The State Wants the Plant, Not the Pile

FRONT MATTER Thesis

On 7 November 2025 China suspended six export-control announcements it had issued a month before. It left the April 2025 rare-earth licence regime untouched: those exports stayed legal, but every one of them now needed a permit an official could withhold. The suspension was the headline. The licence regime it left standing is the machine.

The document is Ministry of Commerce and General Administration of Customs Announcement 2025 No. 70¹, dated 7 November 2025. It suspends Announcements 55, 56, 57, 58, 61 and 62 — the 9 October 2025 package that had extended Chinese jurisdiction to foreign-made goods containing 0.1% or more Chinese rare-earth value — from that day until 10 November 2026. Announcement No. 18 of 4 April 2025², which placed samarium, gadolinium, terbium, dysprosium, lutetium, scandium and yttrium — seven of the seventeen rare earths, the chemically near-identical metals that high-strength permanent magnets are made from — and their oxides, alloys, compounds and magnets under export licence, is not on the suspension list. The US Geological Survey confirms it³ in the February 2026 Mineral Commodity Summaries: "In November, China suspended the October export controls for 1 year. The April export controls remained in effect."

Which announcement survived matters more than which ones were suspended. The popular version of resource nationalism is a state filling a warehouse with metal and refusing to sell. Very little of that is happening. What is happening is narrower, more durable and much harder to reverse: governments on both sides of the divide are acquiring positions in processing plants, and both are writing licences that decide, case by case, who gets served. China kept shipping magnets to Europe in record monthly volumes through early 2026 while cutting shipments to the United States for four consecutive months, with the licence regime in force throughout. The United States, on 27 August 2026, began requiring that 100% of domestic sales of battery black mass (the metal-bearing powder left when used batteries are shredded) and tungsten scrap go to American buyers. Neither government is hoarding a pile. Both are taking the plant and the permission slip.

Run the test metal by metal and the answers separate cleanly. Rare earths meet four of the five criteria and are close on the fifth. Copper meets none of them fully: it has real state action at the mine gate in the Democratic Republic of the Congo and Indonesia, a genuine processing concentration in China, and a US tariff episode whose price signature has already faded. Silver meets none of them at all, and its own documents say so — the United States Treasury has held the same 498 tonnes since at least 2021, London vault stocks are up 16.6% year on year, and the President declined in January 2026 to put a tariff on processed critical minerals. What puts silver in this conversation is the narrative around it, not anything a state has done.

The headlines have it wrong about what capital is chasing. If the asset being accumulated is processing capacity rather than metal, the scarce input is the ability to build and run a separation plant, the chemical works that pulls one rare-earth element apart from the next: permits, solvent-extraction chemistry, people who have done it before. Capital is abundant; the US Office of Strategic Capital alone has credit-subsidy funding that can support up to $100 billion of lending. Execution is not. Lynas moved its heavy rare-earth expansion from Texas to Malaysia and watched the Malaysian budget go from about A$180 million to about A$294 million. That gap — between money committed and tonnes commissioned — is where the next four years of this trade will actually be decided.

FRONT MATTER Executive takeaways

1. 1The licence regime is a targeting instrument, and it was fired during the truce. US imports of yttrium (one of the seven rare earths China put under licence) from China fell from 333 tonnes in eight months to 17 tonnes in the eight months after April 2025 — a 94.9% cut — and on 6 January 2026, eight weeks into the general suspension, Beijing banned all dual-use exports (goods with both a civilian and a military use) to Japanese military end users outright. High confidence
2. 2Rare earths are transitioning, and the reserve asset is separation capacity, not oxide. Oxide is the part-refined powder rare earths trade in; separation capacity is the plant that sorts one element from another. China mines 69.2% of world rare earths but refines over 90%; the US National Defense Stockpile's entire fiscal-2025 acquisition plan was 450 tonnes of magnet block, about 0.19% of annual world magnet output. High confidence
3. 3Copper's premium is the carrying cost of an unexercised option. The COMEX-minus-LME gap (New York's copper futures price against London's) fell from 40.0 cents per pound on 2025 averages to 11.0 cents in the second quarter of 2026, then widened to 16.0 cents by 31 July — enough to pull a twelve-year record 200,000 tonnes of imports in a month while the market priced only a 14.6% chance of the 2027 tariff landing. High confidence
4. 4Silver fails the test on the documents. The official sector (governments and central banks) was a net seller of 1.5 million ounces in 2025, and combined COMEX and London vault holdings of 1,245 million ounces at end-August 2026 equal about 409 days of forecast world demand. High confidence — in the negative
5. 5Washington now runs the same instrument Beijing does. A Commerce temporary final rule effective 27 August 2026 requires 100% of US sales of black mass and tungsten scrap to go to US persons, with case-by-case exceptions at official discretion, under Defense Production Act authority — the standing statute that lets a president direct where scarce industrial materials go. High confidence
6. 6Policy-floor economics are already thinning where they are meant to be strongest. MP Materials' price-protection income was 54.1% of Materials segment adjusted EBITDA in the second quarter of 2026, and China's NdPr index (neodymium and praseodymium, the pair that goes into neodymium-iron-boron magnets) reached $107.72–110.72/kg on 2 September 2026 — at the floor. Moderate confidence
7. 7The binding constraint on the Western rewiring is execution, not capital. USA Rare Earth issued the government $882.3 million of stock and warrants as a condition precedent — something that had to happen before any money could move — to a $277.0 million grant it had not drawn a dollar of at 30 June 2026. Moderate confidence

FRONT MATTER The system, sized

The physical quantities all come from the February 2026 USGS Mineral Commodity Summaries⁴, which report 2025 estimates. World copper mine production was 23.0 million tonnes and world refinery production 29.0 million tonnes. World silver mine production was 26,000 tonnes. World rare-earth mine production was 390,000 tonnes of rare-earth-oxide equivalentA common accounting unit. Rare earths leave the ground and leave a plant in many different chemical forms, so every tonnage is restated as if it were oxide. Without that convention, quantities from different stages of the chain could not be added together or compared at all.. Those three numbers span three orders of magnitude, and the market structures underneath them have almost nothing in common. That is precisely why the mandate treats them as three separate trials rather than one story about "critical minerals" — a policy label the United States expanded to 60 entries on 7 November 2025, adding copper and silver to a list that already contained rare earths.

Source: USGS Mineral Commodity Summaries, February 2026 (copper, silver, rare earths chapters); Final 2025 List of Critical Minerals, 90 FR 50494, 7 November 2025.

