Critical Minerals: A Practical Reference

Lithium reaches a converter in one of three forms, and a project developer picks between them before anything else is decided. Hard rock means spodumene, a pegmatite mineral that is crushed, floated, and sold as a concentrate quoted at 6% Li2O; the mine can be built in two or three years and it produces from day one. Brine means pumping lithium-bearing salt water into evaporation ponds in the Atacama or the Argentine puna and waiting 18 to 24 months while sun and wind concentrate it from a few hundred parts per million up to several percent. Clay and direct lithium extraction are the frontier: sedimentary clays such as Thacker Pass need an acid leach, and DLE uses a sorbent or membrane to strip lithium from brine in hours instead of a year and a half. All three end up at the same place, a lithium chemical plant that makes carbonate or hydroxide, which is a separate business covered in the next entry. The important thing about this stage is that it is the least concentrated part of the lithium chain.

Lithium is not geologically scarce and the resource base is spread across four continents, so this is one of the few battery inputs where a Western buyer has real choices. Hard rock is fast to build, responds to price within a year, and is dominated by Australia, which is about as friendly a jurisdiction as mining gets. The failure mode is the cost curve, not the geology: hard-rock mines have high operating costs because you are hauling and floating rock to recover about 1% lithium, so when carbonate fell to $8,000–10,000/t through 2024 and 2025 the marginal Australian and African operations stopped covering cash costs and several shut or curtailed. Brine has the opposite failure mode. Cash costs are low, but ponds take 18 to 24 months to fill, so a brine project ramps years later than its schedule says and cannot be turned up when prices spike. Clay and DLE have not yet failed at scale; they have mostly not been proven at scale either.

If you are securing feedstock and you need volume inside three years, buy spodumene. Australian hard rock is the only part of the lithium chain that can be contracted quickly, in a jurisdiction most Western buyers accept, and it is the swing supply that clears the market. If you are building an asset and you care about surviving the trough rather than catching the peak, buy into brine, because a Salar de Atacama or Hombre Muerto operation stays cash-positive at prices that shut hard rock. Take clay or DLE only if you can tolerate technical risk, and diligence the process on recovery from real field brine rather than a laboratory column. And be honest about what securing raw material buys you: it does not get you a battery-grade chemical, because the conversion step is a different plant, a different geography, and the tighter constraint.

Spodumene concentrate is sold and priced at 6% Li2O · Australia has run roughly a third to 40% of world mine supply in recent years, Chile about a fifth · brine evaporation takes 18–24 months from pond to product · lithium carbonate ran about $6,000/t in early 2021, roughly $80,000/t in late 2022, $8,000–10,000/t through 2024 and 2025, and about $26,000/t in Q1 2026 · spodumene concentrate was assessed above $2,000/t CIF China in late January 2026 against roughly $800/t through 2025 · a 75 kWh pack carries on the order of 50 kg of lithium carbonate equivalent.

There isn't a sharp one here, and that is the point of this entry. Australia has run roughly a third to 40% of world mine supply in recent years, Chile about a fifth, with China and Argentina behind them, so the largest single country sits well under half. The tightest single asset is Greenbushes in Western Australia, the highest-grade hard-rock lithium mine in production, but even a full outage there would move price rather than stop the industry, and Australian, Chilean, Argentine, Brazilian, and Zimbabwean supply are substitutable at the concentrate level. Chinese ownership stakes run through much of this supply, including lepidolite mines at home and spodumene offtakes in Africa, but ownership is not the same as location. If you want the real lithium chokepoint, it is one step downstream in conversion, which is roughly 60–72% Chinese. Western mine supply is not the missing piece; a new hard-rock mine takes two to four years to build once permitted, and the projects exist.

Greenbushes, Western Australia (Talison, owned by Albemarle and the Tianqi-IGO joint venture — the highest-grade hard-rock lithium mine in production); Pilbara Minerals' Pilgangoora; SQM and Albemarle on the Salar de Atacama in Chile; Rio Tinto's Rincon and the former Arcadium operations in Argentina; CATL's Jianxiawo lepidolite mine in Yichun, whose suspension in 2025 was one of the triggers for the price rebound; Lithium Americas' Thacker Pass clay project in Nevada; ExxonMobil and Standard Lithium pursuing DLE on Arkansas Smackover brine.

Spodumene is priced off the lithium chemical price with a lag, usually a formula referencing the prior or following month's carbonate assessment, which means the miner and the converter split a margin that swings by an order of magnitude. Carbonate went from about $6,000/t in early 2021 to roughly $80,000/t in late 2022, then down to $8,000–10,000/t across 2024 and 2025, then back to about $26,000/t in Q1 2026, and spodumene followed it from roughly $800/t to above $2,000/t CIF China in late January 2026. Brine sits at the bottom of the cost curve, integrated Australian hard rock in the middle, and Chinese lepidolite and marginal African spodumene at the top, which is why the top of the curve turns off and on with every cycle and sets the price. Capital intensity is moderate by mining standards and lead times are short compared with copper or nickel, so supply overshoots: the 2022 price spike financed a wave of projects that all arrived together in 2024 and crushed the price. If you are underwriting a lithium mine, model the trough rather than the spot price, and check where the asset sits on the cost curve, because that is the only thing that decides whether it survives the next overshoot.

16 minThe True Cost of Lithium Mining | True Cost | Business InsiderBusiness Insider · 1m+ views

10 minLithium Brine Extraction: The Race to Power Tomorrow's WorldThe Deep Dive · 10k+ views

14 minRevolutionising lithium production. From 18 months to 6 HOURS!!Just Have a Think · 50k+ views

Lithium, Mineral Commodity Summaries 2026 (USGS) · Lithium Production in North America: A Review (Argonne National Laboratory)

A cell maker does not buy spodumene or brine. It buys lithium carbonate or lithium hydroxide monohydrate at battery grade, and the plant that makes those is a chemical works, not a mine. From hard rock the route is a decrepitation roast at around 1,100 °C to flip spodumene from its alpha to its beta crystal form, then an acid roast, a leach, several purification stages to strip calcium, magnesium, iron, and sodium, and finally crystallization. From brine the concentrated liquor goes more or less straight to carbonate, and hydroxide is made by reacting carbonate with lime or by an electrochemical route. Which chemical you want depends on the cathode: carbonate feeds LFP and mid-nickel NMC, while high-nickel cathodes need hydroxide because it lets the cathode be calcined at a lower temperature, and nickel-rich layered oxides degrade if you push the sintering temperature up to suit carbonate. The specification is the product. Battery grade means 99.5% and up with sodium and calcium held to tens of parts per million and ferromagnetic particles counted in parts per billion, because a magnetic particle in a cathode becomes an internal short in a cell.

Conversion is the highest-leverage step in the lithium chain: it turns a cheap bulk concentrate into a specified chemical, and the qualification a converter holds with a cathode maker is worth more than the plant. That qualification also cuts the other way, because a new plant has to run at spec for months and then survive a customer audit before it sells a tonne, so nameplate capacity and saleable capacity are different numbers for years. The failure mode is not the chemistry, it is the ramp. Impurity control at the tail of the distribution is a plant-specific art, and the Western record is bad: Albemarle put more than $4 billion into Kemerton in Western Australia and idled it in February 2026, and the Tianqi-IGO Kwinana hydroxide plant was still running near a third of nameplate three years in before its second train was suspended and the investment fully impaired. Chinese converters, running dozens of plants with the same flowsheet and the same reagent supply next door, hit spec and hold it.

If you are a cell or cathode maker sourcing lithium, treat carbonate and hydroxide as separate supply problems rather than one lithium problem, because they have different producers, different prices, and different qualification files. Buy carbonate if your cathode is LFP or mid-nickel NMC, which is most of the market by volume now; buy hydroxide if you are running nickel-rich NMC or NCA, where the lower calcination temperature is not optional. If you are trying to reduce Chinese exposure, this is the step to attack, not the mine, but budget five years and expect to pay a premium, because the plants that have tried it took years longer and cost far more than planned. If you are underwriting a Western conversion project, the questions that matter are who has qualified the product, how many months of on-spec production exist, and what the plant's cost per tonne looks like against a Chinese converter buying the same concentrate.

Battery grade is 99.5% and up, with ferromagnetic particles specified in parts per billion · China holds roughly 60–72% of world lithium chemical conversion capacity · Chinese output in 2025 was on the order of 780,000 t of carbonate and 360,000 t of hydroxide · carbonate feeds LFP and mid-nickel cathodes, hydroxide the high-nickel ones · Kemerton absorbed more than $4B before Albemarle idled it in February 2026 · Tianqi-IGO's Kwinana train 1 ran near 35% of nameplate after three years, and train 2 was suspended · carbonate has traded between roughly $6,000/t and $80,000/t over the past five years.

Conversion is where lithium is actually concentrated. China holds roughly 60–72% of world lithium chemical capacity depending on whose count you use, and made something like 780,000 t of carbonate and 360,000 t of hydroxide in 2025. That is why Australia mines a third or more of the world's lithium and refines a small fraction of it: the rock ships to Chinese converters, comes back as a chemical, and the value added in between stays there. A disruption here is not subtle, because a cathode plant cannot substitute technical-grade material and cannot requalify a new source in less than six to twelve months. The Western alternative is a conversion plant, nominally a two-to-four-year build, and the record says longer. Kemerton took more than $4 billion and was idled in February 2026; the Tianqi-IGO Kwinana plant was fully impaired after years below nameplate. The counterexample is Covalent's Kwinana refinery, built for less than its original estimate and producing hydroxide from July 2025, which shows the problem is executable rather than impossible.

Ganfeng Lithium and Tianqi Lithium (the two largest converters, both Chinese, both with plants outside China as well); Yahua and Shengxin, which supply hydroxide to Korean and Japanese cathode makers; Albemarle's Kemerton in Western Australia, idled in February 2026; the Tianqi-IGO Kwinana hydroxide plant, whose second train was suspended; Covalent (Wesfarmers and SQM) at Kwinana, first hydroxide in July 2025; SQM's Salar de Carmen carbonate plants in Chile; Tesla's in-house lithium refinery at Robstown, Texas.

The converter's margin is the spread between the chemical price and the concentrate price, and because spodumene contracts index to a lagged carbonate assessment, that spread inverts in a falling market and blows out in a rising one. Chinese converters win on operating cost for unglamorous reasons: sulfuric acid, soda ash, and lime are cheap and local, waste gypsum and sodium sulfate have somewhere to go, power is cheap, and the plants sit next to their customers. A Western plant pays more for every one of those inputs and often has to pay to dispose of residues that a Chinese plant sells. That is why the honest Western business case is not lower cost, it is a customer paying a premium for a non-Chinese qualified source, usually under a government-backed offtake or a floor price. If you are looking at a conversion company, the value sits in qualification and utilization rather than in the resource, and a plant running at 35% of nameplate has no business at any lithium price.

8 minBig Tesla LITHIUM REFINERY Update | Spodumene + Salt = MAGIC!Cleanerwatt · 10k+ views

5 minMetso Lithium Hydroxide ProcessMetso · 1k+ views

18 minFrom Mining to Manufacturing: The Lithium Extraction Process and Refining Technology!Lord Gizmo · 50k+ views

Lithium, 2022 Minerals Yearbook (USGS) · Lithium factsheet (SCRREEN2)

Almost nobody mines cobalt on purpose. It comes out as a byproduct, mostly of copper in the Central African Copperbelt and secondarily of nickel in laterite and sulfide operations, which means the amount produced is set by copper and nickel economics rather than by cobalt demand. In the Congo the ore is leached, the copper is taken out by solvent extraction and electrowinning, and the cobalt is precipitated as a hydroxide intermediate running 20–40% cobalt. That intermediate is shipped, overwhelmingly to China, and refined into cobalt sulfate for cathode precursors or into metal for everything else. The battery market takes the majority of it, through NMC and NCA cathodes and through the lithium cobalt oxide still used in phones and laptops, but the non-battery uses are the ones with no way out: cobalt is the binder that holds tungsten carbide cutting tools together, and it is a load-bearing element in the superalloys that turbine blades and vanes are cast from.

Cobalt does things in a cathode that nothing else does as cheaply (it stabilizes the layered structure, which is why high-nickel chemistries still carry a little of it), and in superalloys and hard metals it is close to irreplaceable. Its weakness is that supply is a policy variable and demand is being designed out at the same time. Byproduct status means a cobalt price spike does not open a cobalt mine; it opens a copper mine, eventually. And the failure mode buyers actually hit is reputational and legal rather than physical. A meaningful share of DRC output passes through artisanal workings that mix into the same trading channels as industrial material, and estimates of that share range from a couple of percent to 20% or more depending on the cobalt price and on who is counting, so a buyer with a due-diligence obligation is auditing a supply chain whose upstream is genuinely hard to trace.

If you are choosing a cathode and cobalt exposure is what worries you, the answer is LFP, which contains none at all, and it is the right default for standard-range vehicles and for stationary storage. If you need the energy density, go high-nickel and accept a small cobalt content rather than trying to eliminate it, because the cell-level cost of removing the last few percent is worse than the supply risk. If you are buying cobalt for superalloys, hard metals, or magnets, stop looking for a substitute and start looking at inventory: those applications cannot switch, the volumes are small enough that a year of cover is affordable, and qualification of an alternative in an aerospace part takes longer than most price cycles last. Wherever you buy, budget for third-party audit and traceability from mine to refinery, because that is now a condition of selling into Europe and to most Western OEMs.

The DRC supplies roughly 70–76% of world mine supply · China refines 75–80% of it · cobalt is recovered as a byproduct of copper and nickel, not mined for itself · the DRC banned exports in February 2025 and replaced the ban with a quota of 96,600 t for 2026, 87,000 t of it allocated pro rata · cobalt hydroxide went from about $5.60/lb in February 2025 to roughly $26/lb in April 2026, near $57,000/t · NMC 111 is one-third cobalt in the cathode metal ratio and NMC 811 is one-tenth, while LFP is zero.

Cobalt is the textbook two-country chain, and both links are tight. The DRC supplies roughly 70–76% of mine output, much of it through Chinese-owned operations such as CMOC's Tenke Fungurume and Kisanfu, and China refines 75–80% of world supply into sulfate and metal regardless of where the ore came from. The live constraint in 2026 is policy rather than geology. Kinshasa banned cobalt exports in February 2025 to defend the price, then replaced the ban with a quota system: 96,600 t for 2026, of which 87,000 t is allocated to producers pro rata and 9,600 t is held back at the regulator's discretion. It worked. Cobalt hydroxide went from about $5.60/lb just before the ban to roughly $26/lb by April 2026, and actual shipments have been running well below the allocated cap on logistics alone. The Western alternatives are real but small: Glencore's Murrin Murrin and Nikkelverk, Sumitomo and Umicore refining outside China, Jervois's Idaho project, and cobalt recovered from Indonesian MHP. None of them changes the DRC's share of the mine, and a greenfield cobalt-primary mine is not something the market will finance.

CMOC's Tenke Fungurume and Kisanfu in the DRC (now the largest cobalt producer in the world); Glencore's Katanga and Mutanda operations; Eurasian Resources Group's Metalkol RTR, which reprocesses tailings; Huayou Cobalt and CNGR as the dominant Chinese refiners and precursor makers; Umicore in Belgium and Sumitomo Metal Mining in Japan as the main non-Chinese refiners; Haynes, Carpenter, and PCC using cobalt in superalloys; Sandvik and Kennametal in tungsten carbide tooling.

Cobalt prices are set by an assessed hydroxide payable rather than an exchange contract for most of the volume: the intermediate trades at a percentage of the standard-grade metal price, and that payable moves as much as the metal price does. Because it is a byproduct, the cost curve is not a cobalt cost curve at all — a copper mine in the Congo will keep producing cobalt at any price the market offers, because the cobalt is a credit against copper costs. That asymmetry is why the price collapsed to around $5.60/lb by early 2025 and why only a state intervention moved it. Who makes money: the DRC government, through royalties and now through the quota mechanism itself, and the Chinese refiners and precursor makers who sit between the intermediate and the cathode. A Western cobalt project needs either a floor price or a customer paying for provenance, because on cost alone it competes with material that is nearly free to produce.

9 minCobalt - Periodic Table of VideosPeriodic Videos · 500k+ views

13 minHow Child-Mined Cobalt From The Congo Powers Our PhonesPopular Mechanics · 10k+ views

Cobalt, Mineral Commodity Summaries 2026 (USGS) · Cobalt in the Democratic Republic of Congo: Market Analysis (World Bank)

Nickel splits into two products that are not interchangeable. Class 1 is nickel at 99.8% and up (briquettes, cathode, powder, and the sulfate crystals a cathode precursor plant dissolves), and it historically came from sulfide ores in Canada, Russia, and Australia. Class 2 is ferronickel and nickel pig iron, 4–15% nickel in an iron matrix, which is all a stainless steel mill needs and which comes from laterite ore smelted in a rotary kiln electric furnace. The change that reorganized this market was making laterite reach the class-1 market. Two routes do it: high-pressure acid leaching of limonite ore to a mixed hydroxide precipitate at 35–40% nickel, which is then refined to sulfate, and converting nickel pig iron to a high-grade matte that can be leached the same way. Both were built in Indonesia, mostly with Chinese capital and Chinese engineering, and both are cheaper than mining sulfide. Stainless steel still takes roughly two thirds of all nickel demand and sets the floor under the price; batteries are the growth market but not yet the big one.

Nickel is the element that buys energy density in a layered-oxide cathode, and there is a lot of it — this is an abundance problem, not a scarcity problem. That is exactly the weakness. Indonesia built so much capacity so fast that the price fell below the cash cost of most of the rest of the world, and the failure mode showed up as mine closures in friendly jurisdictions: BHP suspended Nickel West in Western Australia in 2024, Glencore idled Koniambo in New Caledonia, and several smaller Australian operations went with them. Once those assets are on care and maintenance they are slow and expensive to restart, so the diversification a Western buyer thought it had is gone and will not come back on a price signal alone. The second weakness is that the HPAL and RKEF capacity that replaced them is coal-fired and generates large volumes of tailings, so the cheapest nickel is also the most carbon-intensive, which matters to any buyer selling into Europe.

If your cathode is high-nickel NMC or NCA, you need class-1 units and there is no way around Indonesian intermediate in the mix; contract on MHP or matte rather than on refined metal, since that is where the volume actually is. If your problem is exposure rather than price, LFP removes nickel from the cathode entirely, which is the same move that removes cobalt, and it is the reason nickel demand forecasts have been cut repeatedly. If you are buying for stainless or plating, buy class 2 and ignore the battery-grade conversation, because those are separate markets at separate prices. And if you are underwriting a non-Indonesian nickel project, do not model the current price — model Indonesia's marginal cost, because a producer with 60% of mine supply and an annual quota lever decides what the price is.

Class 1 is 99.8% nickel and up; class 2 is 4–15% nickel in iron · Indonesia went from about 32% of world mine supply in 2020 to roughly 60% by 2024–25 · MHP from high-pressure acid leaching runs 35–40% nickel · stainless steel takes roughly two thirds of nickel demand · LME nickel ran about $14,000–16,000/t through much of 2025, reached roughly $20,000/t on 6 May 2026, and fell back near $16,400/t in June 2026 · Indonesia's 2026 ore quota was approved at 260–270 Mt against 375–379 Mt for 2025, a cut of about 30%.

The chokepoint is a producer with pricing power rather than a shortage. Indonesia went from about 32% of world mine supply in 2020 to roughly 60% by 2024–25, built on Chinese-financed RKEF and HPAL lines in the Morowali and Weda Bay industrial parks, and China refines somewhere around 60% of the world's nickel and most of the battery-grade sulfate. What makes this different from cobalt is that the lever is annual and explicit: Indonesia's RKAB system sets how much ore each mine may extract, and the 2026 approval came in at 260–270 million wet tonnes against 375–379 million for 2025. LME nickel responded by running from roughly $14,000/t in December 2025 to about $20,000/t by early May 2026, then correcting to near $16,400/t in June once the market decided the cut would not bind. A Western alternative exists on paper (restart Nickel West, build a sulfate plant, qualify Canadian or Brazilian material), but restarts take two to four years and none of it works at a price Indonesia can set. This is the entry where the honest answer is that diversification costs money rather than time.

PT Vale Indonesia, Tsingshan's Morowali park, and the Weda Bay complex (the RKEF and HPAL capacity that changed the market); Nickel Industries and Lygend on the HPAL side; Norilsk Nickel, historically the largest class-1 producer outside Indonesia; Glencore's Nikkelverk in Norway and Sumitomo's Niihama refinery, the main non-Chinese sulfate sources; BHP's Nickel West, suspended in 2024; Glencore's Koniambo in New Caledonia, idled the same year; Vale's Long Harbour and Sudbury operations in Canada.

Nickel is one of the few materials on this sheet with a real exchange contract, so LME price is a price someone paid and you can hedge it, but the battery-grade product trades at a premium or discount to that benchmark depending on the sulfate payable, which moves independently. The cost curve is what matters. Indonesian RKEF and HPAL sit at the bottom on cheap coal power, captive ore, and industrial parks that share infrastructure; Western sulfide mines sit near the top and closed when the price went through them. Capital intensity for HPAL is high and the projects have a long history of overrunning, but Chinese contractors have brought build times and costs down enough that the route is now the default. Who makes money: Indonesian and Chinese integrated operators who capture the ore-to-intermediate margin, plus the Indonesian state through royalties and the quota. If you are building a business on nickel outside that system, the model needs either a green premium a customer will actually pay or a government floor, and probably both.

18 minHow China Took Over Indonesia’s Nickel Industry To Fuel Its EVsCNBC · 100k+ views

7 minThe Journey of NickelGlencore · 50k+ views

7 minThe hydrometallurgical processAmbatovy · 10k+ views

Nickel, Mineral Commodity Summaries 2026 (USGS) · Nickel for the Energy Transition: A Developmental Perspective (GIZ)

Natural flake graphite is mined and floated into a concentrate at 94–97% carbon, which is a cheap bulk commodity sold into refractories, foundries, and lubricants. Turning it into a battery anode is a different industry. The flake has to be spheronized (rounded in a high-speed mill to roughly 15–20 µm potato-shaped particles so it packs densely and lithium can intercalate from all sides), then purified to 99.95% and above, usually with hydrofluoric acid or by thermal treatment, and finally coated with a thin layer of amorphous carbon that stops the electrolyte from decomposing on the graphite surface. The product is coated spherical purified graphite, and it is what a cell maker actually buys. Spheronizing yield is the number that decides the economics: only about 30–50% of the flake ends up as spherical product, and the rest comes off as fines that have to be sold into something else. This is the sheet's cleanest illustration of the general pattern, because the mining is spread across five countries and the step after it is not.

Graphite is the incumbent anode for good reasons: it is cheap, it has been in production for thirty years, its voltage curve is flat and well understood, and natural graphite specifically has a lower embodied energy than the synthetic alternative because you skip a 3,000 °C furnace. The weakness is that natural flake is inconsistent. Flake size, crystallinity, and impurity profile vary between deposits and within a deposit, so a cell maker qualifying a new mine is qualifying a new material, and that takes 12–24 months of testing. Cycle life and fast-charge behavior are usually a little worse than synthetic, which is why the premium end of the market moved away from it. The failure mode for a Western project is neither the mine nor the chemistry. It is that the 50–70% of the flake that does not become spherical product has no home outside a market with a large refractory and recarburizer industry to absorb it, and that market is in China.

Use natural graphite when cost per kWh matters more than fast charge and cycle life, which in practice means LFP cells for standard-range vehicles and for stationary storage. Blend it with synthetic when you want part of the cost advantage without giving up all of the cycle life; blends are common and are the usual answer. Go fully synthetic when the cell has to take repeated fast charging or has a long warranted life, and accept the higher price. What you should not do is treat the natural-to-synthetic switch as supply-chain diversification, because both anode materials are made in the same country by many of the same companies. Changing the material does not change the geography, and if that is the problem you are solving, the thing to buy is non-Chinese spheronizing and purification capacity, not a different carbon.

Flake concentrate is 94–97% carbon; anode material is purified to 99.95% and above · spheronized to roughly 15–20 µm · spheroidizing yield is only about 30–50% of the flake · China mines roughly 78–80% of world flake graphite · under 1% of uncoated spherical graphite is produced outside China · a 75 kWh pack carries roughly 75–90 kg of graphite · China put graphite under export licensing from late 2023 and extended it to artificial graphite anode material and equipment from 8 November 2025.

