand will account for 58% of total data center copper demand by 2030 5. Projected 6 million tonne cumulative supply gap by 2035 37. Wood Mackenzie forecasts a 304,000-tonne refined copper deficit for 2025, widening in 2026 4.
Rare earth elements (REEs): Global market approximately $10-12 billion. China produces ~60-70% of REE mining output and controls ~90% of processing/refining 18. Used in permanent magnets (neodymium, dysprosium) essential for wafer stage positioning in ASML lithography tools, cooling fan motors, disk drive actuators, and EV motors. China expanded export controls on 12+ rare earth elements in October 2025 12.
Specialty gases (neon, helium, krypton, xenon): Neon is a critical buffer gas in DUV excimer lasers used in semiconductor lithography. Prior to Russia’s invasion of Ukraine, Ukraine supplied an estimated 45-54% of global semiconductor-grade neon (estimates vary by source; some cite higher figures for total neon output) and approximately 40% of krypton, another lithography gas 8. Prices surged over 500% in Q2 2022 following the invasion, with production facilities in Odesa and Mariupol disrupted. Post-invasion diversification has reduced this dependency (Intel, TSMC, and others have qualified alternative suppliers), but the industry remains vulnerable to supply shocks. As of 2023, Ukraine’s neon exports remained 40% below pre-war levels. Helium is used in semiconductor fab cooling and leak testing; the global helium supply has experienced recurring shortages.
Gallium and germanium: China controls ~98% of primary gallium production and ~60% of germanium 9. Both are critical for compound semiconductors (gallium arsenide, gallium nitride) used in 5G RF chips, power electronics, and photonics. China imposed export controls on gallium and germanium in July 2023, tightening further in 2024-2025 9. A notable diversification effort: Indium Corporation (USA, private) and Rio Tinto (RIO, ASX/LSE) announced a gallium extraction partnership in May 2025, successfully extracting gallium from Rio Tinto’s Vaudreuil alumina refinery in Quebec. The demonstration plant targets 3.5 tonnes per year, potentially scaling to 40 tonnes (5-10% of global gallium production). This is one of the first serious Western attempts to break China’s gallium monopoly.
Polysilicon: China dominates global polysilicon production, with capacity of 3.25 million metric tons by 2024 and an estimated 75-85% of global output depending on methodology 10. Polysilicon is the raw material for solar cells and, via further processing, the ultra-pure electronic-grade silicon used in semiconductor wafers. A US Commerce Department Section 232 investigation into polysilicon was launched in 2025 10. Forced labor concerns in Xinjiang’s polysilicon supply chain add regulatory risk.
2.3 Supply Chain Flowchart
RAW MATERIAL EXTRACTION
|
|---> COPPER (Chile, Peru, DRC, China, US, Australia)
| Mining: Freeport-McMoRan, BHP, Rio Tinto, Codelco (state),
| Southern Copper, Glencore, First Quantum
| Refining: China (~40% of global refining), Chile, Japan
| Uses: Power wiring (27-33 tonnes/MW), PCBs, substrates,
| cables (Chapter 14), cooling, connectors
|
|---> RARE EARTH ELEMENTS (China ~60-70%, Myanmar, Australia, US)
| Mining: Northern Rare Earth, Lynas, MP Materials, Iluka, Energy Fuels
| Processing: China ~90% of global REE processing
| Uses: NdFeB magnets (wafer stages, motors), CeO2 polishing (CMP)
|