Western commitments to this rewiring are large, recent, and mostly not yet spent. The Department of War took $400 million of MP Materials convertible preferred stock in July 2025 and guaranteed a $110 per kilogram floor for a decade. The Office of Strategic Capital lent $620 million to Vulcan Elements and $80 million to ReElement Technologies⁵ in a package the parties valued at $1.4 billion, alongside $50 million of Commerce equity and $550 million of private capital. The Department of Commerce signed a $277.0 million grant and a $1.30 billion loan guarantee with USA Rare Earth on 3 June 2026. Energy Fuels agreed to pay about $718 million in cash plus 65.85 million shares for the German magnet maker Vacuumschmelze, implying roughly $1.9 billion of equity value. The European Commission designated 47 strategic projects in the EU and 13 in third countries in its first round, and received more than 160 applications in its second round⁶, which closed 15 January 2026 — 21 of them for rare earths for permanent magnets. Those projects are being selected against the targets in Regulation (EU) 2024/1252⁷, in force since 3 May 2024: by 2030 the Union is to extract 10% of its strategic raw material consumption, process 40%, recycle 25%, and depend on no single third country for more than 65% of any one material. Japan's state resources agency committed up to C$47.668 million⁸ to a Namibian heavy rare-earth vehicle, with its first investment made on 23 July 2026.

Set against that, the defence tonnage is trivially small and politically decisive. The National Defense Stockpile's potential acquisitions for fiscal 2025 included 300 tonnes of NdPr oxide, 450 tonnes of neodymium-iron-boron magnet block and 60 tonnes of samarium-cobalt alloy. For fiscal 2026 the Annual Materials Plan was not released at all. An F-35 carries more than 900 pounds of rare-earth material; a Virginia-class submarine carries about 9,200 pounds. From 1 January 2027 the Department of War may not buy a samarium-cobalt or neodymium-iron-boron magnet whose entire supply chain — from the mining of the neodymium onward — touches China, Russia, Iran or North Korea. Three hundred tonnes of oxide will not move a market. A statutory sourcing ban on the world's largest defence buyer will.

SECTION 1 How the three prices are made

A copper trader in Shanghai, a silver dealer in London and a magnet buyer in Stuttgart are all said to face a "market price." Only one of them does. Copper has three live futures markets and a continuous physical premium on top of them. Silver has an over-the-counter spot market in London that a New York futures contract shadows. Rare earths have no exchange anywhere in the world. The difference determines what a state has to do to gain control, and how quickly anyone else can tell that it has.

Copper prices form at the London Metal Exchange, the COMEX division of CME Group and the Shanghai Futures Exchange, with each venue clearing metal against its own approved warehouses. Producers then add a physical premiumThe extra amount a buyer pays on top of the exchange price to have actual metal delivered to a particular place at a particular time. The exchange price is for metal somewhere; the premium is the cost of it being here. for delivery to a given region. USGS reports both layers separately⁴: in 2025 the COMEX high-grade first-position price averaged 480 cents per pound, LME grade-A cash averaged 440 cents, and the US producer cathodeRefined copper in the form the market actually trades: flat sheets of finished metal a fabricator can melt and draw into wire. Cathode is the output of a refinery, not of a mine, and the distinction runs through the whole tariff argument in this brief. price including premium averaged 490 cents. A fourth number sits underneath all of them and almost never makes the news — the treatment and refining charge, or TC/RC, which is what a miner pays a smelter to turn concentrateCrushed ore that has been upgraded at the mine into a powder rich in the target metal. It is the form ore travels in, and it is not yet metal — someone still has to smelt and refine it, which is exactly the step that gets fought over here. into metal. It is the miner-smelter split, and in 2026 it went to zero.

Silver works differently. The reference price is set in the London over-the-counterTraded directly between two parties rather than through an exchange. There is no central order book and no published volume, so the price comes from the dealers themselves and an outsider cannot see how much changed hands. market, where metal sits in vaults inside the M25 and changes hands by book entry. COMEX futures trade against that spot price, and the gap between them — the exchange-for-physical, or EFP — is the cost of moving metal between the two pools. When metal will not move, the EFP widens and the London lease rate, the cost of borrowing physical silver for a month, goes up. Those two numbers are the market's stress gauge, and both went to extremes twice in fifteen months. Metals Focus, writing for the Silver Institute in April 2026⁹, describes 2025 as a year "marked by elevated lease rates and regional liquidity tightness, culminating in a squeeze in October, stressing parts of the supply chain for refining and manufacturing."

Rare earths have no equivalent. There is no futures contract, no warehouse warrantThe document of title to a specific lot of metal sitting in an exchange-approved warehouse. Futures are settled by handing over warrants rather than by moving metal, which is why a market can run short of warrants while the world is not short of the metal itself., no daily settlement. Four different things get called the price. Price-reporting agenciesPrivate firms that survey buyers and sellers and publish an assessed price for a commodity no exchange quotes. The number is a considered estimate of where deals are being done, not a record of cleared trades — a difference that matters when nobody can check it. publish assessments from surveyed transactions — USGS's rare-earth chapter sources its price table to Argus Non-Ferrous Markets³. The China Rare Earth Industry Association publishes an index. Chinese state groups set a quarterly concentrate price by administrative decision. And since 1 October 2025 the United States has run a fourth number: a $110 per kilogram floor for NdPr that the Department of War pays MP Materials the difference on. None of these is a clearing price in the sense a copper trader would recognise. Each is a claim about what the price should be, made by an institution with a position.

A copper squeeze is visible in warehouse data the same afternoon. A rare-earth squeeze shows up when a national customs agency publishes trade statistics, weeks later, and even then only in tonnages that mix restricted and unrestricted grades. China stopped publishing its rare-earth mining and separation quota volumes after 2024. USGS's estimate of Chinese mine production is 270,000 tonnes for both 2024 and 2025 — the same number twice, which is what an estimate looks like when the underlying disclosure has gone dark. A market that cannot see the supply schedule prices on headlines. That gives whoever writes the headlines a reason to keep writing them.

SECTION 2 What an export control actually is

Announcement 18 did not ban anything. It required a licence — and a licence is a queue, with an official at the end of it.

