AGI infrastructure is fundamentally a physical build-out. Every data center, every transformer, every mile of cable, every server rack starts as a hole in the ground. The recursive self-improvement loop demands exponentially more compute hardware, which demands exponentially more raw materials. This module maps the metals and mining supply chains feeding that build-out. Key dynamics:
| # | Narrow Sector | AGI Demand Impact | Supply Constrained? | # Companies | Verdict |
|---|---|---|---|---|---|
| 1 | Copper Miners | Very Strong | Very High | 7 | HIGH |
| 2 | Aluminum Producers | Strong | Moderate | 4 | HIGH |
| 3 | Silver Miners | Strong | High | 5 | HIGH |
| 4 | Uranium Miners | Very Strong | Very High | 5 | HIGH |
| 5 | Rare Earth Miners & Processors | Strong | Very High (geopolitical) | 3 | HIGH |
| 6 | Steel Producers | Strong | Low–Moderate | 5 | MEDIUM |
| 7 | Lithium Miners | Moderate–Strong | High | 4 | MEDIUM |
| 8 | Specialty Metals (Cobalt, Nickel, Tungsten, Tin) | Moderate–Strong | High | 4 | MEDIUM |
| 9 | Silicon Metal & Polysilicon | Moderate–Strong | Moderate | 3 | MEDIUM |
| 10 | Graphite & Carbon Materials | Moderate | High | 3 | MEDIUM |
| 11 | Mining Equipment Manufacturers | Strong | Low | 5 | HIGH |
| 12 | Commodity Traders & Diversified Miners | Strong | Low | 5 | HIGH |
| 13 | Mineral Processing & Refining | Strong | High | 4 | MEDIUM |
| 14 | Gold Miners | Weak | Moderate | 5 | LOW |
| 15 | Industrial Minerals (Aggregates, Cement, Sand) | Moderate | Low | 3 | LOW |
| 16 | Platinum Group Metals (PGMs) | Weak–Moderate | High | 3 | LOW |
Copper miners extract copper ore from open-pit and underground mines, process it into concentrate, and sell refined copper cathode or concentrate to fabricators. Copper is the most conductive practical metal and is irreplaceable in electrical wiring, power distribution, busbars, transformers, and cable systems. Every data center, every grid upgrade, every transformer, and every mile of high-voltage transmission line depends on copper.
AI demand driver: Enormous. A single hyperscale data center uses 4,000–20,000 tonnes of copper. The US alone needs 50–80+ GW of new data center capacity, each GW requiring ~5,000–10,000 tonnes of copper. Grid expansion to serve these data centers doubles copper demand again. Total incremental AI-driven copper demand could reach 2–4 million tonnes/year by 2028–2030, against a global market of ~26 million tonnes.
Supply constrained: Severely. New copper mines take 10–15 years from discovery to production. Chile and Peru (40% of global supply) face water shortages, community opposition, and declining ore grades. There is no quick supply response. This is one of the tightest bottlenecks in the entire AGI supply chain.
FCX = Freeport-McMoRan (largest US copper pure-play); SCCO = Southern Copper (lowest-cost major); TECK = Teck Resources; HBM = Hudbay Minerals; TGB = Taseko Mines; CPER = United States Copper Index Fund (ETF proxy).
Aluminum is produced by refining bauxite ore into alumina, then smelting alumina via electrolysis into aluminum metal. It is the second most-used metal after steel, valued for its light weight, conductivity, corrosion resistance, and thermal properties. In AI infrastructure, aluminum is used for server chassis, heat sinks, data center structural framing, cable trays, busbars (as copper substitute), and cooling system components.
AI demand driver: Strong and growing. Aluminum is the primary copper-substitution metal — as copper prices spike, utilities and data center builders switch to aluminum for power cables and busbars where possible. Aluminum also dominates in server heatsinks, rack enclosures, and liquid cooling cold plates. Each hyperscale campus uses thousands of tonnes of aluminum.
Supply constrained: Moderately. Global smelting capacity is concentrated in China (~60%). Western smelting has been declining for years due to high electricity costs. Several US/European smelters have permanently closed. New capacity requires cheap, reliable baseload power — exactly the same power being diverted to data centers.
AA = Alcoa (integrated miner-smelter); CENX = Century Aluminum (US smelter, benefits from reshoring); ARNC = Arconic (rolled aluminum products); KALU = Kaiser Aluminum (specialty flat-rolled).