Mining is not the constraint, though China's share of it is high at roughly 78–80%, with Mozambique, Madagascar, Brazil, and Tanzania supplying most of the rest. The constraint is spheronizing, purification, and coating, which is essentially all Chinese: under 1% of uncoated spherical graphite is made anywhere else. So a Mozambican or Tanzanian mine ships flake to China, and China ships back anode material. China put graphite under export licensing from late 2023, tightened it for US-bound shipments, and in October 2025 extended controls to artificial graphite anode material along with graphitization furnaces, coating equipment, and the associated technology, effective 8 November 2025. A disruption does not stop cells immediately, because buyers hold inventory and licenses have generally been granted, but it prices and delays every shipment and tells Beijing who is buying. The Western build is a two-to-four-year plant plus 12–24 months of cell-maker qualification, and it is under way at Syrah's Vidalia plant in Louisiana, Novonix, Posco, and Renascor's Siviour project. The volumes involved are still a small fraction of demand.

Syrah Resources' Balama mine in Mozambique feeding its Vidalia, Louisiana anode plant (the main integrated non-Chinese chain); Northern Graphite in Canada; Renascor's Siviour project in South Australia, which is built around on-site purified spherical graphite; Talga in Sweden; BTR New Material Group and Shanshan, the two largest Chinese anode makers; Posco Future M in Korea, one of the few non-Chinese producers at scale.

Flake concentrate is priced by flake size and carbon content in bilateral contracts, not on an exchange, and it is cheap — nearly all the value in the chain is added after the mine, with coated spherical graphite selling for several times the concentrate price. That is why a mine without downstream processing is a low-margin business selling into its own competitor. The cost curve outside China is set by three things: the price you can get for the 50–70% of flake that does not spheronize, the cost of purification (HF is cheap and heavily regulated, thermal purification is the reverse), and power. Chinese producers have all three advantages plus twenty years of process tuning, and they have shown they will run at low margin through a downturn to hold share. If you are underwriting a Western anode project, the questions are whether it has a qualified cell-maker offtake, what it does with its fines, and whether the price in the model is the Chinese price or a policy-supported one.

10 minWhy Graphite Is The New Gold As Tech Wars Ramp Up | Risky BusinessBusiness Insider · 1m+ views

10 minHow graphite is processedNorthern Graphite · 10k+ views

Graphite (Natural), Mineral Commodity Summaries 2026 (USGS) · From minerals to materials, Supplementary report: Graphite (CSIRO)

Synthetic graphite starts as needle coke, a highly aligned carbon made from petroleum residue or coal-tar pitch, which is milled, shaped, and then graphitized by holding it near 3,000 °C until the disordered carbon rearranges into stacked graphene layers. The traditional route is the Acheson furnace, a resistance-heated bed that takes weeks per cycle; lengthwise graphitization furnaces are the modern alternative and run faster with better uniformity. After graphitization the anode material is milled to size, classified, and carbon-coated in the same way natural graphite is. The defining feature of the process is electricity. Graphitization consumes thousands of kilowatt-hours per tonne and is the single largest line in the anode cost, which is why the industry concentrated where power is cheap rather than where the coke is. The same needle coke also goes into the graphite electrodes that electric arc furnace steel mills consume, so an anode buyer is bidding against steelmakers for the same feedstock, and premium needle coke supply is itself concentrated in a handful of producers.

Synthetic graphite is more consistent than natural flake because you make the crystal rather than find it, and consistency is what a cell maker is buying: tighter particle size distribution, controlled surface area, better first-cycle efficiency, longer cycle life, and better fast-charge tolerance. That is why it has taken the majority of anode volume even though it costs more. The weaknesses are energy and carbon. Thousands of kilowatt-hours per tonne on a Chinese grid gives the material an embodied-emissions figure several times that of natural graphite, which is a problem for anyone selling into the EU battery regulation. It is also the more expensive product, typically by a wide margin, so LFP cells built to a cost target lean toward natural or blended anodes. And the failure mode for a new entrant is subtle: hitting the target crystallinity and surface chemistry reproducibly at 3,000 °C is a furnace-engineering problem that takes years to get right, which is why "we will site it where power is cheap" has proved easier to say than to build.

Choose synthetic when the cell has to fast charge, when the warranty runs long, or when the cathode is high-nickel and you need every point of first-cycle efficiency — performance cells and premium EVs are its market. Choose natural or a blend when the target is $/kWh and the duty cycle is gentle. If you are trying to cut Chinese exposure, synthetic is the more attackable of the two anode routes, because graphitization is fundamentally a power problem and the West has cheap firm power in places China does not: Norwegian hydro, Quebec hydro, US Gulf Coast gas. But do not expect the switch to be free. Plan on 12–24 months of cell-maker qualification and a price above the Chinese landed cost, and check whether the project has secured needle coke, because the coke market is tight when EAF steel is running hard.

Graphitized at roughly 3,000 °C in Acheson or lengthwise furnaces · graphitization consumes thousands of kWh per tonne and is the largest single cost in the anode · China holds roughly 98% of world graphitization capacity · synthetic now takes the majority of anode volume · the same needle coke feeds EAF graphite electrodes, so anode buyers compete with steelmakers · China's export controls extended to artificial graphite anode material, graphitization furnaces, and coating equipment from 8 November 2025.

Graphitization, at roughly 98% Chinese capacity, which is tighter than any step in the natural-graphite chain except spheronizing. The reason is not technology, it is electricity: Chinese producers clustered in Inner Mongolia and Sichuan where industrial power is cheap, and they built enough capacity to serve the whole world with margin to spare. China's October 2025 rules put artificial graphite anode material under export licensing along with the graphitization furnaces and coating equipment used to make it, effective 8 November 2025, which is the part that matters for anyone planning to build outside China — you now need a license to buy the machines as well as the material. The Western answer is credible and slow. Vianode in Norway runs on hydro, Novonix is building in Chattanooga and Australia, Anovion has a Georgia site, and Posco Future M produces in Korea. A greenfield graphitization plant is a three-to-four-year build plus qualification, and total announced non-Chinese capacity is still a small fraction of demand.

BTR New Material Group and Shanshan, the two largest anode makers, both Chinese; Vianode (Elkem, Hydro, Altor) in Norway, built specifically around low-carbon hydro power; Novonix in Chattanooga, Tennessee; Anovion in Bainbridge, Georgia; Posco Future M in Korea; GrafTech and Showa Denko on the electrode side, which competes for the same needle coke; Phillips 66 and Mitsubishi Chemical as needle coke suppliers.

The cost stack is roughly needle coke plus power plus furnace throughput, and the power line is what separates producers. Chinese graphitization runs on industrial tariffs a Western plant cannot match except with hydro or a dedicated contract, so the Western pitch has to be a bundle: lower embodied carbon, supply security, and a price premium a cell maker will accept because its customer's regulator requires it. Needle coke prices are cyclical and correlated with EAF steel activity rather than with batteries, which adds a cost input that does not move with the anode market. Capital intensity is high for a materials plant and utilization is everything, since a graphitization furnace running at half load still pays for itself in fixed cost. If you are looking at an anode company, the two numbers that decide it are the delivered power price and whether there is a binding offtake from a cell maker that has already qualified the material.

15 minGraphite Electrodes Manufacturing (Production) Process. Производство графитированных электродовGraphite & Carbon · 50k+ views

8 minHow Is High-Purity Graphite Made and Where Is It Used?History of Simple Things · 50k+ views

Graphite in batteries (European Carbon and Graphite Association) · 2021–2024 Four-Year Review of Supply Chains for the Advanced Batteries Sector (US Department of Energy)

NdPr oxide is the material the entire permanent-magnet chain hangs on, and it comes out of three kinds of rock. Bastnäsite, a rare-earth fluorocarbonate, is the ore at Bayan Obo in Inner Mongolia and at Mountain Pass in California; at Bayan Obo the rare earths are recovered alongside an iron ore operation, which is part of why the economics work. Monazite, a phosphate, occurs at Mount Weld in Western Australia and as a heavy mineral in the sands mined for titanium and zirconium. The ore is concentrated by flotation, cracked with acid or caustic to break the mineral open, and leached into a mixed rare-earth solution — and then it has to be separated element by element, which is a different business covered two entries down. Neodymium and praseodymium usually travel together as didymium because magnet makers use them interchangeably in the alloy, and they typically make up only 15–25% of the rare-earth content of a deposit. The rest is mostly cerium and lanthanum, which are abundant, cheap, and hard to sell.

Mining light rare earths is not hard. The deposits are large, the ore is amenable to conventional flotation, and there are economic occurrences on every continent, which is why China's roughly 69% share of mine supply is the least concentrated link in the magnet chain. The problems start immediately after. Monazite carries thorium and uranium, so a monazite operation is handling a radioactive residue and needs a licensing path most Western jurisdictions treat as a nuclear matter rather than a mining one — this is what delayed Lynas in Malaysia for years. The economics have a second trap: separating out the NdPr you want leaves you holding four or five times as much cerium and lanthanum, and if you cannot sell those, their disposal cost lands on the NdPr. So the failure mode for a new light rare-earth mine is not that it cannot dig the rock. It is that it produces a basket, gets paid for a fraction of it, and has nowhere to send the separation work.

If you need NdPr and you want it from outside China, the feedstock is available now — Mountain Pass, Mount Weld, Nolans, and monazite from mineral sands are real sources you can contract. The question to ask a supplier is not where the mine is, it is where the separation happens, because a Western concentrate separated in China is a Chinese supply chain with extra shipping. If you are designing a product and NdPr price is your exposure, the levers in order of practicality are: cut the magnet mass through better motor design, move heavy rare earths out with grain-boundary diffusion, and only then consider a rare-earth-free magnet, which costs you power density. If you are underwriting a mine, check the basket: what fraction of revenue comes from NdPr, what the plan is for cerium and lanthanum, and whether there is a price floor or offtake underneath it.

China holds roughly 69% of world rare-earth mine supply, the least concentrated link in the magnet chain · NdPr is typically only 15–25% of the rare-earth content of a deposit · NdPr oxide has run roughly $50–175/kg over the past five years · it was about $74/kg in December 2025, near $120/kg in mid-2026, and about $97/kg on 4 August 2026 · the July 2025 Department of Defense agreement with MP Materials sets a $110/kg NdPr floor for ten years as a contract for difference, alongside a $400M equity stake · a new mine takes 10–18 years from discovery to production, but the separation plant behind it takes 3–6.

Not here, and saying so is the useful part. China mined roughly 69% of world rare earths in 2025. That is high, but it is the lowest share of any link in this chain, and it is falling as Mountain Pass, Mount Weld, and monazite from mineral sands ramp. The concentration tightens at every step after the mine: about 91% of separation and refining, and about 94% of sintered magnet output. What that means practically is that a mine outage is a price event and a separation outage is a production event. The nearest thing to a chokepoint at this stage is thorium: monazite's radioactivity is what determines whether a project can get permitted in the West, and it has added years to more than one program. The Western supply that exists is being held up by policy rather than by markets. The Department of Defense's July 2025 deal with MP Materials guarantees $110/kg for NdPr for ten years through a contract for difference, roughly double the trough price MP was competing against — and NdPr oxide fell below that floor again in early August 2026, which is exactly the situation the floor was written for.

Bayan Obo in Inner Mongolia (the largest rare-earth operation in the world, run alongside an iron ore mine); MP Materials' Mountain Pass in California; Lynas Rare Earths' Mount Weld in Western Australia, processed in Malaysia and increasingly at Kalgoorlie; Arafura's Nolans project in the Northern Territory; Iluka Resources' Eneabba refinery, built around monazite from mineral sands; Energy Fuels' White Mesa mill in Utah, which processes monazite alongside uranium.

There is no exchange contract for NdPr. Prices are assessments published by reporting agencies against Chinese domestic and export quotations, so the number in your model is an estimate of what a market you cannot trade in was paying. It moves hard: roughly $50–175/kg over five years, about $74/kg in December 2025, near $120/kg by mid-2026 on supply-security buying, and back to about $97/kg on 4 August 2026. Chinese production runs under a state quota system, which means the supply response to a high price is an administrative decision rather than a market one. A Western mine's economics come down to the basket: NdPr is most of the revenue, cerium and lanthanum are most of the mass, and the separation charge is the swing cost. That is why the deals getting financed all carry a floor price, an equity stake, or a committed offtake, and why a project pitching on spot economics alone should be read skeptically.

15 minRare Earths Are China’s Trump Card In The Trade War — How The U.S. Is Trying To Fix ThatCNBC · 500k+ views

Rare Earths, Mineral Commodity Summaries 2026 (USGS) · Rare-Earth Elements, Professional Paper 1802-O (USGS)

Dysprosium and terbium are the scarce end of the rare earths, and almost all of the world's supply comes from one unusual deposit type. Ionic adsorption clays are weathered granites in which rare-earth ions sit loosely bound to clay particles, so they can be washed out with an ammonium sulfate solution rather than cracked out of a hard mineral with acid at temperature. That makes them cheap to work and easy to mine informally, and it makes the tailings and leachate a serious pollution problem. The deposits that matter are in southern China (Jiangxi, Guangdong, Fujian) and across the border in Kachin State in Myanmar and in Laos. What the metals buy is heat resistance. Substituting a few percent of dysprosium or terbium for neodymium in a sintered magnet raises its coercivity, so the magnet keeps its magnetization above about 150 °C instead of losing it, which is exactly the condition inside an EV traction motor, a robot joint under load, or a missile fin actuator. Nothing else in the magnet does that job.

A few percent of dysprosium turns a magnet that fails in a hot motor into one that works, and it does it without much loss of remanence when it is applied by grain-boundary diffusion rather than alloyed through the bulk. That is a very good return on a small mass. The weakness is that there is very little of it and it comes from the worst possible places. Ionic clay mining in Myanmar operates inside an active conflict zone with no environmental control, so supply depends on which armed group holds the ground that month, and the ethical exposure for a Western buyer is real and largely undocumented. Even inside China these deposits are being depleted and the state has been consolidating and restricting them. The failure mode for a designer is quiet: a magnet grade specified with dysprosium that becomes unavailable is not a price problem, it is a redesign, because dropping to a lower grade changes the motor's thermal envelope and the fallback is usually a bigger, heavier machine or an active cooling loop.

Specify dysprosium or terbium when the magnet's working temperature genuinely exceeds what an N-grade neodymium magnet holds, which usually means a traction motor, an actuator in a hot bay, or anything with a long duty cycle and limited cooling. Before you do, ask for a grain-boundary-diffused grade, because it hits the same temperature rating with 50–80% less heavy rare earth and costs less than bulk alloying. If the application can tolerate it, design the heat out instead: better rotor cooling, a lower current density, or accepting a slightly larger magnet at a lower operating temperature all remove the requirement entirely. Reserve the highest heavy-rare-earth grades for parts where a thermal excursion is a safety issue. And if you sell into defense or automotive, treat the heavy content of every magnet in your bill of materials as a licensed item, because since April 2025 it is one.

Ionic clay deposits in southern China, Myanmar, and Laos supply nearly all world heavy rare earths · separation of the heavies is roughly 99% Chinese · dysprosium and terbium keep a sintered magnet working above about 150 °C · grain-boundary diffusion delivers the same temperature rating with 50–80% less heavy rare earth · Chinese heavy rare-earth imports from Myanmar fell about 50% in early 2025 during fighting in Kachin State · China's 4 April 2025 controls covered seven medium and heavy elements: samarium, gadolinium, terbium, dysprosium, lutetium, scandium, and yttrium.

This is the tightest link on the sheet on both axes at once. The clays are effectively a two-and-a-half-country resource, and separation of the heavies is roughly 99% Chinese, so there is no meaningful supply that does not pass through Chinese hands. Myanmar makes it worse rather than better: it is the largest single feed into Chinese heavy separation, and it is inside a war. Fighting between the junta and the Kachin Independence Army cut Chinese heavy rare-earth imports from Myanmar by about half in early 2025, and dysprosium and terbium prices moved on that alone. On 4 April 2025 China placed seven medium and heavy elements (samarium, gadolinium, terbium, dysprosium, lutetium, scandium, and yttrium) under export licensing, along with magnets containing them, and shipments stopped for long enough that automakers idled lines. The Western alternative is thin. Lynas began separating dysprosium and terbium in Malaysia in 2025, the first commercial production outside China; Serra Verde in Brazil is producing from ionic clay; Aclara and Ucore have projects. Together they are a small fraction of demand, and a new separation plant is a three-to-six-year build.

The ionic clay districts of Jiangxi and Guangdong provinces; Kachin State in Myanmar, which supplies a large share of the feed into Chinese heavy separation; Lynas Rare Earths, the first commercial producer of separated dysprosium and terbium outside China; Serra Verde in Goiás, Brazil, the first ionic clay operation outside Asia; Aclara Resources' Penco and Carina ionic clay projects in Chile and Brazil; Ucore's Louisiana separation facility; magnet makers such as Shin-Etsu and Proterial that publish low-dysprosium grades built on grain-boundary diffusion.

Dysprosium and terbium are assessed rather than exchange-traded, the volumes are small, and the market is opaque enough that quoted prices and transacted prices can diverge widely — terbium in particular trades in tonnes per year, not thousands of tonnes. Because supply comes from clays that are cheap to leach, the cost of production is low and the price is set almost entirely by policy and by conflict, which is why it moves in steps rather than along a cost curve. Demand is a small mass inside a large product: a few tens of grams of dysprosium in an EV motor is a rounding error on the vehicle's bill of materials until it is unavailable, at which point it stops the line. That asymmetry is the whole investment case for substitution work and for grain-boundary diffusion. If you are looking at a heavy rare-earth project, the number that decides it is separated-oxide output, not resource tonnes, and the buyer of record will usually be a government.

2 minKey differences between rare earths from ionic clays vs hard rockAclara Resources (TSX:ARA.TO) · 1k+ views

16 minIllegal Rare Earth Mining in Myanmar | The Index PodcastOrganized Crime Dispatch (by GI-TOC) · 5k+ views

Rare Earth Elements and U.S. Supply Chains (Congressional Research Service) · Global Critical Minerals Outlook 2025 (IEA)

Rare earths come out of the ground as a mixture and have to be pulled apart one element at a time, which is hard because they are chemically almost the same. Neighboring lanthanides differ only in the filling of an inner electron shell, so their separation factors in solvent extraction are barely above 1, and getting a clean product means repeating a slightly favorable separation over and over. In practice that is a cascade of a hundred or more mixer-settler stages, each one a tank where an aqueous rare-earth solution is contacted with an organic extractant, with the chemistry, pH, and flow ratios tuned element by element and deposit by deposit. Once you have a separated oxide it still is not a magnet input. The oxide is converted to fluoride and reduced to metal, either by molten-salt electrolysis (the usual route for neodymium and didymium) or by calciothermic reduction with calcium metal (used for dysprosium, terbium, and samarium-cobalt alloys), and the metal is then melted with iron and boron and strip cast into thin alloy flake. This is the step that actually decides who can make magnets.

Solvent extraction is old, well-understood chemistry, it runs at ambient temperature and pressure, and once a cascade is tuned it produces 99.9%-plus oxides continuously and cheaply. That is the strength. Everything else about it is a weakness for a new entrant. A cascade holds a large inventory of loaded organic and part-separated rare earth, so the working capital tied up in the plant is substantial and takes months to fill before you sell anything. It generates acidic raffinates and, on a monazite feed, thorium-bearing residues that need a licensed disposal path. Most importantly, the tuning is tacit knowledge (the published flowsheets tell you the principle, not the hundred adjustments that make a specific ore behave), and the people who have done it repeatedly are mostly in China. The failure mode is not an explosion, it is a plant that runs for two years at the wrong purity or the wrong recovery and never pays back the capital.

There is no when. If you want separated rare-earth oxides, this step happens, and if you want magnets, the metal and alloy steps happen too. The decision is who does it for you. If you are a buyer, the question to put to any supplier is where the separation and the metal-making occur, in that order, because a mine outside China whose concentrate is separated inside China gives you no supply security at all. If you are building, start from an existing permitted chemical site with acid handling and a residue disposal route, hire people who have run a cascade, and plan for three to six years and a long commissioning ramp. If you are investing, the asset worth owning in the rare-earth chain is a running separation plant with qualified output, not an ore body — the ore bodies are the easy part.

Neighboring lanthanides have solvent-extraction separation factors barely above 1, so a plant runs 100 or more mixer-settler stages in cascade · China holds about 91% of world separation and refining and a similar share of metal and alloy · oxide is reduced to metal by molten-salt electrolysis or calciothermic reduction · a Western separation plant is a 3–6 year build, against 2–4 years for an ordinary chemical refinery · MP Materials restarted Mountain Pass in 2017 and shipped concentrate to China for separation for years afterwards · there is no substitute for this step.

This is the chokepoint, and it is the reason a Western mine on its own solves nothing. China holds about 91% of world rare-earth separation and refining and a similar share of metal and alloy production, and it got there deliberately over roughly thirty years with cheap power, permissive permitting, and sustained state support through price cycles that would have bankrupted a private company. The clearest evidence of how binding it is: MP Materials restarted Mountain Pass in 2017 and still shipped its concentrate to China for separation for years, because owning a mine and owning a separation plant are different problems. Outside China, Lynas is the main separator, running Mount Weld feed through its Malaysian plant and increasingly through Kalgoorlie, with Solvay's La Rochelle plant, Neo Performance Materials in Estonia, and new US projects behind it. A disruption at this step does not raise the price of magnets; it stops them, because there is no inventory buffer between separated oxide and the magnet plant that consumes it. Building the alternative takes three to six years, a permitting path for thorium residues, and process know-how that is not written down.

China Rare Earth Group and Northern Rare Earth, the consolidated state entities that hold most Chinese separation capacity; Lynas Rare Earths' Kuantan plant in Malaysia and its Kalgoorlie cracking and leaching facility; MP Materials' separation circuit at Mountain Pass, commissioned after years of shipping concentrate to China; Solvay's La Rochelle plant in France, restarted for magnet-grade oxides; Neo Performance Materials in Estonia and its magnet plant in Narva; Less Common Metals in the UK, one of the few Western producers of NdFeB alloy strip; Energy Fuels' White Mesa mill in Utah.

Separation is a chemical conversion business with thin margins, high working capital, and heavy exposure to a price it does not set. Chinese separators operate under production quotas and have historically been willing to run through downturns, so the marginal cost that sets the world price is a Chinese cost including cheap power, cheap reagents, and residue handling a Western regulator would not permit. That is why every Western separation project in the current wave carries something artificial underneath it: a Department of Defense contract for difference, an equity stake, a guaranteed offtake, or a customer paying a documented premium for non-Chinese oxide. Capital cost per tonne is lower than a mine's but the build is slower than a refinery's, three to six years against two to four, because permitting dominates. If you are evaluating one of these companies, the things that matter are the permit, the power price, the residue plan, and whether anyone on the team has commissioned a cascade before.

2 minSeparating Rare Earth Elements (Byte-sized Science)Simply Science · 50k+ views

2 minCritical Materials Recovery - Solvent ExtractionIdaho National Laboratory · 1k+ views

From minerals to materials, Supplementary report: Rare earths (CSIRO) · Rare Earth Elements: A Review of Production, Processing, Recycling, and Associated Environmental Issues (US EPA)

A sintered neodymium-iron-boron magnet is the strongest permanent magnet made at commercial scale, with energy products in the 35–50 MGOe range, and it is a powder-metallurgy part rather than a cast one. The alloy is strip cast into thin flake to control grain structure, then charged with hydrogen, which forces its way into the lattice and shatters the flake into coarse powder — hydrogen decrepitation. That powder is jet milled to a few microns, aligned in a magnetic field while it is pressed so the grains' easy axes point the same way, sintered near 1,080 °C, annealed, machined to final dimensions, and coated, usually nickel-copper-nickel or epoxy, because the alloy corrodes readily. Most automotive and industrial grades then get grain-boundary diffusion: dysprosium or terbium is painted on the surface and diffused along the grain boundaries where it does the most good, which buys the same high-temperature performance as bulk alloying while using 50–80% less heavy rare earth. The finished magnet is where all the upstream concentration in this chain finally lands.