The text is short. MOFCOM and the General Administration of Customs Announcement No. 18 of 2025², dated 4 April 2025 and effective the same day, lists seven elements — samarium, gadolinium, terbium, dysprosium, lutetium, scandium and yttrium — each in three item categories covering metals, alloys and targets; oxides and mixtures; and compounds and mixtures. Named subcategories include samarium-cobalt alloys, terbium permanent magnets and dysprosium-iron alloys. The operative sentence says only that "the exporter shall apply for the license from the competent commercial authority of the State Council." Exports remain lawful. They are simply now discretionary.

Six months later Beijing widened the instrument in three directions at once. Announcement No. 61 of 9 October 2025¹⁰ claimed jurisdiction over goods made outside China when Chinese-origin rare earths account for 0.1% or more of their value, and over goods made abroad using Chinese rare-earth technology. The translated text¹¹ states that applications for items destined for foreign military users "will not be permitted," and the Chinese original applies case-by-case review to semiconductor applications at 14 nanometres and below, memory of 256 layers and above, and military artificial-intelligence development. Announcement No. 62¹², the same day, put the technology itself under control: mining, smelting and separation, metal smelting, magnet manufacture including samarium-cobalt, neodymium-iron-boron and cerium magnets, and secondary recovery from scrap, plus the assembly, maintenance and upgrading of production lines. Licensing, investment, knowledge sharing and joint research all count as transfers.

Then Beijing put most of that back in the drawer. Announcement No. 70 of 7 November 2025 suspended Announcements 55, 56, 57, 58, 61 and 62 until 10 November 2026, implementing the Kuala Lumpur understanding with Washington. What survived is the April regime on seven elements, plus the ordinary dual-use list. Read together, the sequence is a demonstration rather than a policy: Beijing showed it could reach into a German magnet inside a Korean motor inside an American vehicle, then withdrew the reach for a year while keeping the licence desk open.

The trade data answers what the licence desk actually does. CSIS, reviewing the first year on 27 April 2026¹³, traced yttrium: US imports from China ran 333 tonnes across the eight months before the April 2025 controls, and 17 tonnes across the eight months after — a 94.9% cut — recovering only to 20 tonnes in February 2026. Over the same period Chinese magnet exports set records, with November 2025 volumes up 13% while shipments to the United States fell 11% and shipments to the European Union rose 60%. That is triage. Beijing kept the metal moving and chose who received it. CSIS's own conclusion is the right one: licensing "has been uneven, granting certain countries a more stable supply than others."

China has, three times, done the harder thing and imposed an outright ban — and the three cases are the closest available experiment on what a real embargo does. In December 2024 it banned all exports of gallium, germanium and antimony to the United States. USGS's 2026 chapters record the results with unusual clarity. American germanium metal imports fell 67% in 2025 to 7,000 kilograms¹⁴, while germanium dioxide imports rose 55% to about 17,000 kilograms as buyers shifted to Canada. Antimony ran from $9.80 a pound in August 2024 to $27.50 in June 2025 and back to $20.30 by November. And gallium — where China holds 99% of world primary low-purity production¹⁵ — saw US imports more than double in 2025 at an average unit value of $580 a kilogram, 30% above 2024, because in November 2025 China lifted the gallium ban for one year.

Beijing embargoes the metals where its share approaches totality and Western volumes are tiny, and licenses the metal where its share is high but the tonnage matters commercially. Both concessions were time-boxed: the gallium ban was lifted "for 1 year," the same phrase and the same November 2025 arrangement as the rare-earth suspension that runs to 10 November 2026. Everything Beijing conceded that month it conceded on a twelve-month lease.

Then, eight weeks into the suspension, Beijing pointed the instrument at a single country. On 6 January 2026 MOFCOM issued Announcement 2026 No. 1¹⁶, prohibiting the export of all dual-use items to Japan for military users, military purposes, and any other end user whose activities would strengthen Japan's military capability. It takes effect on publication, and it warns that organisations and individuals "from any country or region" who transfer Chinese-origin dual-use items to such Japanese entities will face legal consequences. The stated reason is Japanese leaders' remarks on Taiwan. CSIS, writing on 13 January 2026¹⁷, notes what is at stake: Japan imported over 5.2 million kilograms of rare-earth metals from China in 2024, 63% of its total, and is the world's largest importer.

A licence regime can be aimed at one country while it stays relaxed for everyone else, and a headline about a general suspension cannot carry that. A general control gets negotiated away in a summit; a country-specific, end-use-specific prohibition issued under the same statute does not, because it is not part of anyone's trade deal. Beijing suspended the measures aimed at the United States and, nine weeks later, aimed a narrower and legally cleaner one at Japan. Anyone modelling this as an on-off switch has the wrong model.

The enforcement layer arrived in 2026, and it matters more than the announcements. MOFCOM Announcement 2026 No. 26¹⁸, issued 24 June and effective 1 July 2026, set up an online reporting portal and a hotline, listed thirteen categories of violation including third-country routing and component separation, and offered rewards for verified reports. Morgan Lewis records the practice alongside it¹⁹: two Japanese nationals detained in Dalian in May 2026 over allegedly smuggled rare-earth items, and the chairman of a Chinese precision-optics firm placed under compulsory measuresThe term in Chinese criminal procedure for coercive steps taken against a person before any trial — detention, bail, or restriction of movement. It is not a conviction; it is the point at which an export-control question stops being commercial and becomes personal. in June 2026 for declaring germanium-bearing lenses as ordinary optical glass, with roughly three years of export records reviewed. A licensing regime becomes a control regime when the penalty for going around it becomes personal. That threshold was crossed in the first half of 2026.

SECTION 3 The chokepoint map

Chile digs up nearly a quarter of the world's copper. China digs up less than a twelfth. Then the ore moves, and the shares invert. USGS puts 2025 world copper mine production at 23.0 million tonnes and refinery production at 29.0 million tonnes⁴; China accounts for 1.8 million tonnes of the first and 14.0 million tonnes of the second. It refines almost eight times what it mines. That inversion — not the mine share — is what people mean when they call China a copper chokepointA single step that every unit of supply has to pass through, held by few enough hands that whoever holds it can decide who gets served and on what terms. A large market share on its own does not make one. The step also has to be hard to build somewhere else, and profitable enough that holding it costs the holder nothing.. The step where the inversion happens has a name worth keeping hold of: the midstream. It is everything between the pit and the finished part — smelting, refining, separating, alloying — and it is where this brief keeps finding the governments. A country can sit on a mountain of ore and still have to send it abroad to become usable metal.