|---> GALLIUM & GERMANIUM (China ~98% Ga, ~60% Ge)
| Gallium: byproduct of aluminum refining (China dominant)
| Germanium: byproduct of zinc refining (China, Belgium)
| Uses: GaAs/GaN compound semis, 5G RF, photonics, fiber optics
|
|---> SPECIALTY GASES (Various global suppliers)
| Neon: Air Liquide, Linde, Iceblick (Ukraine, diversifying)
| Helium: US, Qatar, Australia, Russia
| Fluorine gases (NF3, SF6, CF4): Linde, Air Products
| Uses: DUV laser gas (neon), fab process gases, cooling (He)
|
|---> POLYSILICON (China ~80%, Germany, US, South Korea)
| Producers: GCL Technology, Tongwei, Daqo, Wacker Chemie,
| OCI, Hemlock Semiconductor
| Uses: Solar cells (majority), electronic-grade Si --> wafers
|
+---> OTHER METALS & MATERIALS
Aluminum: data center structures, heat sinks
Steel/rebar: physical construction (Chapter 17)
Tin: solder for packaging (Chapter 9)
Cobalt: battery cathodes (backup power)
Lithium: battery storage at data centers
|
v
DOWNSTREAM: Chapters 3-19 (every layer of the buildout)2.4 Key Companies
2.4.1 Copper Producers
| Company | Ticker | Exchange | Approx. Mkt Cap | Role | Key Metric |
|---|---|---|---|---|---|
| Freeport-McMoRan | FCX | NYSE | ~$70.0B | World’s largest publicly traded copper producer | Grasberg mine (Indonesia); significant AI-adjacent demand growth |
| BHP | BHP | ASX / NYSE | ~$215B | Major diversified miner; copper a key growth segment | Copper production ~1.7M tonnes/year; OZ Minerals acquisition expanded copper portfolio |
| Rio Tinto | RIO | LSE / NYSE | ~$171B | Diversified miner with growing copper exposure | Oyu Tolgoi (Mongolia) ramping; Resolution Copper (Arizona, pending) |
| Southern Copper | SCCO | NYSE | ~$153B | Major copper producer (Peru, Mexico); Grupo Mexico subsidiary | One of world’s largest copper reserve bases |
| Glencore | GLEN | LSE | ~$60.0B | Trader and miner; copper, cobalt, zinc | Major copper producer and trader; recycling operations |
| First Quantum Minerals | FM | TSX | ~$15.0B | Mid-cap copper producer; Cobre Panama mine (now restarting) | High-growth copper exposure |
| Codelco | Private | State-owned (Chile) | N/A | World’s largest copper producer; Chilean state-owned | ~1.6M tonnes/year; declining grades driving capex needs |
2.4.2 Rare Earth Producers
| Company | Ticker | Exchange | Approx. Mkt Cap | Role | Key Metric |
|---|---|---|---|---|---|
| China Northern Rare Earth | 600111 | Shanghai SSE | ~$4.9B | World’s largest REE producer; Chinese state-affiliated | Controls ~40% of China’s REE quota |
| Lynas Rare Earths | LYC | ASX | ~$12.8B | Largest non-Chinese REE producer; Mt Weld mine (Australia) | Only integrated mine-to-magnets operation outside China; processing in Malaysia and Kalgoorlie |
| MP Materials | MP | NYSE | ~$12.0B | Only active REE mine in the US (Mountain Pass, California) | Restarted processing in 2023; building magnetics facility in Fort Worth, TX |
| Iluka Resources | ILU | ASX | ~$2.4B | Developing Eneabba rare earths refinery (Australia); also mineral sands | Australian government co-funded; targeting 2026 production |
| Energy Fuels | UUUU | NYSE | ~$5.3B | Uranium producer pivoting to REE processing at White Mesa (Utah) | Small-scale but US-based REE processing capability |
2.4.3 Specialty Gas Suppliers
| Company | Ticker | Exchange | Approx. Mkt Cap | Role | Key Metric |
|---|---|---|---|---|---|