Nothing else gives you this much field in this little mass, which is why NdFeB is in essentially every EV traction motor, every hard drive, every robot joint, and every direct-drive wind turbine. The design consequence is that motors built around it are smaller and lighter than any alternative at the same torque, and swapping the magnet out is a machine-level redesign rather than a part substitution. The weaknesses are all physical and all well known: the alloy corrodes, so it must be coated and the coating is a warranty item; it is brittle and chips in handling; and its magnetization falls with temperature, so a grade that works at 80 °C may fail at 150 °C. That last one is what drives the heavy rare-earth content. The failure mode that has actually bitten manufacturers is not technical, though. It is a magnet grade specified with dysprosium in it becoming an export-licensed item, which happened in April 2025 and idled vehicle assembly lines within weeks.

Default to sintered NdFeB whenever torque density or power density is the binding constraint and the operating temperature is manageable: traction motors, servo and robot actuators, generators, precision positioning. Specify the lowest grade that meets your thermal envelope, and ask for grain-boundary diffusion rather than a bulk heavy rare-earth alloy, because it gets you the same rating for a fraction of the dysprosium. Go to bonded NdFeB when the part is small and complex enough that machining a sintered block is wasteful, and accept roughly half the energy product. Go to ferrite or samarium-cobalt only when you cannot use NdFeB at all: the temperature is above what it holds, corrosion resistance without coating is required, or rare-earth exposure is itself the thing you have been told to eliminate. Read the next entry before you commit to that, because the penalty is larger than it sounds.

Energy product 35–50 MGOe · about 94% of world sintered magnet output is Chinese · strip cast, hydrogen decrepitated, jet milled to a few microns, aligned and pressed in a field, sintered near 1,080 °C, then machined and coated · grain-boundary diffusion cuts heavy rare-earth content by 50–80% for the same temperature rating · 1–3 kg of magnet per EV traction motor, a few kg spread across a humanoid robot's joints, and hundreds of kg per MW in a direct-drive wind turbine · MP Materials' 10X campus in Northlake, Texas is a $1.25B build commissioning from 2028 toward roughly 10,000 t/yr.

This is the most concentrated step on the whole sheet: about 94% of world sintered magnet production is Chinese. Everything upstream compounds into it, since a magnet plant needs separated oxide, metal, and alloy, and those are roughly 91% Chinese too. When China's April 2025 controls covered magnets containing the seven listed medium and heavy elements, shipments stopped while licenses were processed, and automakers in the US, Europe, and India idled lines within weeks — that is what a disruption at this step looks like, and it took about a quarter to work through. The Western build is finally real and it is slow. MP Materials' Independence plant in Fort Worth began commercial metal production in 2024 and made its first alloy flake and finished magnets on commercial equipment in 2025; its 10X campus in Northlake, Texas is a $1.25 billion project commissioning from 2028 and taking MP toward roughly 10,000 t/yr, with the output committed to the Pentagon for ten years. e-VAC is building in Sumter, South Carolina, Noveon runs in San Marcos, Texas, and Neo Performance Materials has a plant in Estonia. Total announced non-Chinese capacity is on the order of tens of thousands of tonnes a year against a world market several times that.

JL MAG, Ningbo Yunsheng, and Zhongke Sanhuan, the largest Chinese magnet makers; Shin-Etsu and Proterial in Japan, the main non-Chinese incumbents and the source of much of the grain-boundary-diffusion patent estate; MP Materials' Independence plant in Fort Worth and the 10X campus in Northlake, Texas; e-VAC Magnetics in Sumter, South Carolina, supplying General Motors; Noveon Magnetics in San Marcos, Texas; Neo Performance Materials' plant in Narva, Estonia; Vacuumschmelze in Germany.

Magnet pricing is roughly the rare-earth content at prevailing oxide prices plus a conversion charge, so the magnet price tracks NdPr with a lag and the maker's margin is the conversion piece. That structure means a magnet plant does not make money on rare-earth price spikes; it makes money on utilization, yield, and the grades it can hit. Chinese producers have scale, an integrated supply of alloy, and enough capacity that they can compete on price whenever a Western entrant appears, which is why every Western plant announced since 2022 has a long-term offtake, a government contract, or both underneath it. The unit economics are also sensitive to machining scrap: a sintered block cut to a finished shape can lose a substantial fraction of its mass, and whether that swarf is recovered in-house changes the cost meaningfully. If you are evaluating a magnet business, look for a signed multi-year offtake at a stated price, a secured alloy supply, and evidence that the grain-boundary-diffusion process is qualified, because that is the grade automotive actually buys.

5 minRare Earth Magnets - HOW they're madeMP Materials · 50k+ views

9 minHow the World’s Strongest Magnets Are Made (Neodymium Explained)Worldnite Journey · 100k+ views

9 minNeodymium Magnets: How They Work and Why They’re Everywhere!History of Simple Things · 100k+ views

Manufacture of Modern Permanent Magnet Materials (Arnold Magnetic Technologies) · Rare Earth Magnets and Motors: A European Call for Action (European Raw Materials Alliance)

These are the two magnets you reach for when neodymium-iron-boron is not available or not suitable, and they sit at opposite ends of the price range. Ferrite, also called ceramic, is strontium or barium hexaferrite: iron oxide mixed with a carbonate, calcined, wet milled, pressed in a magnetic field, and sintered. It contains no rare earths and no cobalt, the raw materials are among the cheapest in industry, and it has been made in enormous volumes for loudspeakers, small motors, and holding applications for sixty years. Samarium-cobalt is the opposite: a sintered rare-earth magnet in either the SmCo5 or the Sm2Co17 family, made by much the same powder route as NdFeB but with samarium in place of neodymium and cobalt making up roughly half to two-thirds of the mass. It gives up energy product to gain temperature capability and corrosion resistance. The whole point of this entry is what the substitution costs, because both of these are genuinely available alternatives and neither is free.

Ferrite's strengths are price and supply: a few dollars a kilogram against tens for NdFeB, no rare earths, no cobalt, no corrosion coating needed. Its weakness is that its energy product is roughly 3.5–5 MGOe against 35–50 for NdFeB, so for the same air-gap flux you need far more magnet volume and a heavier rotor, and the machine grows. It also has an unusual failure mode: ferrite loses coercivity as it gets colder rather than hotter, so a ferrite motor can irreversibly demagnetize on a cold start, which is a real automotive constraint rather than a footnote. Samarium-cobalt keeps working to 300–350 °C where NdFeB has long since given up, holds its magnetization with very little temperature drift, and resists corrosion well enough that it often ships uncoated. Its weaknesses are that it costs more than NdFeB per unit of energy, it is brittle and chips and cracks in machining and assembly, and it carries two supply exposures instead of one — samarium went onto China's export control list in April 2025, and cobalt is the material in entry 003.

Pick ferrite when volume and mass are cheap and cost is the binding constraint: pumps, fans, appliance motors, holding and separation, and traction motors where the designer has explicitly accepted a larger machine to eliminate rare earths. If you go that way, check the cold-temperature demagnetization limit early, because it constrains the design more than the room-temperature numbers suggest. Pick samarium-cobalt when the part sees temperatures above roughly 150–180 °C, when the environment is corrosive and a coating cannot be relied on, or when magnetic output has to stay stable across a wide temperature range — missile and aircraft actuators, downhole tools, high-temperature sensors, and traveling-wave tubes. Do not pick either one as a supply-chain fix on its own. Both are still made overwhelmingly in China, SmCo swaps neodymium exposure for samarium and cobalt exposure, and only ferrite actually removes rare earths from the bill of materials.

Ferrite energy product roughly 3.5–5 MGOe, samarium-cobalt 16–32 MGOe, sintered NdFeB 35–50 MGOe · SmCo works to 300–350 °C against roughly 150–200 °C for automotive NdFeB grades · ferrite costs a few dollars a kilogram against tens of dollars for NdFeB · SmCo is roughly half to two-thirds cobalt by mass · ferrite loses coercivity as temperature falls, so cold-start demagnetization is the design limit · samarium was one of the seven elements China placed under export licensing on 4 April 2025.

Swapping magnet chemistry does not move the country. China makes the large majority of the world's ferrite magnets as well as its neodymium ones, because ferrite is a low-margin, high-volume, energy-intensive ceramic business that Western producers exited decades ago, and the samarium-cobalt supply chain runs through the same separation plants as everything else in entry 009. Samarium specifically was named in China's 4 April 2025 export controls, so the "rare-earth-free" pitch does not apply to SmCo at all. What ferrite does buy is a different kind of exposure: its inputs are iron oxide and strontium or barium carbonate, which are cheap, widely available, and not export controlled, so a ferrite plant can in principle be built anywhere on ordinary industrial timelines of two to three years, with no separation step behind it. That is a much shorter path than the three to six years a rare-earth separation plant takes. The cost of walking it is a heavier machine, and that is the trade a buyer has to be willing to make before the alternative supply chain is worth anything.

Tesla said at its 2023 Investor Day that its next-generation drive unit would use a permanent-magnet motor containing no rare earths; it has not named the material publicly, and ferrite is the usual assumption. TDK, Proterial, and Hitachi Metals in ferrite, alongside a large Chinese industry led by producers such as BGRIMM and Sinomag. Arnold Magnetic Technologies, Electron Energy Corporation, and Vacuumschmelze in samarium-cobalt for aerospace and defense. SmCo is standard in missile fin actuators, aircraft generators and starter-generators, downhole logging tools, and traveling-wave tube focusing magnets. Niron Magnetics is developing iron-nitride magnets as a third rare-earth-free option, targeting energy products above ferrite's, and is not yet shipping at scale.

Ferrite is a commodity: prices track iron oxide, strontium carbonate, and industrial power, they move within a narrow band, and margins are thin enough that the business only works at very high volume with cheap energy. Samarium-cobalt is the opposite — small volumes, long-qualified aerospace and defense customers, prices set in bilateral contracts, and a cost that swings with cobalt and samarium rather than with any exchange. Because samarium is a low-value byproduct of separating the more wanted rare earths, its supply is set by NdPr demand, so an SmCo buyer is exposed to a market it does not participate in. Blended across the two, price movement over the past five years has been moderate compared with the neodymium chain. The commercial question for anyone designing around these is straightforward: ferrite trades capital and mass for supply security, SmCo trades money for temperature, and neither is a cheaper version of NdFeB.

3 minHow to produce ferrite magnetic arc, magnet?Zhengxi Press Channel · 10k+ views

9 minHOW IT'S MADE: MagnetsHow It's Made · 100k+ views

Permanent Magnet Materials and Current Challenges (Arnold Magnetic Technologies) · Progress and prospects of hard hexaferrites for permanent magnet applications (Journal of Physics D: Applied Physics)

Copper is the default electrical conductor, at roughly 100% IACS conductivity, ductile enough to draw into hair-thin wire and cheap enough to use by the tonne. World mine production was about 23 million tonnes in 2025. Most of it starts as sulfide ore at 0.4–1% copper, which is crushed, ground, and floated into a concentrate running 20–30% copper, then smelted to blister and electrolytically refined to 99.99% cathode. Oxide ores take a shorter route: heap leach, solvent extraction, and electrowinning straight to cathode with no smelter involved. Refined output was about 29 million tonnes in 2025, more than mine supply, because scrap feeds the same refineries. This is the volume material on this sheet. Most of the other entries are measured in thousands of tonnes a year, and several of them in kilograms.

Copper's conductivity, corrosion behavior, formability, and infinite recyclability are why it has never been displaced from motor windings, busbars, or building wiring. It is also one of the few materials here with a liquid exchange market, so a buyer can hedge it. The weakness is not scarcity. Ore grades have fallen for decades, the average new project is lower-grade and deeper than the one it replaces, and permitting a greenfield mine takes 10–18 years from discovery to production, so supply cannot respond inside a demand cycle. The failure mode that actually bites is at the smelter: treatment and refining charges collapsed below zero through 2025, which means smelters were paying miners for concentrate, and a business model that only works with negative input costs eventually stops.

Use copper wherever current density, space, or joint reliability decide the design: motor and transformer windings, high-current connectors, PCB traces, heat exchangers, and anything that gets crimped or soldered in the field. Switch to aluminum where the constraint is weight or dollars per meter — overhead transmission, large busbars, long feeder runs, and vehicle harnesses have all moved that way. The penalty is real and predictable: aluminum conducts at about 61% of copper's rate, so you size up roughly 1.6x in cross-section, and you have to manage creep and oxide at every termination. If you are building a product where copper is 60–80 kg per unit, as in an EV, treat the copper price as a design input rather than a line item, because a doubling changes the economics more than most component choices do.

Mine production about 23 Mt in 2025, refined output about 29 Mt · Chile roughly 5.3 Mt of mine supply, China about 14 Mt of refined · concentrate grades 20–30% copper from ore at 0.4–1% · LME cash averaged about $9,700/t in 2025 and traded above $14,000/t in August 2026 · aluminum at roughly 61% of copper's conductivity · 60–83 kg per EV against about 20 kg per combustion car · scrap supplies roughly 30% of US copper.

Smelting and refining, not mining. Mine supply is genuinely spread out — Chile about 23%, the DRC 14%, Peru 12%, China 8%, and the US, Russia, Zambia, and Indonesia behind them — and that is the only diversified mine supply on this sheet. Refining is a different picture: China produced roughly 14 million of the world's 29 million tonnes of refined copper in 2025, about 48%, and it has kept building smelters. The evidence that capacity has outrun concentrate is the treatment and refining charge, which is the fee a smelter deducts for turning concentrate into metal. The 2025 annual benchmark settled at $21.25 per dry tonne, and spot charges spent most of the year below zero, reaching roughly minus $60 per tonne in November 2025. A disruption here does not stop copper reaching the world; it strands concentrate at ports and forces Western fabricators to buy cathode from the country they are trying to diversify away from. The Western alternative is slow. The US has two primary smelters, a new one takes about 4–6 years to permit and build, and it needs somewhere to sell sulfuric acid, which is usually the reason the project does not happen.

Escondida (BHP, Chile) and Collahuasi are the largest single mines; Grasberg (Freeport, Indonesia) had a bad 2025 and Indonesian output fell by roughly 300,000 t; Kamoa-Kakula (Ivanhoe Mines and Zijin, DRC) is the highest-grade major project built this century. On the smelting side, Jiangxi Copper's Guixi complex is the world's largest, Aurubis Hamburg and Atlantic Copper Huelva are the European reference plants, and Rio Tinto Kennecott in Utah and Freeport's Miami smelter in Arizona are the two US primaries. Codelco remains the largest single producer by company. Congo moved to restrict concentrate exports in 2026, which is one of the reasons the price ran to records that year.

Copper is priced on the LME and COMEX, which puts it in a different category from most of this sheet: the number is real, it is hedgeable, and forward curves exist. The 2025 LME cash average was about $9,700/t, and prices ran above $14,000/t in August 2026 on data-center and grid demand against static mine supply. Mining margins follow the cost curve, and the industry's economics rest on byproduct credits — gold, molybdenum, and silver often decide which mines are in the bottom quartile. Smelting is the opposite business: a fee-for-service industry with heavy capital, low margins, and no pricing power, which is why Chinese smelters pledged output cuts for 2026 rather than continue at negative treatment charges. Secondary supply already carries about 30% of US consumption, and scrap is the cheapest new tonne available to anyone who can build a collection network. If you are underwriting a copper project, the two things that decide the answer are grade decline in the existing fleet and whether the permitting jurisdiction can actually deliver a mine inside 15 years.

18 minHow Copper Is Made — From Rock to 99.9% Pure MetalBuildWitt · 500k+ views

5 minMetso Flash SmeltingMetso · 10k+ views

5 minHow Copper is made animation | Karthi ExplainsKarthi Explains · 50k+ views

The World Copper Factbook 2025 (International Copper Study Group) · Copper, Mineral Commodity Summaries 2026 (USGS)

Aluminum reaches metal through three stages that sit in three different places. Bauxite is strip-mined and shipped as ore, about 440 million tonnes in 2025. The Bayer process digests it in hot caustic soda to make alumina, about 150 million tonnes, at a ratio near 2.5 tonnes of bauxite per tonne of alumina. Hall-Héroult smelting then dissolves that alumina in molten cryolite at around 960 °C and electrolyzes it to metal, roughly 74 million tonnes in 2025, at a ratio near 1.9 tonnes of alumina per tonne of aluminum. The real input at the last stage is electricity, 13–15 MWh per tonne, drawn continuously, because a potline that loses power for a few hours freezes solid and is effectively destroyed. A smelter is a power contract with a potline attached, which is why they sit next to hydro dams, cheap coal, or stranded gas rather than next to the ore.

Aluminum gives you a third of steel's density, good corrosion resistance from its own oxide film, easy extrusion, and about 61% of copper's conductivity at roughly half the weight for the same current. It also recycles on about 5% of the energy of primary metal with no loss of properties, which is why secondary supply keeps taking share. The weaknesses come from the same electricity dependence that defines the process. Power price sets the cost curve, so a smelter in a market with volatile electricity is a stranded asset waiting for a bad winter, which is what happened to several European smelters during the 2022 energy spike. Terminations are the engineering weak point: aluminum creeps under clamp pressure and grows an insulating oxide, so aluminum wiring fails at joints rather than in the middle of a run.

Pick aluminum over copper when weight or cost per meter dominates and you control the joints: overhead transmission lines, large busbars, heat sinks, vehicle structures, and packaging. Pick it over steel when corrosion resistance and weight matter more than stiffness and cost, and remember that aluminum's modulus is about a third of steel's, so a like-for-like substitution deflects three times as much. Stay with copper in motor windings, fine-pitch connectors, and anything field-terminated by people who will not torque to spec. If you are sourcing rather than designing, the question that matters is which stage you are exposed to: bauxite, alumina, and primary metal have different geographies and different failure modes, and a company that has secured metal supply has not necessarily secured alumina.

Bauxite about 440 Mt in 2025, alumina about 150 Mt, primary aluminum about 74 Mt · roughly 2.5 t bauxite per tonne of alumina and 1.9 t alumina per tonne of metal · smelting draws 13–15 MWh/t · China about 45 Mt of primary output against a 45 Mt capacity cap · Guinea roughly a third of world bauxite · recycled aluminum uses about 5% of primary energy · US import prices averaged about $31/t for bauxite and $595/t for alumina in 2025.

There are two, and they tighten in sequence. Alumina refining is about 62% Chinese — roughly 93 million of 150 million tonnes in 2025, with Australia a distant second near 17 million. Primary smelting is about 61% Chinese, roughly 45 million tonnes, and it is capped: Beijing has held a 45 million tonne ceiling on domestic primary capacity for years, and 2025 output finally ran into it. That cap is the single most consequential fact in the aluminum market, because it means the supplier that sets the world price can no longer grow supply on demand, and every additional tonne has to come from Indonesia, India, the Gulf, or a restart in Europe or North America. Upstream of both, Guinea supplies roughly a third of world bauxite, and Guinean policy moves the whole chain: the government revoked one large concession in August 2025 after a dispute over an unbuilt refinery. The Western alternative is uncomfortable. A greenfield smelter takes 4–6 years and needs a decade-long power contract at a price most Western grids cannot offer, so the realistic Western response is restarts and recycling rather than new primary capacity.

Guinea's Boké region and Australia's Weipa and Gove are the reference bauxite operations; Rio Tinto, Alcoa, and Emirates Global Aluminium are the integrated Western producers. Chalco and Hongqiao dominate Chinese alumina and metal. Alcoa's San Ciprián smelter in Spain restarted in 2025 after stopping in 2022, and a 500,000 t/yr smelter in North Kalimantan, Indonesia was due to start around the same time. On the recycling side, new large-scale flat-rolled recycling mills in the US south are the model most Western capacity additions now follow.

Aluminum trades on the LME, so the metal price is transparent, but the margin is set upstream by alumina and power. Alumina has its own assessed price and its own squeezes; US import prices averaged about $595/t in 2025, and a refinery outage can move it far faster than metal. Cash cost for a primary smelter is roughly 30–40% electricity, so the cost curve is really a power-price curve, and a producer with a legacy hydro contract is structurally advantaged in a way no operational improvement can match. Secondary aluminum is the cheapest supply available, and its share keeps rising because the energy saving is close to 95%. If you are building a business on aluminum, the durable positions are a long-dated cheap power contract, a permitted alumina refinery, or a scrap collection and sorting network. Owning bauxite alone is the weakest position in the chain, since the ore is abundant and the value is added twice downstream of it.

7 minHave you ever wondered how aluminium is made?Emirates Global Aluminium · 100k+ views

4 minBayer Process of Aluminium | Purification of BauxiteThe Science Chef Academy · 10k+ views

3 minExtraction Of Aluminium Using Electrolysis | Environmental Chemistry | FuseSchoolFuseSchool - Global Education · 100k+ views

Aluminum, Mineral Commodity Summaries 2026 (USGS) · Bandwidth Study on Energy Use and Potential Energy Saving Opportunities in U.S. Aluminum Manufacturing (US Department of Energy)

Titanium ore is abundant and cheap; titanium metal is neither, and the gap between the two is the Kroll process. Ilmenite and rutile are mined from mineral sands at about 9.4 million tonnes of ilmenite a year, mostly for white pigment, and cost tens of dollars a tonne as concentrate. To make metal, the mineral is chlorinated to titanium tetrachloride, purified by distillation, and then reduced with molten magnesium at around 800–900 °C inside a sealed steel retort. The reaction runs for several days and produces a porous mass called sponge, which is crushed, blended with alloying elements and scrap, and vacuum-arc remelted twice or three times into an ingot. Every step is a batch, every batch is slow, and the magnesium chloride byproduct has to be electrolyzed back to magnesium and chlorine to make the economics work at all. Kroll has resisted replacement since the 1940s despite repeated attempts at continuous electrolytic routes.

Titanium's strength-to-weight ratio, fatigue behavior, and corrosion resistance in seawater and chlorides are why it holds airframe fittings, engine fan blades and discs, and chemical plant internals. It also sits happily against carbon-fiber composites, which corrode aluminum, so composite airframes pull titanium content up. The weakness is process cost: sponge lands around $12/kg, and by the time it is remelted, forged, and machined at buy-to-fly ratios that can exceed 10:1, a finished aerospace part costs many times its raw material. The failure mode for a buyer is qualification rather than price. Aerospace-grade sponge and the melt shops behind it are approved plant by plant against specific specifications, so nominal world capacity of about 470,000 t/yr badly overstates what a given airframer can actually buy this year.

Choose titanium when you need strength at low weight in a corrosive or hot environment and you can carry the machining cost: landing-gear and pylon fittings, hydraulic tubing, compressor sections, heat exchangers in seawater, and implants. Skip it where stiffness rather than strength governs, since titanium's modulus is barely half of steel's and a composite or a steel part will usually be lighter for a stiffness-driven design. Skip it too where a coated steel or a nickel alloy solves the corrosion problem for less. If your program is defense or aerospace, treat second-sourcing sponge as a two-to-three-year project rather than a purchasing decision, and check where your melt shop's sponge actually comes from, because the answer for most Western buyers is Japan.

World sponge production about 370,000 t in 2025 against roughly 470,000 t of capacity · China about 260,000 t, Japan about 53,000 t, Russia about 25,000 t, Kazakhstan about 16,000 t · ilmenite mine production about 9.4 Mt with China roughly a third · Kroll reduction with molten magnesium at 800–900 °C, several days per batch · US sponge imports about 44,000 t in 2025 at roughly $12/kg, all imported · Japan supplied about 73% of those imports.

Sponge production, and behind it magnesium. China made about 260,000 of the world's 370,000 tonnes of titanium sponge in 2025, roughly 70%, with 320,000 t of capacity against Japan's 65,200 t. The United States produced none: the last domestic plant closed in 2024, and two larger idled facilities in Nevada and Utah have been down since 2020 and 2016. That leaves Western aerospace resting on Japan (Toho Titanium and Osaka Titanium), Kazakhstan's UKTMP, and Saudi Arabia, which together supplied essentially all of the 44,000 t the US imported in 2025. Russia's VSMPO-AVISMA was historically the leading exporter of aerospace-grade sponge and mill products until 2022 sanctions and self-sanctioning rerouted Western buying, which is the clearest demonstration that this chokepoint can move. A disruption in Japan would not be solved by restarting a mothballed US plant, because the restart is 2–3 years and the aerospace qualification behind it is longer. The deeper problem is that Kroll consumes magnesium, and magnesium is more than 80% Chinese, so titanium sits downstream of another chokepoint.