Silver miners extract silver ore, often as a byproduct of copper, lead, or zinc mining. Silver is refined into bars, coins, and industrial-grade material. Silver has the highest electrical and thermal conductivity of any metal, making it critical for electrical contacts, soldering in electronics, silver paste in photovoltaic cells, and high-reliability connections in advanced semiconductor packaging.
AI demand driver: Strong and underappreciated. Silver is consumed in solar panel manufacturing (each GW of solar uses ~20 tonnes of silver), and hyperscalers are building massive solar farms to power data centers. Silver also goes into high-reliability electrical contacts, circuit board soldering, and advanced chip packaging. Industrial silver demand has been in structural deficit for 4+ years.
Supply constrained: Yes. ~70% of silver is mined as a byproduct — you cannot increase silver supply without increasing copper/zinc/lead mining. Primary silver mines are scarce and concentrated in Mexico and Peru. The silver market has been in physical deficit since 2021.
HL = Hecla Mining (largest US silver producer); CDE = Coeur Mining; PAAS = Pan American Silver; MAG = MAG Silver; FSM = Fortuna Mining (formerly Fortuna Silver Mines).
Uranium miners extract uranium ore through conventional open-pit/underground mining or in-situ leach recovery (ISR), then process it into yellowcake (U3O8) for sale to nuclear fuel fabricators. Uranium is the fuel for nuclear fission reactors, which provide 24/7 carbon-free baseload power — exactly what AI data centers need. This sector is also covered in Module 01 (Energy & Power) but is included here as it is fundamentally a mining/extraction business.
AI demand driver: Very strong. Hyperscalers (Microsoft, Google, Amazon, Meta) are all signing nuclear PPAs and pursuing SMR deployments to guarantee 24/7 baseload power for AI training clusters. Nuclear restarts (Three Mile Island, Palisades) add incremental uranium demand. The nuclear renaissance is being driven directly by AI power needs.
Supply constrained: Extremely. Global uranium production covers only ~75% of reactor demand; the deficit has been filled by inventory drawdowns and secondary sources that are depleting. Restarting idled mines takes 2–3 years. New mine development takes 8–12 years. Kazakhstan (40% of supply) faces geopolitical risk.
CCJ = Cameco (largest Western producer); UEC = Uranium Energy Corp (US ISR producer); DNN = Denison Mines; NXE = NexGen Energy (Arrow deposit); EU = enCore Energy (US ISR).
Rare earth elements (REEs) are a group of 17 metallic elements critical for high-performance permanent magnets (neodymium, praseodymium, dysprosium), electronics, fiber optics, and advanced defense systems. Mining is relatively straightforward, but separation and processing are chemically complex and environmentally challenging. China controls ~60% of mining and ~85–90% of refining/processing.
AI demand driver: Strong. Rare earth permanent magnets are essential for the motors in data center cooling fans, precision robotics (AGI-driven automation), EV motors (autonomous vehicles), and wind turbines (data center power). Lanthanum and cerium are used in optical glass for fiber optics and camera systems. The coming wave of AGI-powered humanoid robots will be voracious consumers of rare earth magnets.
Supply constrained: Extremely, for geopolitical reasons. China has weaponized rare earth exports before (2010 Japan embargo) and imposed new export controls in 2023–2025. Building non-Chinese processing capacity takes 5–10 years. The US has exactly one operating mine (MP Materials' Mountain Pass) and minimal processing capability. This is a national security bottleneck.
MP = MP Materials (only US rare earth mine, building US magnet factory); UUUU = Energy Fuels (uranium + rare earth processing); TMRC = Texas Mineral Resources (development stage, Round Top deposit).
Steel producers transform iron ore and scrap metal into steel through blast furnace (integrated) or electric arc furnace (EAF/mini-mill) processes. Steel is the structural backbone of data centers — used for building frames, server racks, raised floors, cable trays, rebar in foundations, and transmission tower construction. EAF producers using recycled scrap are dominant in the US market.
AI demand driver: Strong. Each hyperscale data center campus uses 20,000–50,000+ tons of structural steel. With 100+ campuses under construction or planned in the US alone, that is millions of tons of incremental steel demand. Steel rebar goes into massive concrete foundations. Steel racks hold every server. Transmission towers for grid expansion are steel.