Rare earths map the same way, only steeper. China mined 270,000 of the world's 390,000 tonnes of rare-earth oxide in 2025, or 69.2%. On the separation step — the solvent-extraction cascadeThe industrial method for pulling apart elements that are chemically almost the same. A solution of mixed ore meets an organic solvent that grips one element very slightly harder than its neighbour; run that step hundreds of times in series and the elements come out one at a time, each at high purity. The knowledge sits in operating experience rather than in patents, which is why it travels so badly. that pulls individual elements apart — the IEA's Global Critical Minerals Outlook 2026²⁰ puts China's share of refined supply "over 90%" for rare earths, alongside gallium, graphite and manganese. The same report finds that across energy minerals generally the average share of the top refining country reached 72% in 2025, up from 70% in 2023, and that the top three refining nations now hold 86% of key mineral processing against 82% in 2020. Concentration is rising, not falling, five years into a diversification push.

Silver has no such map. Mexico leads at 24.2% of 2025 mine output, Peru follows at 13.8%, China is third at 13.1%; no country reaches a quarter. Refining is dispersed — USGS counts 24 US refiners alone²¹ producing about 2,100 tonnes of commercial-grade silver in 2025 — and turning doréThe crude bar a mine pours on site: a rough mix of gold and silver with impurities still in it. It goes on to a refinery to be purified and cast into bars the market will accept. into good-deliveryBars meeting the London market's published standard for weight, purity and accredited refiner, so they can change hands without being assayed again. It is a paperwork standard as much as a metallurgical one — and a standard many refiners can meet is the opposite of a chokepoint. bars is not a technology anyone monopolises. Silver's real concentration is in vaulting and clearing, in London and in COMEX warehouses — a liquidity structure rather than a sovereign one. A state cannot seize a chokepoint that does not exist.

China's copper midstream loses money, and that is where the analogy to rare earths breaks. The IEA's March 2026 commentary²² reports that the annual copper treatment-charge benchmark settled at zero dollars per tonne in January 2026, against about $21 in 2025, and that spot charges have been negative since 2024. China's smelting share went from roughly 15% to about 50% since 2005, taking more than 90% of the world's smelter-output growth, and the result is that Chinese smelters now pay miners to take concentrate. Beijing halted roughly two million tonnes of planned new smelting capacity, top smelters agreed to cut output by more than 10% in 2026, and one custom smelter needed emergency government support to stay open. Ero Copper's own numbers confirm which side gains: its second-quarter 2026 release²³ cites improved treatment and refining terms worth more than $20 million of savings in 2026.

A chokepoint you are subsidising is a liability dressed as a weapon. Beijing can, in principle, refuse to smelt foreign concentrate; it would be refusing to lose money, and the concentrate would go to Japan, Korea, India or Indonesia, whose smelters are starved of feed. Compare rare-earth separation, where the same market share earns real margin, faces no rival with idle capacity, and is protected by a technology export ban that was drafted specifically to stop the capability leaving. The market shares look alike; the bargaining positions are opposite. Any analysis that treats "China processes half the world's copper" as equivalent to "China separates nine-tenths of the world's rare earths" has confused a share with a power.

SECTION 4 The state-action record, dated

Beijing licenses exports of processed material and bans the export of processing technology. Washington restricts what its defence contractors may buy, guarantees a price for domestic output, takes equity and warrants in the companies it funds, and — since 27 August 2026 — requires that all domestic sales of two recyclable feedstocks go to American buyers. These are the same family of instrument, deployed from opposite ends of the same chain. Three of those instruments get muddled constantly and they behave nothing alike. A tariff is a tax: it raises the price, and anyone willing to pay it still gets the metal. A ban removes the trade. A licence does neither — the trade stays legal, and an official decides shipment by shipment who is served, which is how it can be aimed at one buyer without any other buyer noticing.

The President wrote the thesis of this brief into Proclamation 11001. Paragraph 6 of the proclamation of 14 January 2026²⁴ reads: "Mining a mineral domestically does not safeguard the national security of the United States if the United States remains dependent on a foreign country for the processing of that mineral." The same paragraph records that the United States was 100% net-import reliantThe share of what a country consumes that it has to bring in from abroad, once its own exports and any change in stocks are netted off. At 100% it produces none of its own net supply, whatever it may dig out of the ground and ship away again. for 12 critical minerals as of 2024 and at least 50% reliant for a further 29, and that although America is the world's second-largest producer of mined rare-earth oxides, its limited processing capacity means those oxides are still exported for refining and reimported. Then, having found a threat to national security, the President imposed nothing. He directed Commerce and the Trade Representative to negotiate, told them to "consider price floors for trade in critical minerals," and reserved the option of "minimum import prices" later. The 180-day clock in paragraph 14 ran out on 13 July 2026. No tariff has followed as of this brief's cutoff.

Two USGS chapters explain the choice of materials in the August 2026 allocation order. "Tungsten has not been mined commercially in the United States since 2015."²⁵ China put selected tungsten items under export control in February 2025, and Rotterdam ammonium paratungstate went from $331 to $675 per metric tonne unitThe unit tungsten concentrate is priced in: one percentage point of a tonne of contained metal, not the tonne itself. A quote on this basis is therefore far smaller than a price per tonne, and the two must never be compared as though they were the same thing. across the year, with 65% concentrate from $266 to $551. For a country with no tungsten mine, scrap is the only domestic source there is. An order that would be merely protectionist in copper is close to existential in tungsten. Black mass is the same logic applied to batteries. Washington did not pick the biggest markets; it picked the two where recycling is the domestic supply chain.

Indonesia, the DR Congo and Chile are trying to move up the chain and tax the cycle rather than to accumulate leverage over buyers, and their policy gets conflated with Beijing's constantly. The policy has a name in the countries that use it: downstreaming. Forbid the export of raw ore or concentrate and a smelter has to be built at home to buy it, so the jobs and the processing margin stay in the country instead of leaving with the rock. It is aimed at a producer's own economy. It is not aimed at anyone else's supply, though it tightens it all the same.

SECTION 5 Exchange stock is not world stock

A metal that moves from a London shed to a Kentucky shed has not been consumed. It has been relocated — and the second shed may be closed to the rest of the world.

The words registered, eligible and on-warrant mean three different things, and inventory analysis routinely mixes them up. Registered metal carries a warrant and can be delivered against a futures contract. Eligible metal meets exchange specification and sits in the same warehouse but is not warranted. On-warrant is the LME's equivalent of registered, and the LME also reports cancelled warrants, which are tonnes booked for withdrawal but not yet gone. Headlines routinely compare a COMEX total against an LME on-warrant figure and call the difference a shortage. It is a definitional mismatch, and the single most common error in this field.