Toho Titanium and Osaka Titanium Technologies (Japan) are the reference aerospace sponge producers; UKTMP (Kazakhstan) and Toho's Saudi joint venture supply most of the rest of the Western market. VSMPO-AVISMA (Russia) still holds the largest integrated sponge-to-forging capacity in the world. On the mill side, ATI, Howmet Aerospace, and TIMET serve Boeing and Airbus programs; the 787 and A350 both carry high titanium fractions because of their composite structures. IperionX in Virginia began commercial production of titanium powder from scrap in late 2024, aiming at 1,400 t/yr by 2027, which is a scrap-based route rather than a Kroll replacement.

Titanium sponge is sold on multi-year contracts rather than an exchange, so there is no hedge and prices move in steps: US import sponge averaged about $11–13/kg from 2021 through 2025. That stability is deceptive, because the aerospace market clears on availability rather than price during an upcycle, and lead times stretch long before quoted prices move. Capital intensity is high and utilization matters enormously, since a reduction retort produces nothing while it is cooling. Money in this chain is made downstream, at the melt shop and the forge, where qualification and scrap recycling loops are the barriers to entry; sponge itself is close to a commodity with Chinese cost leadership. If you are financing a Western sponge project, the question is not whether you can build the plant but whether an airframer will sign a decade-long offtake at a price that covers Western power and labor, and the historical answer has usually been no.

5 minTitanium: Kroll MethodInnovations in Manufacturing at ORNL - Archived · 100k+ views

9 minHow Is Titanium Made?History of Simple Things · 1m+ views

16 minHow Millions Of Tons Of Titanium Are Made - From Ore To The Strongest MetalThe Factoran · 100k+ views

Titanium and Titanium Dioxide, Mineral Commodity Summaries 2026 (USGS) · Electrolytic Routes to Titanium: Methodological Innovations, Key Findings, and Prospects for Sustainable Production (Materials)

Magnesium is the lightest structural metal, about two thirds the density of aluminum, and it is made in two very different ways. Most of the world's supply comes from the Pidgeon process: calcined dolomite is mixed with ferrosilicon, packed into externally heated steel retorts, and reduced under vacuum at around 1,200 °C, with magnesium vapor condensing out over several hours per charge. It is labor-intensive, coal-fired, and energy-hungry at roughly 35–40 MWh per tonne, which is why it migrated to places with cheap coal and cheap labor. The alternative is electrolysis of magnesium chloride from brine or seawater, which is cleaner and continuous but capital-heavy, and outside China it survives only in a handful of plants. World primary output was about 1.1 million tonnes in 2025. Most of that magnesium never becomes a magnesium part: it goes into aluminum alloys as an alloying element, into steel desulfurization, and into the Kroll process as a reductant.

As a die-casting alloy magnesium gives thin walls, excellent castability, good damping, and the lowest weight of any structural metal, which is why steering-wheel armatures, seat frames, and instrument-panel beams are magnesium. As an alloying element it is close to irreplaceable: essentially every 5xxx and 6xxx aluminum alloy needs a few percent magnesium to reach its strength, and there is no substitute that gives the same result. The weaknesses are galvanic corrosion against almost every other structural metal, low modulus and creep resistance at temperature, and flammability of fines during machining. The failure mode for a buyer, though, is not metallurgical. It is that supply sits in a small number of counties in one province of one country, and every downstream aluminum alloy and titanium plant in the world depends on it.

Use magnesium castings where mass matters more than stiffness and the part lives in a dry, protected environment: interior automotive structures, portable tool and camera housings, aerospace gearbox casings. Do not use it in a wet, salted, or dissimilar-metal-contact environment without a coating strategy you have actually validated, because galvanic corrosion at fasteners is the usual field failure. Prefer aluminum when you need corrosion tolerance or stiffness, and prefer engineering plastics when you need cheapness and complexity. As a purchaser, treat magnesium differently from the metals you can substitute your way out of: you cannot design magnesium out of an aluminum alloy specification, so the right response is inventory and a second supplier, not a materials review.

World primary output about 1.1 Mt in 2025 against roughly 1.8 Mt of capacity · China about 950,000 t, roughly 86% of supply, against roughly 60% of world magnesite mining · Pidgeon retorts consume roughly 35–40 MWh/t · European free-market price averaged about $2,500/t in 2025 against roughly $5,000/t in 2021 and a spike near $10,000/t in late 2021 · a few percent magnesium in essentially every 5xxx and 6xxx aluminum alloy · about 69% of US primary consumption goes into castings.

Primary production, and it is the most concentrated single-country dependency on this sheet outside the rare-earth chain. China made roughly 950,000 of the world's 1.1 million tonnes in 2025, about 86%, and most of that comes from Pidgeon retorts clustered in a handful of counties in Shaanxi province. The rest of the world's output is small and scattered: Russia around 60,000 t, Brazil and Israel around 20,000 t each, Turkey and Kazakhstan smaller still. The rock is a milder problem than the metal. China mined about 12.7 million of the world's 21 million tonnes of magnesite in 2025, roughly 60%, and Pidgeon retorts run on dolomite anyway, which is abundant almost everywhere. The concentration is in the reduction step. What a disruption looks like is on the record. In autumn 2021, Chinese power curtailment under the "dual control" energy policy shut a large share of Shaanxi's retorts, the European price went from roughly $2,000/t to about $10,000/t in weeks, and European automotive and aluminum producers publicly warned they were weeks from stopping production. Nothing structural changed afterward. The US had one primary smelter, in Utah, which stopped producing in 2022 and whose operator filed for Chapter 11 in September 2025. A brine-based plant in Arkansas is in development with Defense Production Act funding and produced pilot-scale metal in July 2025, but a commercial plant is several years out, and even at full size it would not displace a meaningful share of Chinese supply.

Shaanxi's Fugu and Shenmu county retort clusters are where most world supply physically comes from. US Magnesium in Utah was the last US primary producer; Dead Sea Magnesium (Israel) and Rima Industrial (Brazil) are the surviving non-Chinese electrolytic and thermal plants. Magrathea Metals is the Defense Production Act-funded brine project that produced qualifying pilot metal in 2025. On the demand side, the 2021 shortage hit Volkswagen, BMW, and the European aluminum extruders hardest, and it is still the case study every automotive purchasing organization cites when arguing for strategic inventory.

Magnesium is assessed rather than exchange-traded, with a European free-market price and a US spot Western price that carry a large and persistent duty-driven gap: the US price ran near $3.20/lb in 2025 against a European average around $2,500/t. Cost is dominated by coal or electricity and by ferrosilicon, so Chinese producers set the marginal cost and a Western plant has to beat it on power or not at all. The market is small in dollar terms, on the order of a few billion a year, which is exactly why nobody built redundancy: the cost of a shortage falls on the aluminum and automotive industries downstream, and none of them are large enough buyers of magnesium itself to justify funding a plant. Recycling helps at the margin — about 26,000 t of US secondary magnesium came from old scrap in 2025 — but scrap returns follow the die-casting fleet rather than the alloying demand that dominates consumption. If you are underwriting a Western magnesium project, the only credible thesis is a government-backed floor price or an offtake from a buyer that has already been burned once.

17 minMaking Magnesium MetalScrap Science · 10k+ views

8 minBeyond Strength and Lightness: How Magnesium Alloys Are MadeHistory of Simple Things · 10k+ views

Magnesium Metal, Mineral Commodity Summaries 2026 (USGS) · Magnesium factsheet (SCRREEN2)

Silicon starts as quartzite, a common rock, and gains value entirely through purification. Lump quartzite and a carbon reductant (coal, charcoal, wood chips) are fed into a submerged-arc furnace at around 1,800 °C, where the carbon strips the oxygen and metallurgical-grade silicon at 98–99% purity is tapped off. That grade is good enough for aluminum alloys and silicones, which take most of it, but nowhere near good enough for a wafer. To go further, the silicon is converted to trichlorosilane and run through the Siemens process, where the gas decomposes onto heated silicon filaments inside a bell jar and grows polysilicon rods over several days. Solar grade comes out around 9N and semiconductor grade around 11N, meaning fewer than a few parts per trillion of the impurities that matter. Fluidized-bed reactors make granular polysilicon more cheaply at solar grade. Each additional nine is a separate process step, and the two grades behave like two separate industries.

Silicon is abundant, non-toxic, forms a stable native oxide that made the whole planar transistor possible, and it is the only material with a mature multi-trillion-dollar manufacturing base behind it. There is no substitute for it in either chips or crystalline solar cells. The weaknesses are all in the purification chain rather than the element. Metallurgical silicon production is electricity-hungry and produces carbon dioxide directly from the reduction, and Siemens polysilicon consumes large amounts of power again. The industry's failure mode is not scarcity but violent cyclicality: polysilicon is capital-intensive with high fixed costs, so producers run flat out through downturns and sell below cash cost, which is exactly what happened through 2025 and into 2026.

There is no material decision to make here — if you are making chips or crystalline silicon PV, you buy polysilicon. The decisions are about grade and sourcing. If you need semiconductor grade, your supplier list is short and mostly non-Chinese, and you should expect to qualify a plant rather than a product. If you need solar grade, your supply is overwhelmingly Chinese, and the questions are traceability and import law rather than availability. If you are selling into the US solar market, plan polysilicon sourcing around the Uyghur Forced Labor Prevention Act from the start, because a module held at the border is a total loss and the documentation burden runs back to the quartz. Thin-film PV using cadmium telluride is the only route that avoids polysilicon entirely, and it trades away efficiency and supplier choice.

Metallurgical silicon about 98–99% pure from a submerged-arc furnace at roughly 1,800 °C · solar polysilicon at about 9N and semiconductor grade at about 11N · world silicon metal output about 4.6 Mt in 2025, with China roughly 4.0 Mt · China roughly 80% of all silicon materials including ferrosilicon, and about 93–95% of solar polysilicon · polysilicon ran near $40/kg in 2022 and about $5–6.50/kg through 2026 · US silicon metal price averaged about 130 cents per pound in 2025, roughly 21% below 2024.

It splits in two, and that split is the most useful thing in this entry. Metallurgical silicon is about 87% Chinese, roughly 4.0 million of 4.6 million tonnes in 2025, and solar polysilicon is tighter still at roughly 93–95%, concentrated in Xinjiang, Inner Mongolia, and Sichuan where power is cheapest. Semiconductor-grade polysilicon is the exception, and one of the few high-purity steps on this sheet still held outside China: Wacker (Germany and Tennessee), Hemlock (Michigan), Tokuyama (Japan and Malaysia), and OCI (Korea and Malaysia) supply most of it. So a Western chipmaker and a Western module maker face completely different risks from the same element. For solar, the disruption already happened in a legal rather than a physical form — the Uyghur Forced Labor Prevention Act split the market into Xinjiang and non-Xinjiang polysilicon, with separate prices and separate supply chains. For semiconductors, the exposure is thinner than it looks: those plants buy metallurgical silicon that is largely Chinese, so the high-purity step is Western while the feedstock beneath it is not. Building non-Chinese polysilicon takes 2–4 years and, on current prices, loses money the whole time without a policy floor.

Wacker Chemie (Burghausen, Germany and Charleston, Tennessee), Hemlock Semiconductor (Michigan), Tokuyama (Japan and Malaysia), and OCI (Korea and Malaysia) supply most non-Chinese polysilicon and essentially all semiconductor grade. Tongwei, GCL Technology, Daqo New Energy, and Xinte Energy are the large Chinese solar polysilicon producers; GCL's fluidized-bed granular product is the main alternative to Siemens rods. Ferroglobe, Elkem, and Wacker are the reference Western silicon metal smelters, and Brazil, Norway, and Canada supply most US silicon metal imports. On the demand side, Shin-Etsu Handotai, SUMCO, and GlobalWafers pull the semiconductor ingots, and LONGi, JinkoSolar, and Trina consume the solar grade.

Polysilicon is the clearest illustration on this sheet that concentration and scarcity are different things. The price went from single digits per kilogram in 2020 to roughly $40/kg in 2022 as solar demand outran capacity, then collapsed as Chinese capacity expanded far faster than demand, and by August 2026 it was near $5/kg with much of the industry selling below cash cost. That is a capacity problem rather than a resource problem. Capital intensity is high and power is the dominant operating cost, so the cost curve sorts producers by electricity price more than by anything they do on site. Semiconductor-grade polysilicon is a much smaller, higher-margin business that runs on long-term contracts and qualification lock-in rather than spot pricing, which is why its producers survived a solar downturn that bankrupted others. If you are building a business here, the solar side is a commodity race that Chinese producers currently win on power and scale, and the semiconductor side is a specialty business where the moat is a customer's qualification file.

5 min“It’s Like Working In A Volcano”: How Silicon Is Made | Extreme JobsABC Science · 100k+ views

6 minTurning Sand into Sillicon MetalPrussian Blue · 50k+ views

5 minpoly silicon production process(多晶矽材生產過程)globalchampion · 100k+ views

Silicon, Mineral Commodity Summaries 2026 (USGS) · Special Report on Solar PV Global Supply Chains (IEA)

High-purity quartz is quartz sand refined to total impurities in the low parts per million, and its main job is to hold molten silicon without contaminating it. In the Czochralski process, polysilicon is melted at about 1,420 °C in a fused-quartz crucible and a seed crystal is slowly pulled and rotated out of it over roughly a week to grow a single-crystal ingot. The crucible is in contact with the melt for that entire pull, so any aluminum, sodium, titanium, or iron in the quartz has time to migrate into the silicon and ruin it. Ordinary quartz sand cannot do this at any price. The Spruce Pine pegmatites in North Carolina are geologically unusual: they crystallized slowly enough, and with few enough trace elements, that the quartz can be processed into crucible grade at commercial volumes. Two companies mine and refine it there, Sibelco under the IOTA brand and The Quartz Corp, a joint venture of Imerys and Norsk Mineral.

The material does something nothing else does economically: it survives a week at 1,420 °C against molten silicon while contributing almost no contamination, which is why the semiconductor and solar industries both standardized on it. Its weaknesses are economic and geographic rather than technical. Volumes are tiny relative to the value they enable, prices are set in bilateral contracts rather than on an exchange, and the industry is famously secretive, so published figures are estimates and should be treated that way. The failure mode is a single-location outage. Both mines sit in the same small area of the same county, exposed to the same weather and the same power grid, and Hurricane Helene proved in September 2024 that this is not a theoretical concern.

This is not a material anyone chooses; it is a dependency to map. If you pull silicon ingots, audit how many weeks of crucible inventory you and your crucible supplier hold, because the exposure sits at your supplier rather than on your own purchase orders. If you build synthetic quartz into a plan, verify which grade you actually mean: synthetic and cultured quartz already serve some semiconductor consumables, but for large Czochralski crucibles they cost roughly 5–10x natural high-purity quartz, which prices them out of solar entirely and makes them a partial answer even in chips. For everything else on your bill of materials, this entry is mainly useful as a calibration point. A chokepoint held by a friendly country with two competing operators is a different kind of risk from one held by a strategic rival, and it deserves a different response.

Spruce Pine supplies roughly 70–90% of world high-purity quartz, most estimates above 80% · crucibles hold molten silicon at about 1,420 °C for a Czochralski pull of roughly a week · crucibles have been estimated at about 30% of the manufacturing cost of a monocrystalline silicon ingot · synthetic quartz costs roughly 5–10x natural at crucible grade · production is on the order of tens of thousands of tonnes a year, with two operators · mines shut on 26 September 2024 after Hurricane Helene and restarted within about two weeks.

A near-monopoly held by the United States, which is the reason this entry is on the sheet. Estimates of the Spruce Pine share of world crucible-grade quartz run from about 70% to about 90%, and the honest answer is that nobody outside the two companies knows precisely, because neither publishes volumes. Hurricane Helene shut both operations on 26 September 2024, and for about two weeks it was a front-page semiconductor risk; Sibelco restarted production and shipments in early October and ramped back to full rate, and no chip fab actually stopped. That near-miss is the useful data point. A disruption here does not halt wafer production immediately, because crucible makers and ingot pullers hold inventory, but a multi-month outage would work its way through both the chip and solar supply chains with no available substitute. Be precise about what the dependence covers: it is the crucible, not the silicon. The silicon in a wafer comes from polysilicon made in Germany, the US, Japan, Korea, and China, and Spruce Pine quartz never enters the chip. Alternatives are under development in Norway (Drag), India, and elsewhere, plus expanded synthetic capacity, and Sibelco committed about $200 million in 2023 to roughly double its Spruce Pine capacity. None of them arrives quickly, because qualifying a new quartz source with a crucible maker and then with an ingot puller is a multi-year exercise in trace-element chemistry.

Sibelco's IOTA products and The Quartz Corp are the two Spruce Pine operators. Downstream, crucible makers such as Shin-Etsu Quartz, Momentive, and a growing set of Chinese producers turn the sand into fused-quartz crucibles, which then go to ingot pullers including Shin-Etsu Handotai, SUMCO, GlobalWafers, and the large Chinese wafer companies. Norway's Drag deposit, operated by The Quartz Corp, is the most-cited non-US source, and Australian and Indian projects have been announced repeatedly without displacing Spruce Pine. Hurricane Helene in September 2024 is the reference disruption, and China's steadily widening export-licensing list is the reason people started asking who else holds materials like this.

The market is small in revenue and enormous in leverage: tens of thousands of tonnes a year of a material that gates hundreds of billions of dollars of wafer and module production. There is no exchange price and no price reporting agency assessment worth much, because volumes move on multi-year bilateral contracts between a handful of counterparties, which is also why this material is invisible in commodity databases and in most critical-minerals policy documents. Prices are stable in a way that nothing else on this sheet is, and that stability is the point of the entry: the two operators have priced for long-term relationships rather than for extraction, and their customers have not felt the need to build alternatives. Compare that with a chokepoint of similar tightness held by a strategic rival, where buyers pay a two-tier price and carry inventory against a policy decision. Niobium is the other entry on this sheet that makes the same point, from Brazil rather than the United States, and with an even tighter single-company share. So the concentration number on its own does not tell you how exposed you are; you also have to ask who holds it and what they want from the relationship.

8 minWhy You Need Sand from This Town to Make a ComputerHalf as Interesting · 100k+ views

6 minSibelco Spruce Pine IOTA HPQSibelco Group · 5k+ views

8 minHow to make pure, synthetic quartzNikkei Asia · 100k+ views

Geology of the Spruce Pine District, Avery, Mitchell, and Yancey Counties, North Carolina (U.S. Geological Survey) · Quartz resources in Norway (Geological Survey of Norway)

Nobody mines gallium. Bauxite carries it at an average of about 50 parts per million, and during Bayer-process alumina refining it accumulates in the recirculating caustic liquor, where it can be stripped out with an ion-exchange resin or a solvent-extraction circuit bolted onto the refinery. A smaller amount comes from zinc-processing residues. The output is low-purity primary gallium, which is then zone-refined and crystallized to 6N or 7N before it is any use in a semiconductor. That means gallium supply is set by aluminum economics and by who chose to install a recovery circuit, not by gallium demand. World primary production was around 900 tonnes in 2025 against installed capacity near 1,700 tonnes, so the constraint is not geology at all. It is which refineries bother to run their gallium circuits.

Gallium's value comes from its compounds. Gallium nitride has a wide bandgap, high breakdown field, and high electron mobility, so a GaN power transistor switches faster and runs hotter than a silicon MOSFET, which is why fast chargers, data-center power supplies, and RF power amplifiers moved to it. Gallium arsenide has direct-bandgap optical properties and very high electron mobility, which makes it the substrate for laser diodes, high-efficiency multijunction solar cells, and the transmit-receive modules in AESA radar. The weakness is that the whole category is a byproduct: a price spike raises recovery rates at existing refineries but opens no new mines, and the response time is however long it takes to build and commission a recovery circuit. Purity is also a separate industry from production, so having gallium metal is not the same as having semiconductor-grade gallium.

Design with GaN when switching frequency, power density, or thermal headroom decides the product — chargers, onboard vehicle chargers, data-center power delivery, and RF above a few gigahertz. Use silicon instead where you can afford the efficiency and size penalty and want a supply chain with no policy overhang; silicon superjunction and IGBT parts still win on cost in most sub-kilowatt applications. Use silicon carbide instead when you need high voltage and high current in traction or grid inverters, where SiC has better thermal and breakdown behavior and a different supply chain. For radar, electronic warfare, and satellite solar there is nothing to switch to at high frequency and high power, so the correct response is inventory and qualified second sources rather than a materials study. If you are a Western buyer, price your GaN and GaAs content off the Rotterdam number, not the Chinese one.

World primary production about 900 t in 2025 against roughly 1,700 t of installed capacity · China roughly 99% of primary low-purity output, with Japan and Russia the only other producers · bauxite carries gallium at about 50 ppm · Rotterdam assessments ran roughly $2,100/kg in early 2026 and about $2,270/kg in August 2026, against Chinese domestic prices around $235–280/kg · US average import unit value about $580/kg in 2025, up roughly 30% on 2024 · integrated circuits took about 73% of US gallium consumption and optoelectronics about 26%.

Recovery from Bayer liquor, and it is as concentrated as anything on this sheet: China accounts for roughly 99% of world primary low-purity gallium, with about 1,600 tonnes of the world's 1,700 tonnes of capacity. Japan and Russia together make under 10 tonnes a year. Gallium was the first material China put under export licensing, in August 2023, followed by an outright ban on exports to the United States in December 2024 and a one-year suspension of that ban announced in November 2025 and running to late November 2026. Licensing did not stop material moving; it split the market. In 2026 gallium traded around $2,100–2,270/kg in Rotterdam against roughly $235–280/kg inside China, a gap of eight times or more for the same metal at the same purity, which is a policy price rather than a production cost. The Western alternative is unusually tractable on paper and unusually slow in practice. Most large alumina refineries could recover gallium, and several used to: Germany stopped in 2016, Hungary in 2015, Kazakhstan in 2013, all because it was uneconomic against Chinese supply. Projects have been announced in Australia, Canada, Greece, Kazakhstan, and Korea; the US Department of Energy put up as much as $6 million for research in September 2025 and the Department of War awarded $29.9 million in November 2025 for a Louisiana recovery demonstration plant. Call it 2–4 years to first metal at any single refinery, and longer to reach the purity semiconductor customers need.

Chinese alumina refiners operated by Chalco and Hongqiao supply most of the world's primary gallium. Downstream, Wolfspeed, Infineon, Navitas, and Power Integrations sell GaN power devices; Qorvo and Macom sell GaN RF; Raytheon and Northrop Grumman build the GaN AESA arrays in current radar programs. Azur Space and Rocket Lab's Solaero make GaAs multijunction space solar cells. In the US, the only current gallium refining is a single facility in New York that upgrades imported low-purity metal and new scrap; the Department of War's November 2025 award funds a demonstration plant in Louisiana recovering gallium and scandium from industrial waste.

This is a tiny market in dollars — world primary output at Rotterdam prices is on the order of a couple of billion dollars a year, and at Chinese domestic prices a small fraction of that. Nothing about the price is set by the cost of production, which is a marginal add-on at a refinery that is already running. The two-tier price is the economics: a Western device maker's cost is the Rotterdam number, set by export policy, while a Chinese competitor buys at the domestic number. That gap is a durable cost disadvantage for any device whose bill of materials is gallium-heavy, and it is the strongest argument for Western recovery capacity that exists. The catch is that recovery only pencils out while the gap persists, and the gap can be closed by a policy decision in Beijing faster than a plant can be built, which is exactly why the projects being funded now carry government money rather than private offtake. If you are underwriting one, the question is whether the offtake survives a scenario where Chinese export licensing is lifted entirely.