| Linde | LIN | NASDAQ / XETRA | ~$220B | World’s largest industrial gas company; supplies neon, helium, process gases | Critical supplier to semiconductor fabs worldwide; electronic specialty gases |
| Air Liquide | AI | Euronext Paris | ~$100B | #2 industrial gas; supplies ultra-pure gases for semiconductor fabs | Strong position in electronics gases; expanding in Asia |
| Air Products | APD | NYSE | ~$65.0B | Industrial gas; hydrogen, helium, specialty gases | Hydrogen for data center fuel cells (emerging); helium supply |
| Nippon Sanso (Taiyo Nippon Sanso) | 4091 | TSE | ~$5.4B | Japanese industrial gas; strong in Asian semiconductor supply | Subsidiary of Mitsubishi Chemical; key supplier to Japanese and Taiwanese fabs |
2.4.4 Polysilicon Producers
| Company | Ticker | Exchange | Approx. Mkt Cap | Role | Key Metric |
|---|---|---|---|---|---|
| Tongwei | 600438 | Shanghai SSE | ~$15.0B | World’s largest polysilicon producer (China) | Dominant in solar-grade polysilicon; not directly in electronic-grade |
| GCL Technology Holdings (fmr. GCL-Poly) | 3800 | HKEX | ~$3.0B | Major Chinese polysilicon producer | Rebranded Apr 2022. Granular polysilicon technology; primarily solar |
| Wacker Chemie | WCH | XETRA | ~$10.0B | German chemical company; electronic-grade polysilicon | Supplies electronic-grade polysilicon to wafer makers (Shin-Etsu, SUMCO) via its Nurnberg plant; also solar-grade |
| Hemlock Semiconductor | Private | Private (Dow/Shin-Etsu/Mitsubishi JV) | Private | US-based electronic-grade polysilicon producer (Michigan) | Critical US-based source for semiconductor-grade silicon |
| Daqo New Energy | DQ | NYSE | ~$1.3B | Chinese polysilicon producer | Solar-grade; Xinjiang operations raise forced labor concerns |
2.5 Bottleneck Analysis
Copper (HIGH, tightening): Copper is not a monopoly, but the gap between supply growth and demand growth is widening. S&P Global projects a potential 10 million tonne supply deficit by 2040 5. Each megawatt of data center capacity requires 27-33 tonnes 34. As data centers scale from hundreds of megawatts to gigawatt-class campuses, copper consumption becomes a binding physical constraint. New mines take 10-15 years to permit and develop. Copper hit record prices in early 2026, and US 50% tariffs on copper imports (August 2025) are reshaping trade flows 3. This is not a “someday” bottleneck; it is already manifesting in higher prices and longer lead times for electrical infrastructure.
Chinese-controlled materials (SEVERE for geopolitical scenarios): China’s dominance in rare earths (processing), gallium, germanium, and polysilicon creates a structural vulnerability. The October 2025 expansion of rare earth export controls signals Beijing’s willingness to use material supply as leverage 12. In a scenario where US-China tensions escalate further, disruption to these materials could affect the production of magnets for ASML wafer stages, compound semiconductors for 5G/photonics, and process chemicals for fabs. Mitigation efforts (Lynas in Australia, MP Materials in the US, Japanese stockpiling programs) are underway but years from full effect.
Neon and specialty gases (MODERATE, improving): The post-Ukraine diversification of neon supply has reduced the concentration risk that existed pre-2022. However, specialty gas supply remains fragile, with limited producers and long qualification cycles at fabs. CSIS notes that neon is “especially instructive” as a case study of concentrated supply chains 8.