The same thing happens whenever a price gap opens between two venues: metal moves toward the richer one. Once the metal arrives, the gap closes — but the metal does not return, because returning it means paying freight to sell into the cheaper market. World inventory is unchanged; available inventory outside the United States is lower; and the price that the rest of the world pays reflects the tighter pool. Fragmentation of this kind is real and it is also self-limiting. It ends when someone finally exports the metal back, which they will do only if the gap inverts far enough to pay for two ocean crossings.

SECTION 6 Trial one: copper

The United States imported 1.7 million tonnes of refined copper in 2025. In 2024 it imported 903,000 tonnes. Nothing about American consumption changed by 88% in a year; reported refined consumption went from 1.58 to 1.70 million tonnes. What changed was a tariff expectation, and the response was the largest peacetime relocation of copper inventory on record.

Commerce opened its Section 232 copper investigation on 10 March 2025 and reported to the President on 30 June 2025; a parallel Section 232 investigation into processed critical minerals opened on 22 April 2025²⁶. Proclamation 10962 of 30 July 2025²⁷, published at 90 FR 37727, imposed 50% tariffs on semi-finished copper products and copper-intensive derivatives from 1 August 2025 — $15.5 billion of 2024 imports, sourced 19% from China, 17% Mexico, 15% Canada and 14% the European Union — while explicitly exempting ores, concentrates, mattes, cathodes, anodes and scrap. The market had spent five months pricing a tariff on cathode that never arrived. Traders had already shipped the metal.

The gap between the two exchange prices traces that trade and its unwinding, and it is the cleanest evidence in this brief.

The last three bars carry the copper verdict. In 2025 the COMEX-to-LME gap averaged 40.0 cents a pound. In the second quarter of 2026 Freeport-McMoRan's own 10-Q²⁸ reports COMEX settlements averaging $6.16 and LME settlements averaging $6.05 — a gap of 11.0 cents, roughly a quarter of the prior year's. By 31 July it had widened again to 16.0 cents. Both exchanges then set all-time highs within a day of each other, COMEX at $6.70 on 5 August 2026 and the LME at $6.56 on 6 August 2026.

The dislocation reverted; a smaller premium stayed. Sixteen cents was enough to make the trade work again: US refined copper imports exceeded 200,000 tonnes in July 2026²⁹, the highest month in twelve years, LME stocks fell 14% to 214,550 tonnes by 12 August, and the LME cash-to-three-month spreadThe gap between the price of metal for immediate delivery and the price for delivery in three months. When metal today costs more than metal later, buyers are paying up to get their hands on it now — the plainest sign that the near-term pool is thin. widened to $207.50 a tonne from about $34 at the end of July. COMEX inventories reached 765,700 short tons by 2 September, of which 477,500 were registered. Meanwhile the market's own pricing implied only a 14.6% chance that the recommended 15% refined-copper tariff would take effect on 1 January 2027.

Traders assign roughly one chance in seven to the tariff, and they are still shipping copper to America at a rate not seen in twelve years. An option that will probably not be exercised is holding a ten-to-sixteen-cent premium and close to seven hundred thousand tonnes of metal on the wrong side of an ocean. Copper's regional gap is not a bloc; it is the carrying cost of an unexercised policy option, and it collapses the day the option is closed either way.

Copper's real state exposure sits at the mine gate rather than in the tariff, and it is growing. The DR Congo produced 3.2 million tonnes in 2025, 13.9% of world mine supply, and on 29 June 2026 prohibited exports of copper concentrate outright, extending the logic of the cobalt quota regime it introduced in October 2025³⁰. Indonesia banned copper concentrate exports on 1 January 2025 and requires 100% of Freeport Indonesia's export proceeds to be parked in Indonesian banks — a rule tightened from 30% for 90 days in March 2025²⁸. Between them those two countries controlled 17.0% of 2025 world mine output and both now forbid the export of unprocessed material. That is real and structural, and it is about capturing smelting jobs rather than denying metal to an adversary.

The supply side has done more to the copper price than any policy in this record. Freeport's Grasberg block cave suffered an external mud rush in September 2025; the company collected $0.7 billion of insurance proceeds in the first quarter of 2026 and is still ramping back, with idle facility costs running through the income statement. USGS shows the consequence at the country level: Indonesian mine output fell from 1,010,000 tonnes in 2024 to 710,000 tonnes in 2025. At Rio Tinto's Kennecott — one of only two primary copper smelters left in the United States — refined metal production fell 49% year on year in the second quarter of 2026 and 34% across the half³¹, on mine sequencing, a 23-day concentrator maintenance and, in Rio's words, "reduced availability of high quality copper concentrate." American refined output fell an estimated 9% in 2025 on planned maintenance at both primary smelters. CSIS put the structural problem plainly in July 2025³²: the United States mined 1.2 million tonnes of copper in 2024 and had roughly 585,000 tonnes of smelting capacity, about half its own mine output, across two operating smelters. A country cannot tariff its way to refined copper self-sufficiency when its two smelters are down for repair.

Copper, then, is a metal whose price is set on three transparent exchanges; whose 2025 regional gap has largely closed; whose government holds no stockpile at all — USGS's copper chapter records "Government Stockpile: None" — and whose most consequential state actions are two producer countries forcing local smelting. Add the negative treatment charges. The world is short of mines rather than processing, and short of copper in America because two furnaces are cold. Copper is a metal in a genuine supply squeeze that a tariff briefly dressed up as geopolitics.

SECTION 7 Trial two: silver

The United States Treasury held 498 tonnes of silver at the end of 2021. It held 498 tonnes at the end of 2022, 2023, 2024 and 2025. USGS publishes the number every February²¹, sourced to the US Mint, and it has not moved through the largest silver bull market in half a century. That single unchanging figure is the strongest evidence in this brief, and what it establishes is a negative.

Silver is the weakest of the three claims, and every criterion fails it. There is no export-licence regime anywhere in the silver chain. No state is accumulating: Metals Focus records net official-sector sales of 1.5 million ounces in 2025 and forecasts 1.0 million ounces of net sales in 2026. There is no processing chokepoint — Mexico's 24.2% of mine output is the largest national share, refining is dispersed across dozens of good-delivery refiners, and USGS counts 24 refiners in the United States alone. And when the American government had the clearest possible opportunity to act, in Proclamation 11001 of 14 January 2026²⁴, it found a national-security threat in processed critical minerals and imposed no tariff on any of them.