15 minChina's Gallium & Germanium Export ControlsAsianometry · 100k+ views

6 minIs gallium nitride the silicon of the future?Verge Science · 100k+ views

15 minThe Gallium Nitride RevolutionsAsianometry · 100k+ views

Gallium, Mineral Commodity Summaries 2026 (USGS) · Gallium, Professional Paper 1802-H (USGS)

Germanium is a byproduct twice over. It concentrates in certain sphalerite zinc ores, so it is recovered from the residues and fumes of zinc smelting, and in China a large share comes from the fly ash of lignite coal combustion. Either route produces a crude concentrate that is chlorinated to germanium tetrachloride, distilled, hydrolyzed to germanium dioxide, reduced to metal with hydrogen, and then zone-refined to the 6N and higher purity that optics and electronics need. World output is small, on the order of a couple of hundred tonnes a year, and the production data is genuinely poor: most producers do not report, and the USGS says outright that global estimates are difficult to verify. Uses split three ways. Fiber-optic cable uses germanium dioxide as the core dopant that raises the refractive index; infrared optics use germanium metal because it is transparent from about 2 to 14 microns and can be diamond-turned into lenses; and semiconductor applications use it in silicon-germanium RF circuits, germane gas, and the substrate of the multijunction solar cells on almost every satellite.

For thermal imaging, germanium is close to ideal: high refractive index means fewer elements in a lens, it transmits across both the mid-wave and long-wave infrared bands, and it machines well. For space solar, germanium substrates carry the highest-efficiency triple-junction cells flying. For fiber, a few percent germanium dioxide in the core is what makes single-mode fiber work over long distances. The weaknesses are supply-side. As a byproduct, output tracks zinc smelting and Chinese coal-ash processing rather than germanium demand. Germanium's refractive index also drifts with temperature, which forces athermalization in precision thermal optics. The failure mode for a Western buyer is a price step rather than an outright stoppage: China's export controls have roughly halved the volume reaching the West and put a durable premium on what does arrive.

If you are building a cooled or high-performance uncooled thermal imager, germanium is still the default lens material, and you should design in a chalcogenide alternative early rather than after your first supply scare. Chalcogenide glasses (the GASIR and IG families) and zinc selenide cover much of the long-wave band at lower cost and lower density, and several thermal-camera makers have already requalified around them; the trade is somewhat worse transmission and different dispersion, so it is a redesign rather than a drop-in. In datacom, silicon photonics displaces some germanium-doped fiber applications but not the fiber itself. In space solar there is no substitute: multijunction cells are grown on germanium wafers, and the alternative is a lower-efficiency cell and a bigger array. Match the effort to the exposure — a thermal-camera program can design germanium down, a satellite program cannot.

World output on the order of a couple of hundred tonnes a year, with China the leading producer, commonly put at roughly 60–80% · US average annual price about $4,100/kg for germanium metal in 2025 against $1,991/kg in 2024 and $1,187/kg in 2021 · European 99.999% metal rose from about $3,150/kg in January 2025 to about $5,380/kg in October 2025 · in August 2026 China domestic ran near $3,100/kg against about $6,350/kg from a US warehouse · China's germanium metal exports through September 2025 fell to about 7,520 kg from 36,656 kg over the same period in 2023 · transparent from roughly 2 to 14 microns in the infrared.

Refining, and the number is softer than most on this sheet because the data is poor. China is the leading producer and exporter of germanium metal, with estimates of its share usually falling in the 60–80% range, and the only other commercial refiners of consequence are in Belgium, Canada, Russia, and the United States. China put germanium under export licensing in August 2023, banned exports to the US in December 2024, and suspended that ban in November 2025 for a year while keeping the licensing requirement in place. The effect is visible in the trade data: Chinese germanium metal exports for the first nine months of 2025 were about 7,520 kg, against 18,787 kg in 2024 and 36,656 kg in 2023 over the same months. US imports of germanium metal fell by roughly two thirds in 2025. What that does to a Western buyer is a two-tier price — roughly $6,350/kg from a US warehouse in August 2026 against about $3,100/kg inside China — and long lead times on optical blanks. The Western alternative exists but is small. Zinc concentrates from Alaska's Red Dog mine already carry germanium and go to a Canadian refinery; the Clarksville, Tennessee zinc smelter that recovered germanium has been suspended since November 2023; and the Department of Energy announced funding in August 2025 covering germanium among other byproduct recovery projects. Recycling is real here, unusually: germanium is recovered from infrared optics machining swarf, decommissioned military optics, fiber manufacturing waste, and used space-solar substrates.

Umicore (Belgium) and Teck (Canada, from Red Dog concentrates) are the reference Western refiners. Yunnan Lincang Xinyuan Germanium and Yunnan Germanium are the largest Chinese producers, working coal-ash and zinc feeds. On the demand side, FLIR/Teledyne, Lynred, and Leonardo DRS build the thermal imagers that consume germanium optics, and the robot-sensors sheet's thermal-camera entry is the application view of this dependency. Andover and Vitron supply chalcogenide alternatives. II-VI/Coherent and Umicore make germanium substrates for space solar cells; Azur Space and Rocket Lab's Solaero build the cells on them. A germanium wafer plant in St. George, Utah and a germanium tetrachloride plant in Quapaw, Oklahoma are the two US downstream operations, both running on imported and recycled feed.

Germanium is assessed rather than exchange-traded, and the assessments are thin — a few hundred tonnes a year at a few thousand dollars a kilogram makes this well under a billion-dollar market, which is why no serious futures contract exists and no buyer can hedge. The cost of production is close to irrelevant to the price: recovery from zinc residues is a modest add-on cost at a smelter that is already running, so the market clears on availability and policy. That structure produces the pattern in the numbers, where the metal roughly tripled between 2021 and 2025 without any change in the cost of making it. For a Western optics company, germanium is a large enough share of the cost of a thermal lens assembly that the two-tier price shows up directly in product pricing. The businesses that do well here are the ones with a closed recycling loop, since reclaimed germanium from swarf and old optics is bought at a discount to the warehouse price and requalified in-house. If you are underwriting new Western primary recovery, check whether the host zinc smelter is economic on its own, because a germanium circuit cannot carry a smelter.

7 minGermanium - THE MOST FUTURISTIC ELEMENT!Thoisoi2 - Chemical Experiments! · 100k+ views

7 minPurifying Germanium - Early Semiconductor HistoryEdison Tech Center · 5k+ views

Germanium, Mineral Commodity Summaries 2026 (USGS) · Germanium and Indium, Professional Paper 1802-I (USGS)

Antimony is a brittle metalloid recovered mostly from stibnite, its sulfide ore, and increasingly as a byproduct of gold, lead, and silver mining. Concentrate is roasted or smelted to antimony trioxide, or reduced to metal in a blast or reverberatory furnace and then refined. World mine production was about 110,000 tonnes in 2025, which makes this one of the smaller markets on the sheet. The uses are unglamorous and stubborn. Roughly half of US consumption is antimony trioxide as a flame-retardant synergist, where it works with halogenated compounds in plastics, cables, textiles, and electronics housings; about 40% is antimonial lead in battery grids and ammunition primers, where a few percent antimony hardens soft lead; and the remainder goes into ceramics, glass, and rubber. The one that matters most for energy is solar glass, where antimony trioxide is the fining agent that clears bubbles and controls the redox state during melting.

Antimony is used because it does several unrelated jobs cheaply and well: it hardens lead, it makes halogenated flame retardants work at practical loadings, it fines glass, and it dopes silicon n-type. Nothing about it is exotic, which is why almost nobody thought about it until 2024. The weaknesses are the ones every small, opaque market has. Production tracks gold and lead mining as much as antimony demand, refining is far more concentrated than mining, and there is no exchange contract, so buyers discovered during the 2024–25 spike that they had no hedge and no visibility. The failure mode is a cost shock rather than a stoppage: material kept moving through export controls, but at four to five times the price, and flame retardant and glass formulators had to absorb it or reformulate.

If you are formulating flame retardants, antimony trioxide is the default synergist, and the alternatives are real but formulation-specific: hydrated aluminum oxide, magnesium hydroxide, and phosphorus-based systems can replace it at higher loadings, which usually costs you mechanical properties and sometimes a UL rating. Budget a year for requalification if the rating is safety-critical. If you make solar glass, treat antimony as a hard dependency, because the fining chemistry has few practical substitutes at production scale and the tonnage per gigawatt is small enough that you will pay whatever it costs. For lead-acid grids, calcium, tin, and selenium alloys already replace antimonial lead in many designs. For ammunition primers and armor-piercing cores, there is no substitute anyone is willing to qualify quickly. As a rule of thumb: if antimony is a few percent of your product's cost, ride out the price; if it is a synergist you can reformulate around, start now, because the controls have been suspended rather than removed.

World mine production about 110,000 t in 2025 · China about 40,000 t, Russia about 32,000 t, Tajikistan about 22,000 t · US price averaged $5.31/lb in 2021 and $5.49/lb in 2023, $10.24/lb in 2024, and about $25/lb in 2025 · peaked near $27.50/lb (about $60,000/t) in June 2025 and fell to about $20.30/lb by November 2025 · US uses split roughly 49% flame retardants, 40% antimonial lead and ammunition, 11% ceramics, glass, and rubber · recycling supplied about 12% of US apparent consumption, almost all of it antimonial lead from spent batteries.

Refining and oxide production, with mining a step behind. The mine numbers have moved: the USGS now puts China at about 40,000 tonnes of the world's 110,000 in 2025, roughly 36%, with Russia at about 32,000 and Tajikistan at about 22,000, which is a materially lower Chinese share than the "about half" figure commonly quoted from earlier data. Processing is a different story — China refines considerably more than it mines, importing concentrate to feed its smelters, and it supplied about 55% of total US metal and oxide imports over 2021–24. That processing position is what made the export controls bite. China restricted antimony exports in September 2024, banned exports to the United States in December 2024, and suspended that ban in November 2025 through late November 2026, keeping the licensing requirement. The price went from roughly $12,000/t before controls to a peak near $60,000/t in June 2025, then fell back through 2026 as substitution and new supply arrived, trading in a rough band of $32,000–52,000/t depending on which assessment you read, with one early-August 2026 quote near $51,800/t. The Western build is unusually fast for this sheet because the deposits exist and the processing is not exotic: mining started at the Stibnite Hill mine in Montana in November 2025, an Idaho project broke ground in October 2025 with an $80 million Department of War award, and US Antimony's Montana smelter is already processing imported feedstock. Call it 2–5 years to meaningful non-Chinese supply, which is quick compared with rare earths.

Hunan Gold and the Hsikwangshan mine at Lengshuijiang in Hunan are the reference Chinese operations; Tajikistan's Anzob complex and the antimony recovered at Russia's Olimpiada gold mine are the two largest non-Chinese sources. Perpetua Resources' Stibnite Gold Project in Idaho is the flagship US development, funded partly by the Department of War and pitched as both a gold mine and a domestic antimony source. US Antimony Corporation runs the Montana smelter on imported feedstock and started mining at Stibnite Hill in November 2025. On the demand side, the solar glass makers and the flame-retardant formulators (Israel Chemicals, LANXESS) are the buyers who felt the 2024–25 move most directly.

Antimony is a price-reporting-agency market with assessments in Rotterdam and China and no exchange contract, which is why the 2024–25 move was so violent: a small physical market with no hedging instrument and a policy shock produces exactly that. The whole world market is worth on the order of a few billion dollars a year at peak prices and considerably less at trough prices, so it has never attracted the capital that would build redundancy. The economics that decide new supply are unusual, because most non-Chinese antimony arrives as a credit on a gold mine's revenue rather than as a standalone product, which means the projects that get financed are gold projects with an antimony story attached. That is a strength — gold pays for the mine — and a weakness, since antimony output then follows the gold price. For a buyer, the practical lesson from 2024–26 is that a suspended export ban is not a resolved one; the licensing regime remains, the suspension has an expiry date, and the price has not returned to where it started.

8 minAntimony metal Extraction Element Series: 12Wheeler Scientific · 10k+ views

2 minAntimony: Essential Uses and Impact on Modern IndustriesStanford Advanced Materials · 10k+ views

Antimony, Mineral Commodity Summaries 2026 (USGS) · Antimony, Professional Paper 1802-C (USGS)

These three metals come out of the same rock in ratios the miner does not control, and that single fact explains most of their market behavior. The Bushveld Complex in South Africa holds them in thin reefs — the Merensky, UG2, and Platreef — at a few grams per tonne, with platinum, palladium, rhodium, ruthenium, iridium, gold, nickel, and copper all present together. Ore is milled and floated to a concentrate, smelted to a matte, converted, and then run through a base-metal removal circuit and a precious-metal refinery that separates the platinum-group elements one from another over weeks. Russia's Norilsk deposits are the other major source and carry a palladium-rich ratio, which is why Russia dominates palladium while South Africa dominates platinum and rhodium. Because the ratios are fixed by geology, you cannot make more rhodium without making more platinum, and nobody opens a mine for the metal that happens to be short.

As catalysts these metals are close to unmatched: they oxidize hydrocarbons and carbon monoxide and reduce nitrogen oxides at exhaust temperatures, survive thousands of hours in that environment, and do it at loadings of a few grams per vehicle. Platinum also has the corrosion resistance and high melting point that glass-fiber bushings, laboratory ware, and hydrogen fuel-cell and electrolyzer catalysts need. The weaknesses are structural. Supply is inelastic on a decade timescale, deep-level South African mining is expensive and exposed to power cuts and labor disputes, and output fell about 9% in 2025. Demand is splitting in a way that hurts: palladium is tied to the internal-combustion fleet and shrinks with it, while platinum picks up hydrogen and glass demand. The failure mode for a buyer is price rather than availability, and the price swings are the largest on this sheet.

For a gasoline three-way catalyst, the palladium-versus-platinum decision is a live economic optimization rather than a fixed recipe, and the industry has already run it in both directions. When palladium ran far above platinum, catalyst makers rethrifted back toward platinum, and USGS attributes part of platinum's 2025 price rise to exactly that substitution. If you are designing a new catalyst system, build the formulation so the ratio can be retuned, because the relative price will invert again. In diesel oxidation catalysts, about 25% of the palladium can routinely be swapped for platinum and up to about 50% in some applications. Rhodium is the one you cannot design around: nothing else reduces nitrogen oxides as well, loadings are already minimized, and at a few tenths of a gram per vehicle the cost is tolerable even at $10,000/oz. If your application is a proton-exchange-membrane fuel cell or electrolyzer, you are buying platinum for a growing market rather than a shrinking one, which changes the long-term supply picture in your favor.

World platinum mine production about 170 t in 2025 with South Africa about 120 t, roughly 71% · world palladium about 190 t with Russia about 84 t (roughly 44%) and South Africa about 70 t · annual averages in 2025 ran about $1,200/oz platinum, $1,100/oz palladium, and $5,800/oz rhodium · by August 2026 platinum was near $1,760/oz, palladium near $1,400/oz, and rhodium roughly $8,700–9,500/oz · rhodium averaged about $20,250/oz in 2021 and about $4,660/oz in 2024 · about 140 t of platinum and palladium came from scrap globally in 2025.

Mining, unusually for this sheet, and it is geological rather than industrial. South Africa produced roughly 120 of the world's 170 tonnes of platinum in 2025, about 71%, and the great majority of world rhodium; Russia produced about 84 of 190 tonnes of palladium, roughly 44%, almost all from Norilsk. Refining is more spread out than mining, with major precious-metal refineries in South Africa, the UK, Germany, Japan, and the US, so this is one of the few chains where owning the ore is the harder problem. A disruption is straightforward to picture because it keeps happening: South African load-shedding, a shaft incident, or a wage strike takes output offline for months, and there is no idle capacity anywhere to replace it. Russian palladium has never been formally sanctioned in bulk, but self-sanctioning and payment friction after 2022 pushed Western buyers toward South African and recycled metal. The Western alternative is thin. Stillwater in Montana is the only US primary producer and cut output sharply in 2025, and Canada's and Zimbabwe's mines are small by comparison. A new deep-level shaft is a 7–10 year project at multi-billion-dollar capital cost, so on any realistic timescale the answers are recycling and thrifting rather than new mines.

Anglo American Platinum (Valterra), Impala Platinum, Sibanye-Stillwater, and Northam operate the Bushveld mines; Norilsk Nickel is the Russian producer. Johnson Matthey, BASF, and Umicore make the autocatalysts and also run the refineries that recover the metal at end of life, which is why the same names appear on both sides of the market. Sibanye-Stillwater's Montana operations are the only meaningful US primary source. On the demand side, hydrogen electrolyzer and fuel-cell makers are the growth buyers of platinum, and the glass-fiber industry consumes platinum-rhodium bushings that are leased rather than sold outright, because the metal is worth more than the equipment.

These metals trade with real price discovery — platinum and palladium have futures contracts, rhodium does not — and they are the most volatile things on this sheet. Rhodium's history is the extreme case: a spike near $29,800/oz in March 2021 during the post-pandemic catalyst scramble, an annual average of about $20,250/oz that year, about $4,660/oz on average in 2024, and roughly $9,000/oz again in August 2026. Nothing about the cost of mining changed across that range. Mining economics are dominated by the basket price, the revenue per tonne from all the metals together, so a South African miner's viability depends on a ratio it cannot influence. Recycling is a genuinely large share of supply here, around 140 tonnes of platinum and palladium from scrap in 2025, and it behaves procyclically: converter collection rises with the metal price, which means recycling adds least when the market is tightest. If you are investing rather than buying, the structural trade is that palladium demand declines with the combustion fleet while platinum demand grows with hydrogen, and the two metals are largely produced by different countries, so that transition moves value from Russia toward South Africa.

18 minFrom Mine to Market: How Platinum is REALLY MinedMighty Builds · 100k+ views

3 minInteresting to Know. Mining and production of platinum metalsKrastsvetmet · 50k+ views

15 minPalladium: The King of CatalystsSprott · 5k+ views

Platinum-Group Metals, Mineral Commodity Summaries 2026 (USGS) · Platinum-Group Elements, Professional Paper 1802-N (USGS)

Iridium and ruthenium are the smallest products of the same Bushveld reefs that yield platinum, palladium, and rhodium. They are not mined for their own sake and cannot be: they arrive in fixed proportion with the platinum, are separated out at the end of the precious-metal refining sequence, and their annual supply is set by how much platinum the world happens to be producing. Iridium output is on the order of 7–8 tonnes a year, with South Africa supplying more than 80% of it, roughly 6–7 tonnes. Ruthenium is four to five times more available. Iridium's properties are extreme — the most corrosion-resistant metal known, melting at 2,466 °C — and in oxide form it is the only catalyst that survives the acidic, highly oxidizing conditions at the oxygen-evolution anode of a proton-exchange-membrane electrolyzer. Ruthenium goes into hard-disk magnetic layers, chip metallization at advanced nodes, thick-film resistors, and the dimensionally stable anodes in chlorine cells.

Iridium oxide does a job nothing else does at a useful lifetime: it catalyzes water splitting on the anode side of a PEM stack, which is the acidic half-cell where every cheaper candidate corrodes away. Ruthenium is unusually versatile for its price, and in the last few years it has become the metal semiconductor makers reach for as copper interconnect liners stop scaling. The weakness both share is that supply cannot respond to demand at any price. A tenfold rise in the iridium price produces very little extra iridium, because it would require a tenfold rise in platinum mining, and platinum demand does not justify that. The failure mode for anyone scaling PEM electrolysis is arithmetic rather than engineering: the metal simply does not exist in the quantities the hydrogen roadmaps assume.

Do the arithmetic before you commit a stack architecture. At legacy loadings of roughly 1–2 g of iridium per kilowatt, the entire world's annual iridium supply would build somewhere in the low single-digit gigawatts of PEM capacity a year, and that assumes every gram goes to electrolyzers instead of to spark plugs, crucibles, and electrodes. So if your plan calls for tens of gigawatts, you need one of three things: loadings cut to roughly 0.1–0.3 g/kW, which several catalyst suppliers have now demonstrated, an alkaline or anion-exchange-membrane stack that avoids iridium entirely, or a very large recycling loop. Pick PEM anyway when you need fast dynamic response to variable renewable power, high current density, and high-pressure output, since alkaline is slower and bulkier. Pick alkaline when you have steady power and want to avoid the platinum-group exposure altogether. For ruthenium, there is no design decision at the chip level; there is a procurement decision, and the right one is a long-term contract rather than spot buying.

World iridium supply on the order of 7–8 t/yr, with South Africa over 80% · ruthenium roughly four to five times more available · legacy PEM anode loadings roughly 1–2 g of iridium per kW, with demonstrated catalysts down to about 0.3 g/kW · iridium averaged about $5,160/oz in 2021, about $4,400/oz in 2025, and traded near $8,700/oz in April 2026 · ruthenium averaged about $451/oz in 2024 and about $690/oz in 2025, then reached roughly $1,750/oz by March 2026 · iridium melts at 2,466 °C.

South African mining, and it is tighter than the platinum figure because these are minor coproducts of a coproduct. Over 80% of iridium comes from South Africa, and more than 90% of ruthenium comes from South Africa and Russia combined. Refining is held by the same short list of precious-metal refiners that handle the rest of the platinum-group chain. What makes this chokepoint different from every other one on the sheet is that no policy decision created it and none can relieve it. There is no export license to lift and no second country with an idle deposit; the only way to make more iridium is to mine more platinum, and platinum's own demand decides that. The price history shows how the constraint transmits. Iridium's annual average actually fell about 9% in 2025, which the USGS attributes partly to increased production and partly to fading investor enthusiasm for hydrogen — a supply-constrained metal getting cheaper because the demand story slipped. Ruthenium went the other way, up about 53% in 2025 and roughly tripling in the year to March 2026 as AI data-center buildouts pulled hard-disk and chip-metallization demand, with Metals Focus forecasting a 203,000 oz deficit for 2026. The Western answer is thrifting and recycling, both of which work: catalyst loadings have already come down by most of an order of magnitude, and precious-metal refiners recover these metals efficiently when the scrap is collected.

Anglo American Platinum (Valterra), Impala, and Sibanye-Stillwater produce the iridium and ruthenium that reach the market; Heraeus, Johnson Matthey, and Umicore refine them and sell the catalysts. Heraeus has commercialized PEM anode catalysts at loadings near 0.3 g of iridium per kilowatt. Siemens Energy, Plug Power, Cummins, and ITM Power build PEM stacks and all publish iridium reduction roadmaps; thyssenkrupp nucera and John Cockerill sell alkaline stacks that avoid iridium. On the ruthenium side, Seagate and Western Digital consume it in hard-disk media, and advanced-node ruthenium interconnect and liner work at TSMC, Intel, and Samsung is what turned a quiet byproduct into a contested metal.

Both metals are assessed by refiners and price-reporting agencies rather than traded on an exchange, with wide bid-offer spreads and physical availability that can vanish while a quoted price sits still. The markets are small: single-digit tonnes a year for iridium and tens of tonnes for ruthenium, worth a few hundred million dollars each at typical prices. Supply is a byproduct credit on someone else's mine plan, so the cost of production tells you nothing about the price, and the price tells the producer nothing worth acting on. That combination produces the same pattern as the larger platinum-group metals, only sharper: violent moves, no supply response, and a market that clears through demand destruction and thrifting. For anyone building a hydrogen business, the practical consequence is that iridium intensity per kilowatt belongs in the model as a hard constraint alongside capital cost, and a stack design that cannot reach roughly 0.3 g/kW is not scalable to gigawatts regardless of what it costs. For anyone in semiconductors, the ruthenium move into 2026 is a reminder that a byproduct with a new structural demand source reprices very quickly.

4 minPEM electrolysis at Bosch | Scaling production of green hydrogenBosch Hydrogen Energy · 10k+ views

PGM market report, May 2026 (Johnson Matthey) · Is iridium demand a potential bottleneck in the realization of large-scale PEM water electrolysis? (International Journal of Hydrogen Energy)

Tungsten melts at 3,422 °C, higher than any other metal, and weighs about 19.3 g/cm³, roughly two and a half times as much as steel. Almost none of it is used as pure metal. Two ores matter, scheelite (calcium tungstate) and wolframite (an iron-manganese tungstate), which are gravity- and flotation-concentrated to roughly 65–75% WO3, digested in caustic soda, purified by solvent extraction or ion exchange, and crystallized as ammonium paratungstate. APT is the traded intermediate, quoted per metric tonne unit of WO3, and everything downstream is priced against it. From there the route is calcination to tungsten oxide, hydrogen reduction to metal powder, and carburization with carbon at roughly 1,400–1,600 °C to tungsten carbide powder, which is milled with 6–12% cobalt and liquid-phase sintered into cemented carbide. Cemented carbide takes something like 55–60% of world tungsten; the rest goes into mill products such as wire and welding electrodes, tungsten heavy alloy for mass balances and penetrators, steel and superalloy additions, and chemicals.