Supply constrained: Low to moderate. US EAF capacity is substantial and can ramp relatively quickly. Steel is not a bottleneck the way copper is. However, sustained demand from data center construction, combined with infrastructure spending and reshoring, could tighten the US market.
NUE = Nucor (largest US steelmaker, EAF); STLD = Steel Dynamics (EAF, highest margins); CLF = Cleveland-Cliffs (integrated, auto-focused); X = United States Steel; CMC = Commercial Metals Company (rebar, structural).
Lithium miners extract lithium from hard-rock spodumene deposits (Australia) or evaporative brine operations (South America's "lithium triangle"). Lithium is refined into battery-grade lithium carbonate or lithium hydroxide, the critical cathode material for lithium-ion batteries. Batteries are used in backup power systems (UPS), grid-scale energy storage, and EVs.
AI demand driver: Moderate to strong. Data centers require massive UPS battery banks (typically lithium-ion) for power continuity. Grid-scale battery storage (co-located with solar/wind farms powering data centers) is a growing lithium consumer. The AGI-driven autonomous vehicle wave will also drive EV battery demand. However, lithium demand from AI is secondary to EV demand.
Supply constrained: Currently oversupplied after a capacity surge in 2023–2024, but long-term structural deficit expected as EV + storage demand resumes growth. Prices crashed ~80% from 2022 peak, causing project deferrals that will tighten supply in 2027–2029.
ALB = Albemarle (largest Western lithium producer); SQM = Sociedad Quimica y Minera (Chilean brine, ADR); LAC = Lithium Americas (Thacker Pass, Nevada); ALTM = Arcadium Lithium (spun from Livent/Allkem merger, if still listed separately).
Specialty metals are a diverse group of less-common metals critical for advanced manufacturing. Cobalt is used in battery cathodes and superalloys. Nickel goes into stainless steel, battery cathodes, and high-temperature alloys. Tungsten provides hardness for cutting tools and thermal resistance. Tin is the primary soldering metal for all electronics. Each has a unique supply chain, often dominated by one or two producing countries.
AI demand driver: Moderate to strong, depending on the metal. Tin is a direct play — every circuit board in every AI server is soldered with tin. Cobalt and nickel are battery metals (UPS, grid storage). Tungsten goes into cutting tools for precision manufacturing of server components. Nickel is also critical for stainless steel in data center construction and liquid cooling systems.
Supply constrained: Varies. Cobalt is ~70% DRC-sourced (geopolitical risk). Indonesia dominates nickel (environmental/political risk). Tin supply is tight with Myanmar disruptions. Tungsten is ~80% China-controlled.
VALE = Vale SA (ADR, major nickel/cobalt producer); GLENF = Glencore OTC (major cobalt producer); NOVT = Novanta (precision metal components, indirect); MTRN = Materion (specialty alloys including beryllium-copper, tungsten, advanced materials). Note: pure-play specialty metal miners with US listings are scarce — most exposure comes through diversified miners.
Silicon metal is produced by smelting quartz (silica) with carbon in electric arc furnaces. Metallurgical-grade silicon is used in aluminum alloys and silicones. Upgraded to semiconductor-grade polysilicon (99.9999999% purity), it becomes the wafer substrate for every chip in every AI server. Solar-grade polysilicon (slightly lower purity) goes into photovoltaic cells. The purification chain from rock to wafer-grade material is energy-intensive and technically demanding.
AI demand driver: Moderate to strong. Every AI chip starts as a silicon wafer. However, the silicon wafer cost is a tiny fraction of a finished GPU's value, so silicon metal price increases have limited impact on chip economics. Solar polysilicon demand is larger and more price-sensitive, driven by solar buildouts powering data centers.
Supply constrained: Moderate. China dominates both metallurgical silicon (~70%) and polysilicon (~80%) production. Western polysilicon capacity exists (Wacker, Hemlock) but is limited. Semiconductor-grade silicon is even more concentrated. Tariff/trade tensions could tighten Western supply.
REC = REC Silicon (OTC, polysilicon producer); FSLR = First Solar (uses CdTe not silicon, but indirect competitor/benchmark); WCH = Wacker Chemie (ADR, OTC, major polysilicon). Note: Hemlock Semiconductor is private (Corning JV). US-listed pure-play silicon metal companies are very scarce. Globe Specialty Metals merged into Ferroglobe (GSM on NASDAQ).