Silver does have a genuine and well-documented liquidity problem, and it is easily mistaken for the other thing. The World Silver Survey 2026⁹ describes 2025 as "the year when silver's deficit finally caught up with it, as lower inventories and metal being pulled out of London or tied up in exchange-traded products (ETPs) created explosive conditions for lease rates and prices." The price started 2025 below $29, peaked at $84 in December, and averaged $40.03 for the year, up 42%. The Silver Institute's 10 February 2026 release³³ then records what happened next: silver "breach[ed] $100 for the first time before declining below $80." A metal that runs from $29 to above $100 and back below $80 inside fourteen months is not being administered by anyone.

The supply-and-demand table settles the causation question.

The photovoltaic line and the ETP line matter more than the headline deficit. PhotovoltaicsSolar cells. Silver goes into them as a paste that carries the current off the cell, and how much paste a cell needs is something manufacturers can engineer down — which is what has been happening., silver's largest single industrial use, is forecast down 19.1% in 2026 — the largest annual fall on record — because cell makers keep thinning and substituting the silver paste. Industrial demand for silver is being engineered away at the pace prices reward. That is the ordinary fate of an expensive input, and it is the opposite of a strategic dependency. Then the ETP line. Investors added 278.1 million ounces in 2025, which is 87.3% of the entire physical drain, and Metals Focus forecasts 30.0 million ounces for 2026. What tightened London was a portfolio decision.

And the metal, having been drawn to New York, went home. COMEX silver stocks ended 2025 at 15,000 tonnes — 482.3 million ounces — and stood at 338.2 million ounces on 1 September 2026, of which 99.3 million ounces were registered. London went the other way: the LBMA's own vault series³⁴ shows 28,213 tonnes at the end of July 2026, up 0.5% on the month, which is 907.06 million ounces and 16.6% above July 2025³⁵. Between the two vaulting systems sits 1,245 million ounces, about 409 days of forecast 2026 demand. There is a great deal of silver. Some of it is spoken for by exchange-traded funds and is not free floatThe part of a stock of metal genuinely available to be bought, borrowed or lent. Metal held against a fund's shares sits in the same vault and counts in the same headline total, but nobody else can draw on it — which is how a vault can look full while borrowers cannot find an ounce. — that qualification is real and it is why lease rates can spike while vaults look full. It is not the same as a state taking metal off the market.

So why is silver in this conversation at all? Because on 7 November 2025 the United States added it to the critical minerals list³⁶, and because a designation reads, at a glance, like an action. The Congressional Research Service is explicit³⁷ that the list is a statutory construct under section 7002 of the Energy Act of 2020 — a trigger for permitting and funding programmes, not a control instrument. Copper, lead, potash, rhenium, silicon and silver joined it on the USGS's updated methodology; boron, metallurgical coal, phosphate rock and uranium were added after interagency comment. Sixty entries now. A list that contains metallurgical coal and phosphate rock is not a list of chokepoints.

SECTION 8 Trial three: rare earths

Ore is common. The solvent-extraction cascade that separates one lanthanideThe row of chemically near-identical elements that makes up most of the rare-earth group. Their similarity is the entire problem: it is what makes them almost impossible to tell apart by ordinary chemistry, and therefore expensive to separate. from the next is not, and China spent thirty years and a great deal of pollution learning to run it at scale.

Rare earths are not rare. The seventeen elements occur together in the same minerals and separate from each other only with difficulty, because their chemistry is nearly identical. The industrial answer is a cascade of hundreds of mixer-settlerThe basic vessel of a separation plant: one tank where the ore solution and the solvent are stirred together, and a second where they are left to settle apart again. Hundreds are plumbed in series, and the plant is essentially that one unit repeated until the elements come out singly. stages in which an organic solvent pulls one element slightly harder than its neighbour, repeated until purity reaches four ninesTrade shorthand for a purity written as a run of nines. It is the grade magnet customers specify, and specifying it is how a buyer keeps a supplier honest without inspecting the plant.. It is process engineering, not geology, and it is embodied in operating know-how more than in patents. That is why Announcement No. 62 controls the technology¹², including "production line assembly, maintenance, repair and upgrading," and treats a joint research project as a transfer. A state that owns a process rather than a deposit protects it by controlling the people and the equipment. The seventeen also split into two groups by atomic weight, and that split does more work in this brief than the chemistry suggests. The light ones are led commercially by neodymium and praseodymium, the pair the magnet industry runs on. The heavy ones include dysprosium, terbium and yttrium. Almost every restriction described here, and every large regional price gap, sits on the heavy side.

The concentration numbers follow from that. China mined 69.2% of the world's 390,000 tonnes of rare-earth oxide in 2025 and, per the IEA²⁰, refines over 90%. The gap between those two numbers is twenty-plus percentage points, and it is the subject of this brief. Nigeria, Thailand, Madagascar and Brazil have all raised mine output; almost none of them separate what they dig. Burma's 22,000 tonnes in 2025, estimated by USGS from Chinese import records, goes across the border to be crackedThe first chemical step after mining: the ore mineral is broken open with acid or heat so the rare earths locked inside it can be dissolved out. Separation comes afterwards, and it is the harder half..

There is no exchange in rare earths, so nothing arbitrages a regional gap away, and the gaps are enormous. The IEA's 2026 outlook reports European gallium and heavy rare-earth prices "around five times higher than Chinese domestic prices" and germanium prices almost three times higher. Freight on a drum of dysprosium oxide is trivial against a 400% price difference. That gap is what a licence queue looks like when it is priced.

The USGS annual series, sourced to Argus, shows NdPr oxide at $92/kg in 2021, $124 in 2022, $75 in 2023, $55 in 2024 and $69 in 2025 — a market in oversupply, which is exactly why Washington had to guarantee $110 to get anyone to build. By 2 September 2026 the China Rare Earth Industry Association's index³⁸ put NdPr at $107.72–110.72/kg, dysprosium at $214.50–220.50/kg and terbium at $1,008.75–1,017.75/kg — with the composite index at 262.8, down from a March 2026 peak of about 310. The floor that was set at roughly double the prevailing price is now approximately at the market. That is a different world for every project financed against it, and Part II works through what it means name by name.