Cemented carbide keeps its hardness at the 800–1,000 °C a cutting edge reaches at production feed rates, which is why it displaced high-speed steel and still holds most of the metal-cutting market. Tungsten heavy alloy does the other job well: it puts a lot of mass in a small volume without being radioactive, which is what a balance weight, a radiation collimator, or a kinetic penetrator needs. The weaknesses are mechanical and commercial. Carbide is brittle and cannot be machined conventionally, so every feature has to be pressed in or ground and EDM'd afterwards, and the cobalt binder carries its own supply exposure plus a respirable-dust health problem in grinding shops. The failure mode that catches buyers is qualification rather than performance. A carbide grade is qualified around a specific powder from a specific supplier, so when that supplier's export license disappears you cannot simply buy equivalent powder somewhere else and keep shipping the same tool.

If you are cutting hardened steel, cast iron, or superalloys at production rates, carbide is the default and the question is only which grade and coating. Move to ceramics or cubic boron nitride when the workpiece is harder than about 45 HRC and the cut is continuous, and to polycrystalline diamond for aluminum, composites, and other non-ferrous work, accepting in both cases that the tool is more fragile and much less tolerant of interrupted cuts. If you need density in a fixed volume, tungsten heavy alloy is the practical answer, because depleted uranium is denser and self-sharpening but politically constrained almost everywhere outside a few defense programs. If your exposure is the concern rather than the machining, the highest-return move is a scrap loop: carbide swarf and used inserts are worth recovering at any recent price, and unlike rare earths the recycling routes are mature and already carry a large share of supply.

Melts at 3,422 °C, density about 19.3 g/cm³ · China roughly 80–83% of mine supply, about 67,000 t of a world total near 80,000 t in 2024 · cemented carbide 55–60% of demand, milled with 6–12% cobalt binder · recycled scrap on the order of 30–35% of world supply · APT cif Rotterdam went from roughly $340/mtu in early 2025 to about $3,075/mtu in early August 2026 · Chinese APT exports fell from 782 t in 2024 to 243 t in the first eleven months of 2025 · Sangdong Phase 1 commissioned March 2026 at roughly 2,300 t/yr of concentrate.

Mine supply and APT conversion, and the second is tighter than the first. China mines roughly 80–83% of world tungsten, about 67,000 t of a 80,000 t total in 2024, and converts a larger share again into APT, because most producing countries outside China ship concentrate rather than intermediate. China put tungsten under export licensing on 4 February 2025 and named a short list of about fifteen approved exporters in December 2025. The effect was immediate and measurable: APT exports fell from 782 t in 2024 to 243 t in the first eleven months of 2025, and the cif Rotterdam price went from roughly $340/mtu in early 2025 to $900–940/mtu in January 2026, $1,650–1,900/mtu by mid-February, and about $3,075/mtu in early August 2026. The Chinese domestic price has decoupled from the export price, which is the same two-tier structure gallium and germanium developed after their controls. The Western answer is real but small and slow. Almonty commissioned Phase 1 of the Sangdong mine in South Korea in March 2026, the first production there in over thirty years, at roughly 2,300 t/yr of concentrate with a Phase 2 doubling planned for 2027; Masan's Nui Phao in Vietnam remains the largest non-Chinese producer; Guardian Metal Resources is targeting late 2028 for Pilot Mountain in Nevada, in a country that has not mined tungsten since 2015. Converting concentrate to APT and carbide powder outside China is the harder half, and the existing capacity at Plansee's Global Tungsten & Powders in Towanda, Pennsylvania and Wolfram Bergbau in Austria is the base anyone building a non-Chinese chain has to expand from.

Sandvik Coromant, Kennametal, Ceratizit, Mitsubishi Materials, and Sumitomo Electric are the cutting-tool companies that consume most of the carbide. Upstream, Global Tungsten & Powders (Plansee) in Towanda, Pennsylvania and Wolfram Bergbau und Hütten in Austria are the reference non-Chinese APT and powder plants. On the mine side, Masan High-Tech Materials' Nui Phao in Vietnam, Almonty's Panasqueira in Portugal and Sangdong in South Korea, EQ Resources' Mt Carbine in Queensland, and Group 6 Metals' Dolphin mine on King Island are the operations Western buyers actually contract from. Guardian Metal Resources' Pilot Mountain project in Nevada has taken US Department of Defense funding and published a prefeasibility study in 2026. On the recycling side, Sandvik, Kennametal, and Buffalo Tungsten all run carbide reclaim, mostly by the zinc-reclaim route for clean single-grade scrap and chemical oxidation back to APT for mixed scrap.

Tungsten trades as APT, assessed cif Rotterdam and Baltimore by price reporting agencies rather than on an exchange, with a separate Chinese domestic quote that has diverged sharply since the export controls. World mine output of about 80,000 t of contained tungsten is a small market by tonnage, and at pre-2025 prices the APT layer was worth a few billion dollars a year, which is part of why nobody built redundancy into it. Mining economics are the reason supply responds slowly: run-of-mine grades are typically 0.1–0.5% WO3, the ore is abrasive, and a new hard-rock tungsten mine takes the better part of a decade from discovery to concentrate. Most of the value sits downstream anyway. A solid carbide end mill sells for one to two orders of magnitude more per kilogram than the powder inside it, so a five- or ten-fold move in APT raises a machine shop's tooling bill without stopping it, while squeezing powder producers who sell on annual contracts into a spot market that has repriced. Recycling is the one part of this chain that works better than the rare-earth equivalent, at roughly 30–35% of supply, and it works for a specific reason: carbide scrap is dense, valuable, generated in known places, and already flows back through the toolmakers who sold it. At $3,000/mtu that loop tightens further, and collection rates rise with the price.

17 minTungsten: From Ore to WireGM PULLMAN · 50k+ views

8 minHow Is Tungsten Made?History of Simple Things · 100k+ views

3 minHow to recycle solid carbide with Sandvik CoromantSandvik Coromant · 100k+ views

Tungsten, Mineral Commodity Summaries 2026 (USGS) · Tungsten mineral profile (British Geological Survey)

Niobium is a refractory metal used mainly as a steel additive, in quantities most people would call a rounding error. A few hundred grams per tonne, typically 0.02–0.06% by weight, forms fine niobium carbonitride particles that pin austenite grain boundaries during hot rolling and then precipitate on cooling, so the finished steel comes out stronger and tougher without extra carbon or alloy. That is the basis of high-strength low-alloy steel, which is what pipelines, car bodies, bridges, and heavy vehicle frames are made of. Almost all of it comes from pyrochlore in carbonatite deposits, above all the open pit at Araxá in Minas Gerais, where ore runs on the order of 2–3% Nb2O5 against parts per million in ordinary crust. Concentrate is floated and then reduced aluminothermically with iron oxide to ferroniobium at about 65% niobium, which is what a steelmaker actually buys and drops into the ladle. A much smaller high-purity stream goes to niobium oxide, electron-beam-melted metal, superalloys such as Inconel 718, and superconducting wire and cavities in the form of NbTi and Nb3Sn.

Niobium is the cheapest strength a steelmaker can buy. Adding 0.03% niobium can raise yield strength enough to take 10–15% of the weight out of a structure at almost no change in cost per tonne, which is why it spread through automotive and line-pipe steels without any policy pushing it. It also does the job at lower addition rates than vanadium or titanium, so less of it goes in per tonne of steel. The catch is that the mechanism depends on the mill, not just the recipe: niobium works through controlled rolling, so a plant that cannot hold finishing temperature and cooling rate gets a fraction of the benefit and may get worse toughness instead. The other weakness is one nobody prices. Supply is more concentrated than any Chinese chokepoint on this sheet, and because the price has been flat for fifteen years, very few buyers hold inventory, dual-source, or have a qualified substitute grade sitting on the shelf.

If you are designing structural, automotive, or line-pipe steel and your mill runs thermomechanical controlled processing, niobium microalloying is the default and the decision is only how much and at what finishing temperature. Use vanadium instead when the steel is thick-section, air-cooled, or reheated after rolling, since vanadium precipitates during cooling and does not need the same rolling discipline, and use titanium when you mainly want grain-size control in the weld heat-affected zone. Both cost more strength per unit added than niobium, so switching is a real option rather than a free one. For magnets, pick NbTi below roughly 9 T because the wire is ductile and cheap, and pay for Nb3Sn only when the field demands it, since Nb3Sn is brittle and has to be reacted after winding. And before you act on any concentration number on this sheet, ask who holds it and what they want from you: niobium is tighter than almost anything China controls, and it has never once been used against a buyer.

Brazil supplies roughly 90% of world niobium, with USGS-based tallies putting it as high as 93% · CBMM alone accounts for about 80% of world supply from the single deposit at Araxá · world mine output around 112,000 t of niobium content in 2025 · US average ferroniobium import unit value was $21/kg in 2021 and about $26/kg in 2025, a range of roughly 1.2x over five years · typical HSLA addition 0.02–0.06% niobium, a few hundred grams per tonne of steel · US consumption splits about 77% steels and 23% superalloys, with 100% import reliance, 67% from Brazil and 28% from Canada.

One orebody, one company, one country, and almost nobody worried about it. Brazil supplies close to 90% of world niobium and CBMM alone is roughly 80% of world supply, all of it from Araxá, which is a tighter single-company share than China holds in graphite anode material, magnets, or gallium. A serious outage at Araxá would leave the world's steelmakers short of the additive in most modern structural steel within a quarter, with Canada's Niobec mine and CMOC's Boa Vista operation in Goiás unable to cover more than a slice of it. Yet there is no export license regime, no two-tier price, and no stockpiling program worth the name outside a small US Defense Logistics Agency ferroniobium line. The reason is the supplier's strategy. CBMM has priced for volume growth rather than for extraction, spends heavily on steel-design support so that customers use more niobium rather than less, and faces a real substitution threat from vanadium and titanium if it ever pushes price. Ownership reinforces the point in both directions: 70% sits with Brazil's Moreira Salles family, and the other 30% was sold in 2011 in two 15% blocks, one to a Japanese-Korean consortium including Nippon Steel and POSCO and one to a group of Chinese companies, so the largest customers are also shareholders. The Western alternative is thin. Niobec in Quebec, now owned by Magris Resources, is the only significant non-Brazilian mine, and NioCorp's Elk Creek project in Nebraska has permits and US government interest but has not closed financing, which puts first production toward the end of the decade at the earliest. High-purity quartz makes this same argument from the American side, and the two entries are the reason the sheet separates concentration from exposure: a 90% share held by a friendly commercial supplier with an interest in your growth is not the same risk as a 90% share held by a government that has already used export licensing as leverage.

CBMM at Araxá is the producer that matters, selling standard ferroniobium plus high-purity oxide, metal, and specialty alloys. CMOC runs the Boa Vista niobium operation at Catalão in Goiás, bought from Anglo American in 2016, and Magris Resources runs Niobec at Saint-Honoré in Quebec. NioCorp's Elk Creek project in Nebraska is the most advanced US proposal. On the demand side, Nippon Steel, POSCO, ArcelorMittal, and the API X70 and X80 line-pipe grades are the volume users; Inconel 718, at around 5% niobium, is the superalloy that put niobium into every large jet engine and gas turbine; NbTi wire is in essentially every MRI magnet and the LHC dipoles; Nb3Sn is in ITER's central solenoid and in high-field research magnets; and bulk high-purity niobium is what superconducting radiofrequency cavities such as those in LCLS-II are made from. CBMM and Boston Metal have also run trials on molten-oxide electrolysis for niobium production, which is worth watching but is not supply today.

There is no exchange price for niobium. Ferroniobium moves on contracts against a price the dominant producer effectively posts, and the best public series is the USGS US import unit value, which ran $21/kg in 2021 and 2022, $25/kg in 2023, and about $26/kg in 2024 and 2025. A range of 1.2x over five years is the flattest price on this sheet by a wide margin, in a market where the seller could plainly charge more. The market is also small in revenue: US niobium imports were valued at about $525 million in 2025, and the world market at the ferroniobium level is a few billion dollars a year, which is trivial next to the value of the steel it goes into. That asymmetry is the whole business model. Niobium raises the value of a tonne of steel by far more than it costs, so as long as the price stays low the steelmaker keeps specifying it, and CBMM sells growing volume from a deposit with reserves measured in decades at very high margin. Push the price and the calculation changes, because vanadium and titanium microalloying work, just less efficiently, and steel buyers are extremely price-sensitive. For anyone building on this, the practical conclusion is that the risk here is not commercial but physical and political: the price will probably stay stable, and the thing to plan against is an accident, a tailings failure, or a change in Brazilian export policy rather than a squeeze.

3 minNiobium | What role does Niobium play in steel?Niobium Tech · 10k+ views

7 minWhat is Niobium? (Cradle Resources)The Sophisticated Investor TV · 10k+ views

7 minNiobium | The Emerging Role of Niobium in Advanced Technologies (2017)CBMM · 5k+ views

Niobium (Columbium), Mineral Commodity Summaries 2026 (USGS) · Niobium and Tantalum, Professional Paper 1802-M (USGS)

Uranium concentrate, U3O8, is the first tradeable product of the nuclear fuel cycle and the form utilities actually contract for. Two routes produce it. Conventional mining crushes and grinds ore, leaches it with sulfuric acid or an alkaline carbonate solution, recovers the uranium by solvent extraction or ion exchange, precipitates it, and dries or calcines it to a yellow-brown powder at roughly 75–85% uranium by weight. In-situ recovery skips the mine: oxygenated, acidified or carbonated groundwater is pumped down injection wells into a permeable sandstone-hosted deposit, dissolves the uranium in place, and is pumped back up through production wells to an ion-exchange plant at surface. Almost all Kazakh and all US output uses in-situ recovery, which is why Kazakhstan can produce at very low cash cost from ore grades that would never support a conventional mine. Grades vary by orders of magnitude: the Athabasca Basin deposits in Saskatchewan run in the range of 10–20% uranium, while typical in-situ orebodies run a few hundredths of a percent. From here the concentrate goes to conversion, which is the next entry and a much narrower step.

Uranium is not scarce and its supply is the least concentrated part of the nuclear fuel chain, with meaningful production in Kazakhstan, Canada, Namibia, Australia, Uzbekistan, Niger, and Russia and identified resources measured in decades of demand. Energy density makes the logistics trivial: a 1 GW reactor consumes on the order of 200 t of natural uranium a year, so stockpiling several years of supply is cheap and utilities routinely do it. The weakness is response time. A greenfield uranium mine takes something like 10–15 years from discovery through permitting to first production, so the price signal that justifies it arrives years before the metal does, and the 2007 and 2024 price spikes both ended before much new supply appeared. In-situ recovery has a narrower failure mode: it depends on hydrogeology that only works in specific formations, and on a steady supply of reagents. Kazakh output was capped through 2024 and 2025 partly by a sulfuric acid shortage, which is an ordinary industrial-chemicals problem constraining 40% of world uranium.

There is no alternative fuel for a light-water or heavy-water reactor, so this is a procurement problem rather than a design choice. If you are buying, buy on term contracts and buy early: the spot market carries a small share of world volume and is easily moved by financial buyers, while the term price is what actually clears fuel, and in June 2026 the term price reached $94/lb against a spot price near $86/lb. That inversion is the market telling you it expects tightness. If you are financing a mine, model the incentive price rather than the spot price, because most undeveloped projects need something in the $90–100/lb range to justify construction and the current spot price is below that. If you are designing a reactor and want to know how enrichment level and fuel form change your exposure, that decision belongs to the nuclear-reactors sheet; here the point is that uranium itself is the part of the chain you are least likely to be short of.

World mine production 60,213 tU in 2024 · Kazakhstan 23,270 tU, about 39%, followed by Canada at 14,309 tU and Namibia at 7,333 tU · reactor requirements about 68,900 tU in 2025, with mine supply covering roughly 90% · spot U3O8 about $28/lb in 2020, a peak near $107/lb in early 2024, and $86.48/lb on 8 August 2026 · long-term contract price $94/lb in June 2026, an 18-year high · Kazatomprom 2026 guidance 27,500–29,000 tU on a 100% basis, roughly 10% below 2025 · a 1 GW reactor consumes on the order of 200 tU a year.

Kazakhstan, at about 39% of world mine supply in 2024, which is the loosest chokepoint on this sheet and still the one most worth watching. The concentration that matters is not geological. Kazatomprom is roughly three-quarters owned by the Kazakh state fund Samruk-Kazyna, and its production sits in joint ventures with Rosatom's Uranium One, China's CNNC and CGN, Cameco, and Orano, so a large share of Kazakh output is already committed to Russian and Chinese offtake before it reaches the market. Logistics are the second dependency: Kazakh uranium historically shipped west through St Petersburg, and since 2022 Western buyers have pushed volumes onto the Trans-Caspian route through Azerbaijan and Georgia, which works but is slower, more expensive, and lower capacity. A disruption would not stop reactors quickly, because utilities hold one to three years of inventory and the fuel takes 12–18 months to move through conversion, enrichment, and fabrication anyway. It would show up as a price spike and a scramble for term contracts. The Western alternative is more credible here than anywhere else on the sheet, and it is still slow. Cameco has ramped McArthur River and Cigar Lake back toward licensed capacity, NexGen's Rook I project in Saskatchewan is a large new Western mine working through permitting toward the end of the decade, and several small US in-situ operations have restarted at Christensen Ranch, Rosita, and Lost Creek. Restarting an idled in-situ wellfield takes roughly 12–24 months but adds a few hundred tonnes; building a new conventional mine takes ten years or more.

Kazatomprom is the world's largest producer, operating through joint ventures including Inkai with Cameco, Katco with Orano, and Budenovskoye with Uranium One. Cameco runs McArthur River and Cigar Lake with the Key Lake and Blind River facilities behind them. In Namibia, CGN's Husab and the Rössing mine, now Chinese-owned, make that country the third-largest producer. BHP's Olympic Dam in South Australia produces uranium as a copper byproduct. Orano's position in Niger shows how fast this can change: the company lost operational control of the Somair mine in December 2024 and Niger moved to nationalize it in 2025. On the US side, Energy Fuels' White Mesa mill is the only operating conventional uranium mill in the country, and enCore Energy, Ur-Energy, and Peninsula Energy have restarted in-situ production in Texas and Wyoming.

The uranium spot price is widely quoted and poorly representative. Most volume moves under multi-year term contracts, priced either as a base price escalated to delivery or as a market-related price with a floor and a ceiling, and utilities typically contract two to five years ahead of burn. That structure means the spot price can double without a utility's fuel cost changing much, and it also means a utility that let its contract book run down has to buy into whatever market exists. Costs split sharply by route. In-situ recovery in Kazakhstan and the US has all-in sustaining costs in the region of $20–40/lb, high-grade Athabasca mining is competitive with that despite far higher capital cost, and conventional open-pit operations in Africa and Australia sit higher. Greenfield incentive prices are generally quoted around $90–100/lb or above, which is why a spot price in the $80s has not triggered a construction wave. The other structural change is the disappearance of secondary supply. For two decades the gap between mine output and reactor requirements was filled by downblended Russian weapons material under the Megatons to Megawatts agreement, government inventories, and enrichment underfeeding. That agreement ended in 2013, and when separative work is scarce and expensive, enrichers overfeed instead, which consumes more natural uranium rather than releasing it. Mine supply now covers about 90% of requirements and has to cover more.

10 minUranium from the GroundIllinois EnergyProf · 50k+ views

2 minIn Situ Mining ProcessCamecoCorporation · 10k+ views

Uranium Mining Overview (World Nuclear Association) · Uranium 2024: Resources, Production and Demand (OECD-NEA and IAEA)

Conversion turns U3O8 concentrate into uranium hexafluoride, the only uranium compound that becomes a gas at a convenient temperature and therefore the only feed a centrifuge cascade can enrich. Both commercial routes end the same way and differ in where the uranium gets purified. In the dry route, yellowcake is ground and reduced with hydrogen in a fluidized bed at roughly 540–650 °C to UO2, hydrofluorinated with anhydrous HF at about 480–540 °C to UF4, the non-volatile green salt, and then fluorinated with elemental fluorine to UF6, with volatile impurities stripped out at several points and by distilling the UF6 at the end. The wet route dissolves the yellowcake in nitric acid first and runs it through solvent extraction to remove impurities, then does the same reduction, hydrofluorination, and fluorination steps. Wet or dry describes the purification, not the fluorination chemistry, and the industry's common shorthand gets this backwards: Cameco's Port Hope plant runs the wet route, Orano runs wet, and ConverDyn's Metropolis Works in Illinois is the dry plant. Product UF6 is condensed as a liquid into 14-tonne mild steel cylinders where it solidifies over about five days, and it ships that way to an enricher.

The chemistry is old, well understood, and not capital-intensive by nuclear standards, which is exactly why nobody built spare capacity. There is no technical barrier to adding conversion; the barrier is that a plant only makes money when the conversion price is high, and for most of the 2010s it was not. Handling is the operational weakness. Anhydrous HF and elemental fluorine are among the more unpleasant industrial reagents, UF6 hydrolyzes on contact with moisture to HF and uranyl fluoride, and a conversion plant is therefore a chemical-safety facility with a nuclear license on top. The failure mode this industry actually experienced is commercial rather than chemical: Honeywell idled Metropolis in 2017 because the price did not cover operating cost, Cameco cut Port Hope output, and the UK's Springfields UF6 line closed in 2014. When demand recovered after 2022, restarting those plants took years, not months, and the price went up by roughly an order of magnitude in the meantime.

There is no choice to make and no way around the step, so treat it as inventory management. If you are a utility or a fuel buyer, contract conversion separately and well ahead, because conversion and enrichment are bought as distinct services and a buyer who holds uranium but no conversion slot holds a powder rather than a fuel. If you are evaluating a fuel-cycle investment, note that the long-term conversion price was still climbing in mid-2026 at $55.50/kgU while the spot price had fallen back to $64.50/kgU from a peak of $97/kgU at the end of 2024, which is the market saying it expects the tightness to persist beyond the current spot balance. If you are looking for the step in the nuclear chain most exposed to a single outage, this is a better candidate than uranium mining, because there are only a handful of plants and world output has been running well below licensed capacity for reasons that are operational rather than deliberate.

World licensed conversion capacity roughly 62,000 tU/yr against about 42,000 tU of actual output in 2022 · Russia holds about 20% of licensed capacity and produced about 29% of 2022 output · conversion spot price about $6/kgU at the end of August 2016, peaking near $97/kgU at the end of 2024, and $64.50/kgU on 30 June 2026 · long-term conversion price $55.50/kgU in mid-2026 and still rising · UF6 ships in 14 t mild steel cylinders that take about five days to solidify · Metropolis Works was idled in 2017 and restarted in 2023.

Four commercial operators outside China, and none of them can absorb the loss of another. Cameco's Port Hope in Ontario, backed by the Blind River refinery that makes the UO3 feed, ConverDyn's Metropolis Works in Illinois, Orano's Philippe Coste plant at Tricastin with its Malvési front end, and Rosatom's Russian plants are the whole Western-accessible list, with CNNC's Chinese capacity largely serving domestic reactors. A common claim is that Russia holds around 40% of this step. It does not; that figure belongs to enrichment. Russia's share of conversion is about 20% of licensed capacity and was roughly 29% of 2022 output, which is significant but not dominant. The real problem is utilization. World licensed capacity is on the order of 62,000 tU/yr, and 2022 output was about 42,000 tU, so roughly a third of nameplate capacity was not producing. There is no authoritative public figure for world conversion demand, so resist the temptation to compute a precise gap from those two numbers; what the price did is the better evidence, and conversion went from single-digit dollars per kgU in the late 2010s to a peak near $97/kgU at the end of 2024. The Western alternative is unusual on this sheet in that it is mostly already built. Metropolis restarted in 2023 after a six-year idle, Orano is expanding Philippe Coste toward its higher design rate, and Cameco has been running Port Hope harder. Bringing an idled line back has taken three to six years in practice, and building a new plant would take longer, so the correct expectation is that this step stays tight through the late 2020s rather than being solved by an announcement.