Tungsten hexafluoride / WF6 (HIGH, emerging crisis): WF6 is the precursor gas for tungsten metallization at advanced semiconductor nodes (3D NAND, logic interconnect). SK Specialty (South Korea, subsidiary of SK Group) is the world’s largest WF6 producer, with Japanese suppliers providing approximately 25% of global capacity. In 2026, SK Specialty and Japanese suppliers announced 70-90% price increases driven by Chinese tungsten export tightening. Japanese producers (Josen and others) are cutting output due to raw material constraints. China controls approximately 80% of primary tungsten supply and tightened export controls in early 2025 12. This creates a supply chain vulnerability analogous to the 2022 neon crisis: a critical process gas with concentrated production, subject to geopolitical disruption. Every advanced DRAM, NAND, and logic chip requires tungsten metallization. Unlike neon, where post-Ukraine diversification is underway, tungsten/WF6 diversification has barely begun. The primary Western diversification play is Almonty Industries (AII, TSX, ~CAD $5B 13), which restarted production at the Sangdong tungsten mine in South Korea (December 2025) after 30+ years of closure. At full capacity, Sangdong can supply approximately 40% of non-China tungsten globally. Almonty is redomiciling to Delaware in 2026 to strengthen US positioning and has secured a strategic partnership with American Defense International. This is the only significant non-China tungsten expansion underway.
Water supply for semiconductor fabrication (HIGH, structural): A modern semiconductor fab consumes 20-38 million liters of ultrapure water (UPW) per day, comparable to a small city. Over 40% of existing fabs and more than 40% of fabs under construction post-2021 are located in high or extremely high water-stress zones 14. The UPW market is $11.5B (2025), growing at 11.8% CAGR. Key suppliers include Xylem (XYL, NYSE, ~$31B 15), Pentair (PNR, NYSE, ~$14.6B 15), and Pall Corporation (subsidiary of Danaher, DHR). TSMC’s Arizona fab includes a 15-acre water recycling plant targeting 90% wastewater recovery. Water stress is an emerging site-selection constraint that will increasingly limit where new fabs and data centers can be built. This chapter’s geographic concentration analysis should be read jointly with the water constraints on power and cooling (Chapters 13-15).
Polysilicon (LOW for semiconductors directly): Most polysilicon goes to solar, and the semiconductor industry uses a small fraction of total polysilicon output (electronic-grade is a niche within the polysilicon market). However, the concentration of even electronic-grade production in a few facilities (Wacker in Germany, Hemlock in the US) means a single-site disruption could affect wafer supply upstream of everything in Chapters 03-09.
2.6 Risks
China weaponizing material supply: The most significant risk in this chapter. China’s October 2025 rare earth export controls, its 2023 gallium/germanium restrictions, and its dominance in polysilicon processing give Beijing leverage over the entire semiconductor supply chain. A full embargo would not halt chip production immediately (stockpiles exist, alternative sources are developing), but would create severe disruptions lasting 12-24 months and drive massive price spikes.
Copper as a physical limit on data center scale: If the AI buildout proceeds as planned (multiple gigawatt-scale campuses, each requiring tens of thousands of tonnes of copper), the material simply may not be available at any price. Unlike chip shortages, which can be resolved by building more fabs, copper supply is constrained by geology, permitting, and physics. New mines are getting harder to develop (lower grades, deeper deposits, stricter environmental regulations). This could force data center architects to explore aluminum wiring (lower conductivity), recycled copper (limited supply), or fundamentally different power distribution architectures.
Substitution and recycling: For many materials, substitutes exist but with performance penalties. Aluminum is already standard for facility-level power distribution (busbars, overhead lines) where its approximately 60% conductivity relative to copper is offset by lower weight and cost. The copper constraint is binding specifically at in-rack density, where bend radius, volume, and thermal conductivity requirements rule out aluminum. Ferrite magnets can replace rare-earth magnets but with a fraction of the magnetic strength. These substitutions are possible but degrade performance, increase system size, or both.
Oversupply at mature-node materials: If the AI buildout slows, demand for copper, rare earths, and other materials would soften, potentially causing price crashes similar to past commodity cycles. Miners who invested heavily in expansion would face margin compression.
First principles check: Does the material scarcity thesis hold? Partially. Copper scarcity is real and physics-constrained (you cannot substitute away from conductivity for power wiring). Rare earth scarcity is politically constrained (the resources exist but processing is concentrated in China by policy, not geology). Gas scarcity is operationally constrained (diversification is happening but slowly). The risks are genuine but uneven across materials.