Two USGS chapters published for the first time at this level of detail put the chokepoint at the element level, and they are blunter than any commentary. American net import reliance for heavy rare-earth compounds and metals was 100% in every year from 2021 to 2025.³⁹ Terbium, holmium and lutetium came 100% from China across 2021–24; ytterbium 86%. And on 2025: "at least five companies were developing commercial-scale heavy-rare-earth processing and refining capacity… although none produced sustained commercial-scale quantities." Five companies building, none producing. The yttrium chapter is starker still⁴⁰: "There were no fully commercial facilities in the United States that could separate or refine yttrium," and US imports fell to an estimated 300 tonnes in 2025 from 457 tonnes in 2024 — an independent, Tier 1 corroboration of the customs series CSIS traced.

Dysprosium got cheaper during the control campaign, which cuts against the price narrative. On the Argus annual averages USGS publishes, dysprosium oxide went down in 2025, from $257 to $239 a kilogram, and terbium oxide rose from $812 to $1,010 — well below its 2022 level of $2,050. The IEA's finding that European heavy rare-earth prices run around five times Chinese domestic levels is not inconsistent with that, because the two are different measures: one is an annual free-on-boardPriced at the port the metal leaves from, with the buyer paying freight and insurance from there onward. Two quotes struck on different terms cannot be compared until that difference is stripped out. assessment, the other a spot comparison between two regions during an acute squeeze. Both are true and neither supports "heavy rare-earth prices exploded in 2025." What exploded was the variance between what a buyer in Rotterdam paid and what a buyer in Baotou paid.

The controls are working and eroding at the same time, and the erosion is measurable. The IEA's 2026 outlook notes that rare earths are the one exception to rising concentration — refining concentration showed a "modest decline" between 2023 and 2025, attributed to new US projects and higher Malaysian production. USGS's American numbers show it from the other side: domestic production of separated rare-earth compounds and metals went from 800 tonnes in 2023 to 4,300 tonnes in 2024 to 8,900 tonnes in 2025. Eleven times in two years. The United States now makes about a third of the separated material it consumes, against effectively none three years ago, and its net import reliance for compounds and metals fell from over 90% in 2023 to 53% in 2024 before rising to 67% in 2025 as consumption tripled.

China's leverage produced the prices and the political will that are funding its replacement. Every month the licence desk keeps a Western buyer waiting is a month of subsidy to Lynas, MP, Solvay and Neo. The leverage has a half-life, and Beijing knows it — which is the most plausible reading of why the October 2025 package was withdrawn four weeks after it was issued. Maximum pressure applied continuously would finance the exit. Applied intermittently, with a licence desk that mostly says yes, it keeps the alternative supply chain marginal.

Chinese customs data⁴¹ for January and February 2026 show magnet exports to the European Union at 4,775 tonnes, up 28.4% year on year and 44.4% of China's total, while shipments to the United States fell 22.5% to 994 tonnes — a fourth consecutive monthly decline, leaving America at 9.2% of the total and third in line behind South Korea. Beijing preserved both revenue and leverage by making the restriction selective rather than general. That is the incentive structure resolving itself in customs data, and it is why "China cut off rare earths" was never an accurate description of 2025 or 2026.

SECTION 9 Defence demand, sized honestly

An F-35 carries more than 900 pounds of rare-earth material. A Virginia-class submarine carries about 9,200 pounds. Those numbers are quoted constantly and they measure the wrong thing. The relevant quantity is not what a weapon contains; it is what a law compels.

10 U.S.C. § 4872⁴² bars the Secretary of Defense from procuring covered materials melted or produced in China, Russia, Iran or North Korea, or end items containing them. Samarium-cobalt and neodymium-iron-boron magnets are covered materials. Under DFARS 225.7018-2⁴³, the restriction through 31 December 2026 reaches melting and every subsequent phase — powder formation, pressing, sinteringPressing magnet powder into a mould and heating it until the grains fuse into a solid block, without ever melting it. It is one of the two ways a permanent magnet is made; the other is to bind the powder in a polymer, which the same rule also names. or bonding, magnetisation. From 1 January 2027 it reaches the "entire supply chain from mining of neodymium, iron, and boron through production of finished magnets." Germanium and gallium join the covered list on 18 December 2027. Waivers exist — the Secretary may waive where satisfactory domestic material cannot be procured at reasonable prices within 36 months, or where national security requires it, with notice to the congressional defence committees.

That is the demand that matters, and it is a legal quantity rather than a tonnage. The tonnage itself is small.

The IEA calculates⁴⁴ that the April 2025 controls put "an estimated USD 6.5 trillion per year of downstream production outside China at risk." Three hundred tonnes of oxide against six and a half trillion dollars of exposure. The Pentagon will not fill a 10,000-tonne magnet plant; carmakers, wind developers, robot builders and data-centre cooling will. What defence supplies is a statutory floor under a domestic buyer that cannot legally defect to the cheapest source. Every Western magnet project is being underwritten by a procurement rule and priced against a commercial market that will decide whether it survives.

SECTION 10 The stress precedents, and what they actually indict

On 8 March 2022 the London Metal Exchange suspended nickel trading and cancelled eight hours of trades. The exchange commissioned Oliver Wyman and NERA to find out why. The LME's own response of 10 January 2023⁴⁵ lists the contributing factors the review identified, and they are not the ones the popular account remembers.

The review found large, exposed short positions whose risk was hidden because they were "fragmentation of positions across counterparties and between on-exchange and OTC," so that "position limits and accountability levels did not prevent the build-up." It found that liquidity withdrew, that "the LME's price volatility controls did not control price volatility during the events," and — the line that matters most here — that "basis risks emerged from the use of LME nickel prices to hedge sales" of material the contract does not deliver. The remedy the LME flagged was structural: its Nickel Committee would "continue to explore whether any changes to the LME Nickel contract, additional Class 2 contracts, and/or alternative Class 2 pricing mechanisms may be beneficial to the market."

Indonesia mined 2.6 million of the world's 3.9 million tonnes of nickel in 2025⁴⁶ — 66.7% — almost all of it Class 2The lower-grade nickel products — nickel pig iron and matte — that the LME's contract will not accept for delivery. Class 1 is the high-purity refined metal it will. A benchmark that settles only in Class 1 is pricing a slice of the nickel market, not the market. material, nickel pig iron and matte, that the LME contract cannot deliver. The same chapter records the market in surplus every year since 2022, with an estimated 189,000 tonnes of surplus in the first nine months of 2025 alone, and the annual average LME cash price down 11% in 2025. A benchmark can break upward in a market that is structurally long, because the deliverable pool and the market are different things. Producers hedged it with the Class 1 contract anyway, because that was the only liquid instrument. When the deliverable pool got small relative to the positions written against it, the price of the deliverable pool stopped being a price for nickel and became a price for warrants. The failure was in the link between the benchmark and the physical market it claimed to represent.