Cameco Port Hope is the reference Western plant, taking UO3 from Blind River and also producing UO2 powder for CANDU fuel. ConverDyn markets the output of Honeywell's Metropolis Works, the only US conversion plant and the only commercial dry-route plant. Orano's Philippe Coste facility at Tricastin, the completion of the long-delayed Comurhex II program, replaced older French capacity and is the European supply base. Rosatom converts at its Siberian sites and sells UF6 into markets that will still buy it. Westinghouse's Springfields site in Lancashire is the cautionary example: it produced UF6 until 2014 and then stopped, and the UK now has no commercial conversion. Utilities such as Électricité de France, Constellation, and the Korean and Japanese fleets buy conversion as a separate line item from uranium and enrichment, usually through the same term contracts that cover the rest of the fuel cycle.

Conversion is quoted in dollars per kilogram of uranium as UF6, both as a spot price and as a long-term price, and both are assessed by price reporting agencies rather than exchange-traded. The market is small in absolute terms, a few billion dollars a year at recent prices and a fraction of that before 2022, and that smallness is why the step went unhedged for so long. Cost structure is dominated by fixed costs: a conversion plant is a chemical plant designed to run continuously, so operating one at half rate is close to as expensive as running it full, and the marginal cost of the last tonne is low while the average cost is high. That produces exactly the behavior the industry showed, with plants idled entirely rather than throttled when prices fell below cash cost, and it also explains the size of the rebound, since restarting is expensive and slow enough that operators want a price that pays for the whole cycle. Energy is a real but secondary input. The dry route at Metropolis uses more electricity and less steam than the wet plants, and older cost analysis put the energy component at roughly $1/kgU, which was material when the price was $6/kgU and is a rounding error at $65/kgU. The producers make good money at current prices, and the buyers who signed long-term conversion contracts before 2021 are the ones who came out ahead.

4 minCameco Fuel Cycle - ConversionCamecoCorporation · 5k+ views

2 minCameco Fuel Cycle - RefiningCamecoCorporation · 1k+ views

5 minHow It's Made - Uranium Part 2CamecoCorporation · 1m+ views

Conversion and Deconversion (World Nuclear Association) · Advanced Fuel Cycle Cost Basis – 2017 Edition (Idaho National Laboratory)

Enrichment raises the fraction of uranium-235 in uranium hexafluoride from the 0.71% found in nature to the 3–5% a light-water reactor needs, or to the 5–20% band called high-assay low-enriched uranium. Every commercial plant does it with gas centrifuges: UF6 is fed into rotors spinning at high peripheral speed, the heavier U-238 concentrates near the wall, a countercurrent flow amplifies the tiny separation along the rotor's length, and thousands of machines are plumbed into cascades so the enriched stream moves up and the depleted tails move down. The unit of output is the separative work unit, which measures separation effort rather than mass, so a plant's capacity is quoted in SWU rather than tonnes. The tails assay is a commercial lever as much as a technical one: enriching to a lower tails assay uses more SWU and less natural uranium, and enrichers move it depending on which is scarcer. HALEU is the same process run further up the cascade, and the reason it is a separate problem is that criticality safety, licensing, transport packaging, and cascade design all change above 10% enrichment. Which reactor designs need HALEU and why is covered in the nuclear-reactors sheet.

Centrifuges are efficient and modular. A modern plant uses roughly 50 kWh per SWU against about 2,400 kWh per SWU for the gaseous diffusion plants they replaced, and capacity is added by installing more machines in an existing hall rather than by building a new process, so expansions can be incremental. The weakness is that the technology is proliferation-sensitive, so the machine designs are export-controlled state secrets held by a handful of organizations, and a new entrant cannot buy centrifuges the way it would buy pumps. That is why capacity concentrates. The failure mode showing up now is timing: a centrifuge plant's lead time is dominated by manufacturing the machines and by licensing rather than by civil works, so even a well-funded expansion adds capacity over three to five years, and the Western fleet spent the 2010s in oversupply and shut capacity instead of building it.

Nothing substitutes for enrichment in a light-water reactor, so the decisions are contractual and design-level. If you are buying, split your exposure across suppliers and contract long, because the gap between what a spot buyer pays and what a term buyer pays is unusually wide here: US utilities paid an average of $108.70/SWU across their 2025 deliveries, up about 11% on 2024, while spot SWU was assessed near $188/SWU in August 2025. If you are designing an advanced reactor, treat HALEU availability as a schedule risk equal to your licensing risk and get a DOE allocation or a commercial contract before you finalize a core design, since first cores have slipped for exactly this reason. If you want to reduce exposure without changing supplier, ask your enricher about tails assay: at a high SWU price, buying more natural uranium and running higher tails is the cheaper package, which is also why enrichment scarcity raises uranium demand rather than lowering it.

Natural uranium is 0.71% U-235; light-water fuel is 3–5% and HALEU is 5–20% · Rosatom holds roughly 40–44% of world enrichment capacity, with Urenco and Orano most of the rest · Russia still supplied about 26% of US utilities' SWU purchases in 2025 despite the import ban · US utilities paid an average of $108.70/SWU in 2025, about 11% above 2024, against spot lows near $50/SWU in 2021 and spot near $188/SWU in August 2025 · US domestic enrichment capacity was about 4.3 million SWU/yr before 2025 against roughly 15.6 million SWU/yr of US reactor demand · Centrus produced at a 900 kg/yr HALEU rate at Piketon and is targeting 12 t/yr by 2029 · centrifuges use roughly 50 kWh per SWU against about 2,400 kWh for gaseous diffusion.

Russia, at roughly 40–44% of world enrichment capacity, and it is the single largest concentration in the nuclear fuel cycle. Urenco, with plants in the UK, Netherlands, Germany, and Eunice, New Mexico, and Orano's Georges Besse II in France hold most of the balance, with China's CNNC capacity growing and largely captive. The United States banned Russian enriched uranium imports under legislation signed in May 2024, effective from August 2024, with waivers available through 2027 and a full prohibition from 1 January 2028; even so, Russia supplied about 26% of US utilities' SWU purchases in 2025 under those waivers. A hard cutoff would not be absorbed by the existing Western fleet, which is why the waivers exist. The build is real and dated. Urenco announced a further expansion of its New Mexico plant in June 2026, adding around 2.1 million SWU; Orano is developing a roughly $5 billion enrichment plant at Oak Ridge, Tennessee with US government support; Centrus is expanding Piketon under a $900 million DOE award. All of that arrives from the late 2020s. HALEU is the sharper version of the same story. Russia's TENEX was the only commercial supplier, US production is measured in hundreds of kilograms a year against advanced-reactor demand projected in tens of tonnes a year around 2030, and the interim supply is DOE downblending its own stocks of highly enriched uranium. Centrus met its 900 kg annual target at Piketon and is contracted to reach 12 t/yr by 2029, which is a real answer that arrives after several announced reactors say they need fuel.

Rosatom's TENEX markets Russian enrichment from plants at Seversk, Zelenogorsk, Angarsk, and Novouralsk. Urenco operates Capenhurst in the UK, Almelo in the Netherlands, Gronau in Germany, and Eunice in New Mexico, which is the only operating commercial enrichment plant in the United States. Orano runs Georges Besse II at Tricastin and is developing an Oak Ridge site. Centrus Energy operates the American Centrifuge Plant at Piketon, Ohio, the only US HALEU production, and Global Laser Enrichment is pursuing the SILEX laser process at Paducah, Kentucky as the one genuinely different technology in development. On the demand side, TerraPower, X-energy, Kairos Power, Oklo, and Radiant have all received DOE HALEU allocations, and TerraPower's Natrium schedule in particular has been shaped by when that fuel arrives.

Enrichment is sold in SWU on multi-year contracts, and as with conversion the term price and the spot price tell different stories. The EIA's weighted average of what US utilities actually paid was $108.70/SWU in 2025 while spot ran roughly 70% higher, which means the reported cost of the existing fleet's fuel understates what a new buyer faces. The world market is on the order of $15 billion a year, small enough that the whole business is a rounding error next to the electricity it enables and large enough that a 40% supplier has leverage. Cost structure is capital and electricity: a centrifuge plant is essentially a very large fixed investment in machines with a long life and low variable cost, so the economics reward running flat out and punish idle capacity, and the industry's decade of oversupply drove prices below the level that justified new machines. That is why nothing was built until the political shock made it necessary. For anyone building on this, the practical point is that the enrichment price is not the binding constraint on nuclear economics, since fuel is a small share of a reactor's cost, but enrichment availability can be a binding constraint on schedule. HALEU is where that shows: the metal exists at a price, but not at a delivery date, and no amount of money moves a cascade license faster.

7 minUrenco Virtual Tour - How do we enrich uranium?Urenco · 10k+ views

13 minGetting to the Good Stuff (Uranium Enrichment)Illinois EnergyProf · 100k+ views

2 minFrom LEU to HALEU - Nuclear Enrichment Explained With Maple Syrup 🍁World Nuclear Association · 100k+ views

Uranium Enrichment (World Nuclear Association) · What is High-Assay Low-Enriched Uranium (HALEU)? (US Department of Energy, Office of Nuclear Energy)

Zirconium reaches the market as zircon sand, ZrSiO4, recovered as a coproduct when heavy-mineral sands are mined for the titanium minerals ilmenite and rutile. World output was about 1.2 million tonnes gross weight in 2025, led by Australia at 400,000 t and South Africa at 270,000 t, and most of it goes into ceramic tile opacifier, foundry sand, refractories, and zirconium chemicals. Only a small slice becomes metal, and a smaller slice again becomes nuclear grade. That route runs zircon through carbochlorination to zirconium tetrachloride, then a hafnium separation step, then Kroll reduction with molten magnesium to sponge, vacuum arc remelting into Zircaloy-2 and Zircaloy-4 or the zirconium-niobium alloys M5 and ZIRLO, and finally extrusion and pilger rolling into cladding tube. The hafnium separation is what makes the nuclear chain a different business from the industrial one. Hafnium sits directly below zirconium in the periodic table and has almost the same ionic radius because of the lanthanide contraction, so the two are unusually hard to separate, yet hafnium absorbs thermal neutrons about 600 times as strongly. Reactor-grade zirconium therefore has to be stripped to under 100 ppm hafnium, by solvent extraction of the aqueous chemistry or by extractive distillation of the chlorides. What comes off that step is entry 029.

Zirconium is close to unique in combining very low neutron absorption with good corrosion resistance in 300 °C water and adequate creep strength, which is why every commercial water-cooled reactor built since the 1950s has used it for cladding. The weakness is a chemistry problem rather than a supply problem. Above roughly 1,200 °C, zirconium reacts with steam exothermically to make zirconium dioxide and hydrogen, so a loss-of-coolant accident generates both extra heat and the hydrogen that blew up the Fukushima Daiichi reactor buildings. Long-burnup fuel also picks up hydrogen and forms brittle hydrides in the cladding wall. On the supply side, the failure mode is qualification. Nuclear-grade sponge and finished tube are approved plant by plant against a fuel vendor's specification, so a customer who loses a supplier cannot simply buy from another one.

For a light-water reactor being fueled today, a zirconium alloy is the only licensed option, and the current decision is whether to take a chromium-coated variant. If your plant is chasing accident margin and higher burnup, chromium-coated Zircaloy is the near-term answer and is already flying in lead test assemblies. If oxidation resistance at 1,200 °C matters more to you than neutron economy, FeCrAl cladding is the credible alternative, but budget for the parasitic neutron absorption, which costs roughly a few tenths of a percent of extra enrichment. If you are designing for coolant temperatures well above light-water conditions, silicon-carbide composite cladding is the long-term answer and is not licensed for commercial use, with hermetic sealing and end-cap joining still unsolved at production scale. The `nuclear-reactors` sheet covers which reactor designs those claddings suit; this entry is about whether you can buy the metal.

World zircon mine production about 1.2 Mt gross weight in 2025, down 12% on 2024, with Australia at 400,000 t and South Africa at 270,000 t · only a low single-digit percentage of zircon becomes metal, and less again becomes nuclear grade · reactor-grade zirconium must hold hafnium below 100 ppm · zirconium's thermal neutron absorption cross-section is about 0.18 barns, against 104 for hafnium · Chinese zirconium sponge ran $22–30/kg across 2021–2025 and premium zircon $1,530–2,300/t, both moving less than 2x over five years · four countries hold essentially all nuclear-grade capacity.

Nuclear-grade sponge and tube, held by four countries and a handful of plants. Framatome's Cezus operations in France, ATI in Albany, Oregon, Westinghouse's Western Zirconium in Ogden, Utah, Rosatom's Chepetsky Mechanical Plant at Glazov, and a growing Chinese group around CNNC and Baoji are the whole world list, and USGS records exactly one US producer of zirconium and hafnium metal in Oregon and one in Utah. The mine side is not the problem: Australia at roughly a third of zircon output and South Africa at another fifth or so make the raw material one of the more diversified on this sheet, and the sand itself has no strategic scarcity at all. The tightness is entirely in the conversion and qualification chain. A disruption at the separation step would not stop reactors quickly, since fuel is fabricated 12–24 months ahead and utilities hold cores in inventory, but it would show up within a couple of refuelling cycles and there is no alternative supplier to call. The Western alternative is expansion of existing plants rather than new entry, because the separation chemistry is mature but the qualification and capital burden keeps newcomers out; both Framatome and ATI have announced capacity work, and the realistic timeline is the second half of this decade.

On the mine side, Iluka Resources and Tronox in Australia, Rio Tinto's Richards Bay Minerals and Tronox Namakwa in South Africa, Kenmare in Mozambique, and Chemours in Florida and Georgia supply the zircon. On the metal side, Framatome's Cezus plants at Jarrie and Ugine, ATI in Albany, Oregon, Westinghouse's Western Zirconium in Ogden, Utah, and Rosatom's Chepetsky Mechanical Plant are the producers that matter. The alloys themselves are Zircaloy-2 and Zircaloy-4, Westinghouse's ZIRLO, and Framatome's M5. On the substitution side, General Electric's IronClad FeCrAl and General Atomics' SiGA silicon-carbide cladding are the two development programs worth tracking, alongside chromium-coated cladding from Framatome and Westinghouse.

Zircon is a bulk mineral sold on bilateral contracts with quarterly or half-yearly resets, and its price has been remarkably steady: premium grade cif China ran between $1,530/t and $2,300/t across 2021 to 2025, and Chinese zirconium sponge between $22/kg and $30/kg over the same period. That stability is worth understanding, because it is the opposite of the hafnium that comes out of the same plant. Because zircon is a coproduct of titanium feedstock mining, its supply follows pigment demand rather than nuclear demand, and USGS reported world output falling 12% in 2025 as older heavy-mineral-sand deposits deplete — a decline driven by the titanium market, not by anything nuclear. The money in the chain sits at two ends. Mineral-sands miners earn a commodity margin on roughly $2 billion a year of zircon, and the four metal producers earn a much better one on nuclear-grade tube, protected by a qualification barrier that takes a new entrant the better part of a decade to clear. If you are underwriting anything here, the asset is the qualification and the customer relationship, not the ore.

6 minZirconium: This Unreactive Metal That Powers Nuclear Reactors, Fake Diamonds, and More (Element 40).Daily Dose of Science · 1k+ views

Zirconium and Hafnium, Mineral Commodity Summaries 2026 (USGS) · Zirconium and Hafnium, Professional Paper 1802-V (USGS)

Hafnium is what comes off the step that makes zirconium fit for a reactor. It occurs in zircon at roughly 2% of the zirconium and is chemically almost indistinguishable from it, so nobody mines hafnium and nobody ever will: every kilogram in existence was separated out because someone wanted the zirconium purified. That makes its supply curve vertical. World production is only about 70–75 tonnes a year from four countries, it cannot rise unless nuclear-grade zirconium production rises, and less than 1% of the hafnium contained in mined zirconium is actually recovered. Demand, meanwhile, is mostly not nuclear at all. The largest use by far is single-crystal superalloys for turbine blades, where 1–2% hafnium improves grain-boundary strength and coating adhesion; the next is hafnium oxide, the high-k gate dielectric in every logic chip since Intel's 45 nm node in 2007. Control rods, where hafnium's neutron absorption is the point rather than the problem, are a distant third. The metal is made by chlorinating the separated hafnium tetrachloride and reducing it with magnesium, the same Kroll chemistry as zirconium, then refining by electron-beam melting or iodide decomposition for the crystal-bar grade that superalloys need.

Hafnium's thermal neutron absorption cross-section is about 104 barns, roughly 600 times zirconium's 0.18, and it holds that appetite over a broad energy spectrum while resisting corrosion in hot water — a rare combination that makes it an excellent control-rod material with a long service life. In superalloys it does something no cheap element does: a fraction of a percent measurably extends creep life and stops thermal-barrier coatings spalling. The weakness is entirely economic rather than technical. This is a byproduct with no supply response, so a demand shock has nowhere to go but price, and the market is small enough that a single new aerospace program or a fab buildout moves it. The failure mode to plan for is not scarcity but repricing: hafnium went from $781/kg in 2021 to $6,130/kg in 2023 and back to about $3,800/kg in 2025, and a buyer with no inventory and an annual contract can find a bill of materials rewritten underneath them.

If you need a control rod that lasts, hafnium is the best material available and boron carbide or silver-indium-cadmium are the substitutes, both at a real cost in rod lifetime and replacement frequency. If you are alloying a turbine blade, there is no substitute at all for what the hafnium addition does, and the quantity is small enough that the price barely touches the part cost — so the question is availability of crystal-bar grade, not dollars. If you are specifying a gate dielectric, likewise: the volumes are tiny and the material is irreplaceable. The advice that actually matters is procurement rather than design. Qualify a second source before you need one, hold physical inventory rather than a contract, and model the price as a byproduct with no supply response — a high price does not bring out more metal, it only rations what exists. Anyone whose business case assumes hafnium reverts to its 2021 price is assuming something the supply structure cannot deliver.

World production roughly 70–75 t/yr from four countries, and USGS publishes no world figure at all because the data are not available · present in zircon at about 2% of the zirconium, and under 1% of that contained hafnium is recovered · thermal neutron absorption cross-section about 104 barns against zirconium's 0.18 · unwrought metal averaged $781/kg in 2021, $6,130/kg in 2023, and about $3,800/kg in 2025, with Q4 2025 spot at $6,300–7,000/kg against a 2023 record near $7,100 · roughly 8x price movement over five years, against under 2x for the zircon it comes from · superalloys are about 61% of demand, control rods about 11%, on the last published end-use split.

The separation plants, and there are about five of them. Framatome's Cezus operation in France, ATI in Albany, Oregon, Westinghouse's Western Zirconium in Ogden, Utah, Rosatom's Chepetsky Mechanical Plant at Glazov, and a Chinese group around CNNC and Baoji are the world list. The last published split puts France near half of world output and the United States near 44%, which is the rare case on this sheet where the concentration is real but sits with allies rather than with a strategic rival — the same argument entries 017 and 024 make for high-purity quartz and niobium. China and Germany are the leading exporters of unwrought hafnium, and 54% of US imports come from Germany with 21% from China. What makes this chokepoint unusual on this sheet is that it cannot be relieved by wanting it more: hafnium output is set by nuclear-grade zirconium output, so building hafnium capacity means building zirconium capacity for a nuclear fuel market that is not asking for it. This is the same coproduct trap that constrains iridium, and it is why the price behaves the way it does. China tightened dual-use export controls on hafnium in 2024 and the effect was immediate — Chinese unwrought hafnium exports fell from 5,001 kg in January 2025 to 499 kg that September, a 90% drop, and Rotterdam spot roughly doubled in the following quarter. The Western alternative is incremental expansion at Framatome and ATI, both announced, both realistically second-half-of-decade, and neither large relative to a market this small.

Framatome's Jarrie plant, which produces electrolytic hafnium crystals as the byproduct of its nuclear zirconium line, and ATI in Albany, Oregon are the two Western producers that matter. On the demand side: single-crystal superalloys such as CMSX-4, René N5, and MAR-M 247, which carries about 1.5% hafnium, in the hot sections of essentially every large jet engine and industrial gas turbine; hafnium oxide as the high-k gate dielectric in every logic chip since Intel's 45 nm node; the C-103 niobium-hafnium alloy used in rocket nozzle extensions, which is the reason a hafnium squeeze shows up in launch-vehicle supply chains; hafnium electrodes in plasma cutting torches; and hafnium carbide, among the highest-melting compounds known, in experimental ultra-high-temperature ceramics. Control rods in commercial PWRs are the nuclear use, and the one where a substitute exists.

This is a market of roughly 70–75 tonnes a year, which at recent prices is a few hundred million dollars in total — small enough that no producer builds capacity for it on its own account, and small enough that it does not appear as a line item in most of the industries that depend on it. That combination is exactly what produces violent repricing. Value sits with whoever holds inventory and with the separation plants, which capture a margin on hafnium that is pure upside against a nuclear zirconium business sized by fuel contracts. The demand side is what changed recently: aerospace build rates and semiconductor capacity both rose against a supply that is structurally flat, and the Chinese export controls removed a supplier from a market with four. For an investor, the durable position is a separation plant or a qualified inventory book, not a mine, because there is no mine. For anyone consuming it, the honest planning assumption is that hafnium is priced like a collectible with an industrial use: thin, opaque, and prone to moving faster than a procurement cycle.

3 minHafnium metal - Used in NUCLEAR reactors, exotic bullion and rings!KickAss Science · 1k+ views

Zirconium and Hafnium, Mineral Commodity Summaries 2026 (USGS) · Hafnium factsheet (SCRREEN2)

Black mass is the fine powder left after spent cells and factory scrap are discharged, shredded, and screened, with the steel casings, copper and aluminum foil, and plastics taken out. What remains is cathode metal oxide, graphite, binder, and residual electrolyte. Hydrometallurgical refining dissolves that powder in sulfuric acid with hydrogen peroxide as a reductant at roughly 60–90 °C, drops out iron, aluminum, and copper as impurities, and then runs the pregnant solution through a solvent-extraction train, typically D2EHPA to pull manganese and Cyanex 272 to make the cobalt-nickel split. The separated streams crystallize as battery-grade nickel sulfate, cobalt sulfate, and manganese sulfate. Lithium stays in solution through the whole circuit and comes out last, precipitated from the raffinate as carbonate with soda ash or converted to hydroxide. The products are the same chemicals a mine-fed refinery makes, so they drop straight into an existing precursor plant, which is the reason most Western recyclers chose this route over smelting.

Recovery of the valuable metals is very high, generally 95–99% for nickel, cobalt, and copper, and the process runs near ambient temperature so its energy use is a fraction of a smelter's. It also recovers lithium and can recover graphite, both of which a smelter loses. The weaknesses are lithium and salt. Lithium recovery is usually quoted at 85–95% and it arrives in a dilute stream at the end of the circuit, which is the least valuable place to find it, and every tonne of metal produced generates several tonnes of sodium sulfate that has to be crystallized, sold into a low-value market, or disposed of. The circuit is also tuned to a nickel and cobalt grade, so an LFP or sodium-ion feed breaks the economics rather than just reducing the margin. The failure mode this industry actually hit is commercial: plants were sized for end-of-life packs that have not arrived, and Li-Cycle's Rochester hub ran out of money mid-construction and ended up under Glencore's control.

If your feed is nickel-cobalt production scrap or end-of-life NMC and NCA packs of known chemistry, hydrometallurgy is the right route and the nickel and cobalt pay for the plant. If your feed is mixed, damaged, or unsorted, a smelter will take it and a hydro plant will not, so read the pyrometallurgy entry next. If your feed is LFP, do the arithmetic before committing capital: there is no nickel or cobalt to recover, the value is lithium, copper, and aluminum, and the market prices that accordingly. The payable structure tells you where you stand. In 2024, US black mass at 20–25% nickel drew payables around 75% of contained nickel and cobalt value ex-works and 15–20% nickel material drew 60–65%, while lithium is frequently not paid for at all. Before sizing anything, check feedstock rather than demand: US shredding capacity was around 155,000 t in 2023 and forecast near 230,000 t for 2024, against roughly 90,000 t of end-of-life and manufacturing scrap actually available.