GSM = Ferroglobe (silicon metal, silicon-based alloys, manganese alloys; one of the largest Western silicon metal producers).
Graphite is a crystalline form of carbon used as the anode material in virtually all lithium-ion batteries, in thermal management solutions (heat spreaders, thermal interface materials), in electrodes for electric arc furnace steelmaking, and as a moderator in nuclear reactors. Natural graphite is mined; synthetic graphite is manufactured from petroleum coke at high temperatures. Both forms are critical for battery and thermal management applications.
AI demand driver: Moderate. Graphite is the anode in every lithium-ion battery (UPS systems, grid storage). Graphite-based thermal interface materials are used in GPU heat dissipation. Synthetic graphite electrodes are consumed in EAF steelmaking (data center structural steel). Nuclear graphite is used in some advanced reactor designs (AGI power). Demand is real but spread across multiple end-uses.
Supply constrained: High. China controls ~65% of natural graphite mining and ~95%+ of graphite processing/anode production. China imposed graphite export controls in late 2023. Non-Chinese supply is extremely limited. Battery anode-grade graphite processing outside China barely exists at scale.
SGL = SGL Carbon (ADR/OTC, graphite specialty products); GRA = GrafTech International (synthetic graphite electrodes); GRPH = Graphite Bio / Novalis LifeSciences (pivoted away from graphite — limited relevance). Note: Pure-play natural graphite miners with US listings are extremely scarce. Nouveau Monde Graphite (NMG) is the most relevant development-stage play.
NMG = Nouveau Monde Graphite (Quebec-based, NYSE-listed, developing integrated mine-to-anode-material operation).
Mining equipment manufacturers design, build, and service the heavy machinery used in mining operations: haul trucks, excavators, drills, loaders, crushers, grinding mills, conveyors, and autonomous haulage systems. They also provide aftermarket parts and service, which generates recurring revenue. These companies are the "picks and shovels" of the mining industry — they benefit from any increase in mining activity regardless of which specific commodity is being mined.
AI demand driver: Strong. Every copper mine expansion, every new uranium operation, every lithium project needs heavy equipment. The AGI-driven commodity supercycle will drive a multi-year equipment replacement and expansion cycle. Additionally, mining companies are adopting AI/autonomous systems (autonomous haul trucks, AI-driven geological modeling, predictive maintenance), which drives technology upgrade cycles from the equipment OEMs.
Supply constrained: Low. Equipment manufacturers can scale production over 1–2 year cycles. The bottleneck is in the mines, not the equipment. However, order backlogs are extending and pricing power is increasing.
CAT = Caterpillar (dominant mining equipment OEM, autonomous haulage leader); DE = Deere & Company (expanding into mining); TEX = Terex (aerial platforms, cranes, materials processing); ASTE = Astec Industries (aggregate/mining processing equipment); MTW = Manitowoc (cranes for mine/infrastructure construction).
Diversified miners operate mines across multiple commodities (copper, iron ore, aluminum, nickel, cobalt, lithium, uranium), providing portfolio exposure to the entire metals complex. Commodity traders combine physical trading (buying, storing, transporting, blending, and selling physical commodities) with mining/production assets. This integrated model captures margins across the value chain and benefits from volatility and supply dislocations.
AI demand driver: Strong. Diversified miners and traders benefit from the entire AGI commodity supercycle: copper, aluminum, uranium, nickel, cobalt, lithium, rare earths. They do not need any single commodity to outperform; they win when the whole complex moves up. Commodity traders additionally benefit from the volatility and physical dislocations that supply crunches create.
Supply constrained: Varies by commodity within their portfolios, but the diversification means at least some of their portfolio is always in a tight market.
BHP = BHP Group (ADR, copper/iron ore/nickel/potash); RIO = Rio Tinto (ADR, copper/aluminum/iron ore/lithium); VALE = Vale SA (ADR, iron ore/nickel/copper/cobalt); TECK = Teck Resources (copper/zinc/steelmaking coal); GLNCY = Glencore (OTC, mining + trading, copper/cobalt/zinc/nickel).
Mineral processors and refiners take raw mined ore or concentrate and transform it into usable metals and materials through smelting, refining, hydrometallurgy, or chemical processing. This is the critical middle step between a hole in the ground and a usable industrial material. The processing stage is often the tightest bottleneck — building a new copper smelter or rare earth separation plant takes years and billions of dollars.