Recovery 95–99% for nickel, cobalt, and copper, and 85–95% for lithium · leaching in sulfuric acid with peroxide at roughly 60–90 °C · China holds on the order of 70–75% of world black mass processing capacity · US shredding capacity about 155,000 t in 2023 and near 230,000 t forecast for 2024 against roughly 90,000 t of available feedstock · black mass payables around 75% of contained nickel and cobalt ex-works US for 20–25% nickel material and 60–65% for 15–20% nickel material · the EU classified black mass as hazardous waste effective 5 March 2025, mandatory from 9 November 2026 · China lifted its black mass import ban on 1 August 2025.

Refining capacity, and it is the same chokepoint the primary chain already has. China holds something like 70–75% of world black mass processing capacity, and the sulfate and precursor plants that recycled metal feeds into are the same Chinese-dominated capacity described in the lithium chemicals and nickel sulfate entries. Recycling therefore does not route around the concentration; it re-enters the chain at the point where the concentration is worst. Trade policy in 2025 made this sharper rather than looser. The EU classified black mass as hazardous waste effective 5 March 2025, with the amended waste list mandatory from 9 November 2026, which bars export for recovery to non-OECD countries under the Basel Ban Amendment and puts intra-OECD shipments under prior notification. China went the other way, lifting its import ban on 1 August 2025 and treating black mass that meets new national standards as a non-waste raw material, with separate specifications for nickel-cobalt and LFP material. So the largest buyer opened its door in the same year the largest regulated seller closed part of its own. A disruption here does not stop anything immediately, because recycled metal is a single-digit percentage of cathode feed today, but it decides whether Western shredders have an offtake at all. The Western alternative is under construction rather than announced: Redwood Materials in Nevada and South Carolina, Ascend Elements' Apex plant in Kentucky, Cirba Solutions, and Glencore's rebuild of the Li-Cycle assets. The realistic timeline for meaningful non-Chinese refining is the late 2020s, and it is gated by feedstock volume more than by permitting.

Redwood Materials in Nevada and South Carolina is the largest US operation and takes whole packs and consumer cells rather than buying black mass. Ascend Elements runs the Apex plant in Kentucky and a lithium recovery line in Covington, Georgia. Li-Cycle built spoke shredding plants across North America and Europe before its Rochester refining hub stalled and Glencore took control. Cirba Solutions, Ecobat, and American Battery Technology Company are the other US names. In Europe, Altilium, Tozero, and Librec are the credible startups, and Northvolt's Revolt program is the cautionary tale. In Asia, SungEel HiTech in Korea and CATL's Brunp, GEM, and Huayou Cobalt in China run the largest capacity, and Brunp in particular is integrated straight into cathode production.

Black mass trades as a payable percentage of contained metal rather than at a headline price, which is why the metal cycle passes straight through to the recycler's margin. Nickel and cobalt do almost all of the work: with LME nickel in the mid-teens of thousands of dollars a tonne and cobalt having roughly quadrupled off its early-2025 low, an NMC feed can pay for processing and a low-nickel or LFP feed frequently cannot. Lithium is the swing factor rather than the base, and at the carbonate prices of 2024 it was often ignored entirely in contract terms. Capital intensity is the second problem. A full hydrometallurgical refinery is a chemical plant with a solvent-extraction train, effluent treatment, and a crystallization circuit, and it costs several hundred million dollars at commercial scale, which has to be amortized over a feed stream that does not yet exist at that volume. That mismatch explains the sector's write-offs: capacity was built against forecasts of end-of-life packs from vehicles still on the road, while the actual feed is production scrap whose volume tracks gigafactory yields and falls as those factories get better at making cells. If you are underwriting one of these plants, the questions that matter are where the tonnes come from, what chemistry they are, and who buys the sodium sulfate.

12 minExtraction of Cobalt and Lithium from Lithium-Ion BatteriesAlamist · 10k+ views

7 minWhat *Really* happens to used Electric Car Batteries? - (you might be surprised)JerryRigEverything · 5m+ views

EverBatt: A Closed-loop Battery Recycling Cost and Environmental Impacts Model (Argonne National Laboratory) · Developing a UK lithium-ion battery recycling industry, Faraday Insights Issue 20 (Faraday Institution)

Pyrometallurgical recycling feeds whole cells, modules, or black mass into a smelter and lets high-temperature metallurgy do the sorting. The organics burn, the graphite acts as a reductant and part of the fuel, and the charge separates into two phases: a metallic alloy or matte carrying nickel, cobalt, copper, and iron, and a slag carrying aluminum, manganese, lithium, and whatever else oxidizes. The alloy then goes into the same leach and solvent-extraction refining any nickel matte would, so the back end of the process is conventional base-metal hydrometallurgy and produces the same sulfates. The two reference operations are Umicore's Hoboken plant in Belgium, which runs a dedicated battery line rated at about 7,000 t/yr, roughly 20,000 electric-vehicle packs, alongside the smelter that processes electronic scrap and autocatalysts, and Glencore's Sudbury complex in Ontario, which takes battery material into an existing nickel smelter. Both sit inside plants that already existed for another purpose, which is the central economic fact about this route. Fluorine from the electrolyte and binder makes gas cleaning non-negotiable, so a smelter without an existing scrubbing train cannot simply start taking cells.

The feed flexibility is the whole point. A smelter takes mixed chemistries, unsorted packs, wet or damaged cells, and modules that were never designed to be taken apart, with no discharge step and no disassembly line, which removes the most labor-intensive and most dangerous part of a hydrometallurgical operation. It also handles material a hydro plant would reject outright, including packs pulled from crashed or burned vehicles. The cost is what goes into the slag. Lithium, aluminum, and manganese report there, and graphite and electrolyte are destroyed, so lithium recovery is poor to nil unless a slag-treatment circuit is added, and that circuit is a separate hydrometallurgical plant with its own capital cost against a low-grade feed. Energy use and direct CO2 are also much higher than the ambient-temperature route. The failure mode to watch is regulatory rather than technical: recycled-content and recovery-efficiency rules, including the EU battery regulation's lithium recovery targets, are written in a way that a smelter without slag treatment cannot meet.

Pick pyrometallurgy when your feed is heterogeneous and your value is in nickel and cobalt. If you are handling end-of-life packs from a mixed vehicle fleet, packs of unknown state of charge, or material with a fire risk, this is the route that works, and the discharge and disassembly cost you avoid is real money. Pick hydrometallurgy instead when the feed is clean, single-chemistry production scrap and you need the lithium, because that is where hydro's 85–95% lithium recovery pays and a smelter's near-zero does not. If your feed is LFP, neither route is comfortable, but a smelter is worse, since LFP's value is almost entirely the lithium that goes to slag. The strongest case for pyro is not greenfield at all: if you already own a nickel or copper smelter with permits, gas cleaning, and a refinery behind it, adding battery feed is an incremental project, and the marginal cost per tonne is far below the price a new hydrometallurgical plant would need.

Umicore's Hoboken battery line is rated at about 7,000 t/yr, roughly 20,000 electric-vehicle packs · smelting recovers nickel, cobalt, and copper into an alloy while lithium, aluminum, and manganese report to slag · lithium recovery is close to zero without a dedicated slag-treatment step, against 85–95% for the hydrometallurgical route · no discharge or disassembly step is required, which is where most of a hydro plant's labor cost sits · fluorine from electrolyte and binder makes flue-gas cleaning mandatory · the incremental capital cost of adding battery feed to an existing smelter is a fraction of a greenfield hydrometallurgical refinery, which runs into the hundreds of millions of dollars.

Smelter capacity that can legally and physically take this feed, which is a short list and largely European, Canadian, and Chinese. The constraint is not ore or geography but permits: a plant that melts batteries needs a hazardous-waste permit, a fluorine-capable gas cleaning system, and a base-metal refinery to take the alloy, and very few sites have all three. Umicore Hoboken, Glencore Sudbury, Nickelhütte Aue in Germany, and the Chinese integrated groups are most of the world's capability. That overlap is worth noticing, because the same smelters process spent autocatalysts for platinum-group metals, so battery recycling and PGM recovery compete for the same furnace hours and the same refinery slots. A disruption would push feed toward hydrometallurgical plants that cannot accept most of it, or toward storage. The Western position here is better than in most of this sheet: the capacity exists, it is in OECD countries, and the technology is a century old. What it does not do is solve lithium, and the EU's move to classify black mass as hazardous waste from 2025 constrains where uncomfortable feed can be shipped, which raises the value of domestic smelting capacity rather than lowering it. Anyone planning new capacity should assume three to five years for permitting a greenfield site and one to two years to qualify a battery line at an existing smelter.

Umicore's Hoboken plant is the reference operation and has run battery material since well before electric vehicles were a volume market, alongside its electronic-scrap and precious-metal business. Glencore's Sudbury smelter and the Nikkelverk refinery in Norway take battery feed into an existing nickel chain, and Glencore ended up controlling Li-Cycle's assets after that company's hydrometallurgical hub stalled. Nickelhütte Aue in Germany and Accurec, which pairs a thermal step with hydrometallurgy, are the other European operators. Retriev Technologies in North America and SungEel HiTech in Korea run thermal pretreatment ahead of chemical recovery. In China, the large recyclers are integrated straight into cathode production, so the smelting and refining steps are internal transfers rather than a market. Umicore is also the clearest example of the shared-furnace point, since the same site refines platinum-group metals out of spent autocatalysts.

The economics of this route are the economics of an existing asset. A smelter's fixed costs are already paid for by its primary business, so battery feed is evaluated on marginal terms: does the payable metal in the charge exceed the fuel, reagents, gas cleaning, and lost throughput on other material. That test is much easier to pass than the one facing a greenfield hydrometallurgical refinery, which has to pay back several hundred million dollars of capital on a feed stream that does not yet exist at scale. It also means capacity does not expand smoothly, since it depends on someone else's smelter having spare furnace time. Revenue is dominated by nickel, cobalt, and copper, so this route is directly exposed to the same prices as primary supply, and it is at its most profitable when cobalt is expensive, which is exactly when a mine would also be profitable. Recyclers frequently charge a gate fee on top of paying for metal, and for low-value or hazardous feed the gate fee is the business. The strategic argument for the route is not that it is the cheapest way to recover metal, because it usually is not; it is that it is the only way to deal with the material nobody else will take, and that capability has a price of its own.

2 minUmicore FAQ - How does the battery recycling process work?Umicore · 1k+ views

6 minWhy It's So Hard To Recycle Electric-Car Batteries | World Wide WasteBusiness Insider India · 500k+ views

Lithium-ion battery recycling processes: Research towards a sustainable course (Argonne National Laboratory) · Energy and environmental aspects in recycling lithium-ion batteries: Concept of Battery Identity Global Passport (Oak Ridge National Laboratory)

A sintered neodymium magnet is roughly 30% rare earth by weight, mostly neodymium and praseodymium plus whatever dysprosium or terbium the temperature rating required, which makes end-of-life magnets a far richer source of those elements than any ore. Two routes recover them. The short loop keeps the alloy intact: hydrogen is absorbed into the neodymium-rich grain-boundary phase, the magnet swells and disintegrates into a demagnetized friable powder, and that powder is degassed, blended to correct composition, jet milled, aligned in a field, pressed, and re-sintered into a new magnet. This is hydrogen processing of magnet scrap, developed at the University of Birmingham and commercialized by HyProMag. The long loop dissolves the scrap in acid and runs it back through solvent extraction to separated oxides, then through metal-making and alloying, which is the same route a mine concentrate takes. The short loop is cheap and low-energy and needs clean, known-composition, de-coated feed. The long loop takes mixed, oxidized, or unknown scrap and produces a commodity oxide, at the cost of re-entering the chain exactly where it is most concentrated.

Recycled magnets should be the obvious answer to a chain that is more than 90% Chinese at every step, and the grades support it: a tonne of magnet scrap carries something like 300 kg of contained rare earth, against a few percent in the best ore. The short loop in particular skips separation, metal-making, and strip casting entirely, so it avoids the steps that take a Western entrant three to six years to build. The weakness is that none of that matters without feed. Magnets are 1–3 kg buried inside an electric-vehicle traction motor, 10–20 g inside a hard disk drive, and hundreds of kilograms per megawatt inside a direct-drive wind turbine, always bonded or press-fit inside a rotor that nobody designed to come apart. There is no reverse-logistics chain and no standard part to target. The failure mode is a plant that is technically sound and starved: capacity gets built against a feed forecast, the collection network does not materialize, and the plant runs at low utilization on production scrap it has to bid for.

If you make magnets and generate your own swarf, recycle it, because machining a sintered magnet to final dimensions typically removes 20–30% of its mass and that scrap is clean, single-composition, and already in your building. That is the one part of magnet recycling that works at scale today, mostly inside Chinese plants. If you are sourcing recycled magnets as a buyer, ask which loop the supplier runs and what feed it takes, since a short-loop supplier's output composition is set by its input and cannot be tuned far, while a long-loop supplier is selling you oxide that still has to go through separation and metal-making somewhere. If you are building a business, solve collection first and processing second: partner with electronics recyclers, data-center decommissioners, and vehicle dismantlers before you order equipment. And do not model end-of-life electric-vehicle motors as feedstock this decade, because vehicles built from 2020 onward will not be scrapped in volume until the mid-2030s.

Sintered NdFeB is roughly 30% rare earth by weight, so a tonne of magnet scrap carries on the order of 300 kg of contained rare earth · machining a sintered magnet removes 20–30% of its mass as swarf · magnet content runs 1–3 kg per EV traction motor, 10–20 g per hard disk drive, and hundreds of kg per MW in a direct-drive wind turbine · recycled material supplies well under 1% of the world magnet chain · HyProMag's Birmingham plant runs at about 100 t/yr and its Texas hub targets 750 t/yr of recycled sintered magnets plus about 807 t/yr of co-products within five years of commissioning · Cyclic Materials' Mesa, Arizona plant is sized for up to 25,000 t/yr of end-of-life product feed · NdPr oxide has run roughly $50–175/kg over five years, with a $110/kg floor under MP Materials' US production.

Collection, and then, for one of the two routes, the same separation step that is the chokepoint in the primary chain. Long-loop chemical recycling produces mixed or separated rare-earth oxide, and that oxide has to go through solvent-extraction separation and metal-making, which are roughly 91% Chinese, and then through magnet sintering, which is about 94% Chinese. So a long-loop recycler outside China solves the mining problem and not the processing one, unless it is paired with Western separation capacity such as Solvay's La Rochelle plant or a Western magnet maker. Short-loop processing avoids that trap by design, which is why HyProMag and Noveon both pair recycling with sintering rather than selling oxide. The binding constraint in both cases is upstream of all of it. Nobody collects magnets, because they are small, heavy, glued inside assemblies, and worth nothing to the person holding the assembly, and the one feed that is well organized is hard drives, where data-security rules usually require shredding that mixes the magnets into steel fines. What non-China capacity exists is measured in hundreds of tonnes a year: HyProMag Birmingham at about 100 t/yr, a plant now opened in Germany, and a Texas hub targeting roughly 1,500 t/yr of payable output within five years of commissioning, against a world sintered magnet market in the hundreds of thousands of tonnes. Cyclic Materials is attacking the collection problem directly, with a Mesa, Arizona plant sized to take up to 25,000 t/yr of end-of-life products and supply agreements with electronics recyclers, Vacuumschmelze, and Neo Performance Materials. A useful comparison is the next entry: spent autocatalysts already supply a quarter to a third of platinum-group metal demand, and every reason that worked is missing here.

HyProMag, a subsidiary of Maginito and Mkango, runs the Birmingham plant using hydrogen processing of magnet scrap, has opened a German facility, and is building HyProMag USA in the Dallas-Fort Worth area. Cyclic Materials, based in Kingston, Ontario, runs a two-stage system that produces a magnet concentrate and then a mixed rare-earth oxide, with a US plant in Mesa, Arizona and an offtake relationship with Vacuumschmelze. Noveon Magnetics in San Marcos, Texas makes sintered magnets from recycled feed and raised a $215 million Series C in 2026 to expand. In Europe, Solvay restarted rare-earth separation at La Rochelle in 2025, which gives long-loop recyclers a non-Chinese place to send oxide, and Carester is building separation capacity in France. On the feedstock side, Electronic Recyclers International and similar shredders are the partners that decide whether any of this has volume, and the hard-disk fleets of large cloud operators are the most concentrated single pool of end-of-life NdFeB in existence.

The value in the feed is real and the cost of getting at it is the problem. At NdPr oxide prices of roughly $50–175/kg over the past five years, and with a $110/kg floor under MP Materials' US output from the 2025 Department of Defense agreement, a tonne of clean magnet scrap carries tens of thousands of dollars of contained metal. But collecting that tonne means handling tens of tonnes of end-of-life product, and the labor to extract magnets from rotors and assemblies is manual and does not scale with automation the way shredding does. That is why Cyclic Materials' plant is sized in tonnes of incoming product rather than tonnes of magnet. Capital intensity favors the short loop heavily: hydrogen decrepitation equipment plus a sintering line is a fraction of the cost of a solvent-extraction cascade, and it produces a finished magnet rather than an intermediate. The revenue risk is price. Recyclers are exposed to an NdPr price that China can move with a licensing announcement, and a price collapse hurts them more than it hurts a Chinese integrated producer with mine-to-magnet margin to absorb it. For anyone underwriting this sector, the honest framing is that magnet recycling is currently a supply-security business supported by policy and offtake commitments rather than a cost-competitive one, and the number that decides whether it becomes the latter is tonnes collected, not tonnes of nameplate capacity.

2 minRecycling Magnets and Rare Earth ElementsESCI - European Science Communication Institute · 5k+ views

Addressing Criticality in Rare Earth Elements via Permanent Magnets Recycling (Ames Laboratory) · Short-Loop Recycling of Nd-Fe-B Permanent Magnets: A Sustainable Solution for the RE2Fe14B Matrix Phase Recovery (Materials)

A catalytic converter is a cordierite ceramic honeycomb coated with a high-surface-area alumina washcoat carrying platinum, palladium, and rhodium, typically a few grams per unit. Recycling starts by decanning, cutting the steel shell open to get the monolith out, then crushing, milling, and blending it into a homogeneous powder. Sampling and assay come next, and they are the commercial heart of the process, because metal content varies by vehicle model, by emissions standard, and by how the converter aged, so buyer and seller are negotiating against an analysis rather than a spec. Recovery then goes one of two ways. Smelting with an iron or copper collector in a plasma or electric furnace concentrates the platinum-group metals into a metallic phase while the ceramic goes to slag, or the powder is leached directly in chloride or aqua regia media. Either way the concentrate enters the same precious-metal refinery that handles primary matte from South Africa, where the metals are separated by solvent extraction and selective precipitation. Grades are what make this work: spent autocatalyst runs roughly 1,000–2,000 g/t of combined platinum-group metal against 3–6 g/t in Bushveld ore.

This is the recycling route on the sheet that already supplies real volume. Recycling averaged about 24% of total platinum supply over the five years to 2024, and automotive scrap accounts for 75–84% of all platinum-group metal recycling depending on the metal, which matters because autocatalysts take roughly half of world platinum and about 80% of both palladium and rhodium. Four things made it work, and they are worth naming because the magnet entry is missing all four: value density high enough that shipping a barrel of powder across an ocean is trivial next to its contents, a standard part in a standard location on every vehicle, a mandated end-of-life vehicle process in Europe that requires depollution before scrapping, and refiners who already existed to process the primary metal. The weakness is that collection is procyclical. Converters get held back when prices are low and released when they are high, so the supply that recycling adds arrives late and is smallest exactly when the market is tightest. The failure mode is not technical; it is that a scrap merchant with a shed full of converters is effectively long platinum-group metals and behaves like it.

If you are deciding where to put recovery capital, this is the mature end of the market and the question is whether you can buy feed rather than whether you can process it. Compete on assay accuracy, settlement terms, and payment speed, since the metallurgy is a commodity service and the collectors sell to whoever pays fastest against a fair analysis. If you are a buyer trying to reduce platinum-group metal exposure, do not model secondary supply as a hedge you can add; a quarter of world platinum already comes from recycling and it is priced into the market you are buying in. If you are forecasting long-term supply, watch fleet scrappage rates and the platinum-group basket price rather than electric-vehicle adoption, because the internal-combustion fleet turns over on a 15-to-20-year cycle and the converters coming back today were fitted to cars built well before anyone was buying electric ones. If you want the contrast, read the magnet recycling entry: same logic, none of the preconditions, and under 1% of supply as a result.

Recycling averaged about 24% of total platinum supply over the five years to 2024 · automotive scrap is 75–84% of all platinum-group metal recycling depending on the metal · autocatalysts consume roughly 50% of world platinum and about 80% of both palladium and rhodium · spent autocatalyst grades run roughly 1,000–2,000 g/t of combined platinum-group metal against 3–6 g/t in Bushveld ore · a converter typically carries a few grams of platinum-group metal · US catalytic converter theft reports rose from 1,298 in 2018 to 52,206 in 2021 · rhodium peaked near $29,000/oz in early 2021 and fell back to single-digit thousands.

Refining, not collection, and it is a chokepoint held almost entirely by OECD countries. Collection is fragmented and competitive, running through tens of thousands of scrapyards, dismantlers, and aggregators, which is exactly why it works. The concentration sits at the far end, where the concentrate has to go into a precious-metal refinery capable of separating six chemically similar elements, and that is the same short list that refines primary matte: Umicore in Hoboken, BASF, Johnson Matthey, Heraeus, Sibanye-Stillwater's Columbus operation in Montana, and Dowa and Tanaka in Japan. Those refineries are the shared asset behind several entries on this sheet, since the same furnaces and refining trains handle spent catalysts, electronic scrap, and battery material, so capacity is genuinely contested rather than nominally shared. A disruption at a major refinery would delay metal by months rather than destroy it, because the material is stable and can be stockpiled, and there is no export-license risk of the kind that dominates the Chinese chokepoints. The real threat to this supply is slower and structural: as the internal-combustion fleet shrinks, the feedstock shrinks with it, on a lag of 15–20 years. Palladium and rhodium recycling volumes will peak and then decline, which is one reason the industry's attention is moving to platinum, where hydrogen and glass demand does not depend on tailpipes.

Umicore's Hoboken plant, BASF's precious-metal refining at Seneca, South Carolina, Johnson Matthey in the UK, Heraeus in Hanau, and Sibanye-Stillwater's Columbus, Montana operation are the refiners that set terms in this market. Elemental Holding in Poland, PGM of Texas, and Ecotrade are among the larger aggregators that buy converters and sell assayed lots. The European Union's End-of-Life Vehicles Directive is the regulation that made depollution and converter removal a required step rather than a choice, and it is the single clearest reason European collection rates are high. On the downside, the rhodium spike of early 2021 produced a wave of converter theft, with US reports rising from 1,298 in 2018 to 52,206 in 2021, and a set of state laws in Texas, California, and elsewhere requiring scrap dealers to record vehicle identification numbers and seller identity before buying a converter.

Converters trade on assay against a platinum-group basket, so the collector's margin comes from buying accurately, hedging the basket, and financing inventory between purchase and settlement, which can take weeks. Refiners charge a treatment and refining fee plus a returnable percentage of contained metal, and they set lot-size minimums that push small collectors to sell to aggregators instead. A standard converter has been worth anywhere from tens to several hundred dollars depending on the model and the prevailing basket price, which is precisely why theft tracked the rhodium price so closely. The structural feature that distinguishes this from every other recycling route on the sheet is that the economics work without a subsidy: at normal prices the metal in the feed pays for collection, transport, assay, and refining with margin left over, which is why the network built itself over three decades rather than being planned. The vulnerability is symmetrical. When prices fall far enough, collection slows within months, converters go into storage, and secondary supply drops just as the mining side is also cutting back. Anyone modeling platinum-group metal balances should treat recycled supply as price-elastic and lagging rather than as a stable base, and should assume that the long-run feedstock declines with the combustion fleet even as recovery rates improve.

13 minSmelting Catalytic Converters, Platinum, Palladium, & Rhodiummbmmllc · 500k+ views

5 minRecycling Factories – Precious Metals Extraction of Catalytic ConvertersFactories · 10k+ views

Platinum group metals factsheet (SCRREEN2) · Recovery of Platinum Group Metals from Spent Automotive Catalysts: A Review of Processes and Challenges (Materials)