AI demand driver: Strong, as a derived demand from the entire mining supply chain. Every tonne of copper concentrate, every kilogram of rare earth ore, every pound of uranium yellowcake needs to be processed before it enters the supply chain. Processing capacity is the binding constraint for several critical materials.
Supply constrained: High for several materials. China dominates refining for rare earths (~85%), graphite anodes (~95%), cobalt (~70%), and lithium (~65%). Western refining capacity is scarce and takes 3–7 years to build. IRA/CHIPS Act subsidies are attempting to stimulate Western processing capacity.
MTRN = Materion (beryllium, advanced alloys, precision processing); UUUU = Energy Fuels (uranium processing + rare earth separation); MP = MP Materials (rare earth processing + magnet production); MEOH = Methanex (methanol, used in mining chemical processing). Note: Some tickers repeat from earlier sectors because vertically integrated companies span mining and processing.
Gold miners extract gold from open-pit and underground deposits, process ore through cyanide leaching or gravity separation, and sell refined gold bars/dore to bullion banks and central banks. Gold is primarily a monetary/store-of-value asset (~90% of demand is investment + jewelry). Industrial use is limited to electronics (bonding wire, contacts) and some medical applications.
AI demand driver: Weak. Gold's industrial use in electronics (bonding wire in chip packaging, connector plating) is real but tiny relative to total gold demand. Gold is not a structural input to AI infrastructure the way copper, aluminum, or silver are. The AGI thesis does not create meaningful incremental gold demand.
Supply constrained: Moderately. Gold mine supply is relatively stable at ~3,600 tonnes/year. New projects are expensive and take years. But the investment demand side is driven by macro/monetary factors (central bank buying, inflation hedging), not industrial consumption.
NEM = Newmont (largest gold miner); GOLD = Barrick Gold; AEM = Agnico Eagle Mines; KGC = Kinross Gold; GFI = Gold Fields (ADR). Included for completeness. These are fine companies but they are not AGI plays.
Industrial minerals companies quarry aggregates (crushed stone, gravel, sand), produce cement, and manufacture concrete. These are the fundamental construction materials for any large building project, including data centers. Every data center campus requires massive concrete foundations, reinforced slabs, and aggregate base courses. Cement and aggregates are heavy, low-value-to-weight commodities that are consumed locally.
AI demand driver: Moderate but generic. Data center construction consumes concrete and aggregates, but so does every other construction project. AI does not create a differentiated demand signal for these materials the way it does for copper or uranium. These companies benefit from the general construction boom but have no AI-specific pricing power.
Supply constrained: Low. Aggregate and cement capacity in the US is substantial. Permitting new quarries can be slow, but existing capacity is generally sufficient. Pricing is local/regional and driven by overall construction activity.
VMC = Vulcan Materials (largest US aggregates producer); MLM = Martin Marietta Materials; EXP = Eagle Materials (cement, wallboard).
PGM miners extract platinum, palladium, and rhodium, primarily from deposits in South Africa and Russia. These metals are used mainly in automotive catalytic converters (~60–70% of demand), jewelry, and some industrial/chemical applications. Platinum is also used in hydrogen fuel cell catalysts. PGM mining is concentrated, with Anglo American Platinum, Impala Platinum, and Sibanye-Stillwater dominating global supply.
AI demand driver: Weak to moderate. Platinum has a potential role in hydrogen fuel cells (some data center backup power concepts) and in certain chemical process catalysts. Palladium has minor uses in electronics (multilayer ceramic capacitors). But the dominant demand driver is automotive catalytic converters, which face secular decline from EV adoption. AGI does not create meaningful incremental PGM demand.
Supply constrained: High (South Africa labor/power issues, Russia sanctions), but declining demand offsets supply tightness. The structural outlook is challenged by EV transition reducing catalytic converter demand.
SBSW = Sibanye-Stillwater (only US PGM mine operator, ADR); IMPUY = Impala Platinum (OTC ADR); ANGPY = Anglo American Platinum (OTC ADR).
The following tickers represent the complete US-listed mining, metals & raw materials universe relevant to the AGI build-out. Some tickers appear in multiple sectors (e.g., VALE in both Specialty Metals and Diversified Miners; MP in both Rare Earths and Mineral Processing). Deduplicated unique ticker count is shown below.
Unique tickers: 65 | Mentioned across 16 sectors (some tickers span multiple sectors)