Copper and silver show the same problem. Copper: roughly 694,700 tonnes sit in COMEX warehouses in a country that has exempted refined cathode from tariff but has told the world it may impose 15% from January 2027, while the LME's open tonnage is 107,050 tonnes — about 1.3 days of world refinery output. The global market is not short. The deliverable pool outside the United States is thin, and that is what LME prices are now discovering. Silver: 338.2 million ounces sit in COMEX warehouses, of which 99.3 million are registered; London holds 907.06 million ounces, an unknown but large fraction of it allocated to exchange-traded products and therefore not lendable. Lease rates spiked in October 2025 and again into January 2026 not because silver was gone but because the borrowable slice of it was.

The hoarding narrative has this backwards. The paper-versus-physical divergences of the last two years are better evidence about benchmarks than about states. Nobody confiscated the nickel. Nobody has confiscated the copper or the silver. What has happened, repeatedly, is that the metal eligible to settle the contract stopped being representative of the metal the world actually trades — because of a tariff expectation, a fund allocation or a change in ore chemistry. Prices then did what prices do when the deliverable basis is thin. This is a design problem in the pricing layer, and the LME has said so in writing about nickel. It has not been fixed for copper or for silver.

SECTION 11 What the West has funded, and how long it takes

Money is not the constraint. The Office of Strategic Capital received $500 million of credit-subsidy funding in the 2025 reconciliation act, which supports up to roughly $100 billion of lending for critical-minerals production; the Export-Import Bank board approved a single $2.9 billion loan⁴⁷ for one Idaho antimony-gold project on 21 May 2026. Against that, Lynas produced 19 tonnes of dysprosium and terbium in the June 2026 quarter. The gap between those two facts is the whole execution problem.

Two entries in that table test Western build rates against the record. Lynas signed a US Department of Defense contract in 2023 to build a heavy rare-earth processing plant at Seadrift, Texas, bought a 149-acre site, and held a community open house with the DoD in December 2023. The company's own US project page still describes the facility as "proposed"⁴⁸ and carries nothing later than that December. On 16 March 2026 the arrangement was replaced by a purchase agreement⁴⁹ because of "significant uncertainty as to whether the construction of the Heavy Rare Earth processing facility at Seadrift, Texas would proceed." Three years, a signed government contract, a bought site, and no plant.

Lynas then moved that expansion to Malaysia, where it already runs the largest separation complex outside China — and the budget went from about A$180 million to about A$294 million. The company's own explanation, in its quarterly report of 22 July 2026⁵⁰, names the mechanism precisely: "additional equipment to meet customers' specific product purity and physical characteristic requirements, the higher cost of sourcing equipment outside China, and cost escalation due to the current geopolitical environment." That is a 63% cost increase, incurred by the most experienced non-Chinese separator on earth, on a brownfieldAn expansion on land already in industrial use and already permitted, with power, water and people in place. It is the easy case. A greenfield project starts with bare ground and every approval still to win. site, in a jurisdiction where it has operated for a decade.

The binding constraint is equipment made mostly in China, specifications set by magnet customers, permits, and the small number of people anywhere who have commissioned a solvent-extraction cascade. It is not capital. Announcement No. 62, suspended though it currently is, was aimed exactly there: at the line-assembly and maintenance know-how that turns a budget into a working plant, rather than at the ore or the oxide.

Qualification is the step in the sequence below that sets the schedule. To qualify a material is for the customer to certify it, not the supplier: a magnet maker runs the new oxide through its own process, tests the parts that come out, and only then writes that supplier into the specification for a motor it is already building. It is done one customer at a time and it takes years. That is why a plant can be physically running and still have nothing it is allowed to sell.

1. 01
   Fund

   Loan or grant signed. Fast — months. This is the step the headlines cover.
2. 02
   Permit and site

   Land, environmental approvals, utilities. Seadrift got this far in 2023 and stopped.
3. 03
   Procure

   Mixer-settlers, kilns, chlor-alkali. Mostly Chinese-supplied; the cost line that moved Lynas by 63%.
4. 04
   Commission

   Purity, yield, stability. Energy Fuels has been at pilot scale on Dy and Tb through 2026.
5. 05
   Qualify

   Customer sign-off on a magnet spec. Lynas began samarium qualification in the June 2026 quarter; first orders in the following one.

Sequence and durations drawn from the primary filings and quarterly reports cited above; each step is evidenced by at least one named project's disclosed experience.

SECTION 12 The scorecard

Each cell below is answered from the documents in Sections 6 to 8, against the criteria set out at the top of this brief.

The pattern is not a spectrum with rare earths at one end and silver at the other, with copper in the middle. Copper fails on different criteria from rare earths, for different reasons. It has no state buyer, an open exchange price, and a processing concentration that loses money. Its state exposure sits in Kinshasa and Jakarta, at the mine gate, and takes the form of forced local processing rather than denial of supply. Silver fails on all five and the failures are not close.

SECTION 13 The conclusion, and the consensus it revises

The consensus holds that deglobalisation has turned critical metals into weapons, that states are hoarding them, and that the market is splitting into blocs. Two of those three are wrong on the documents, and the one that is right is right for a reason the consensus does not name.

States are not hoarding. The American stockpile plan for magnet block in fiscal 2025 was 450 tonnes, and no plan was published for fiscal 2026. The Treasury's silver has not moved in five years. USGS records no government stockpile for copper at all. What states are doing is buying into plants and writing rules about who may buy the output: $400 million of preferred stock and a ten-year warrant in MP Materials, $882.3 million of stock and warrants taken from USA Rare Earth as a condition of a $277 million grant, $700 million of Office of Strategic Capital loans into a private magnet maker and its recycler, a Japanese state agency taking direct equity in a Namibian heavy rare-earth vehicle and its ministry proposing to let it invest without any Japanese company alongside. Governments have become minority shareholders and secured lenders in the midstream. That is the transition, and it is not reversible by a change in the metal price.

Nor is the market splitting into blocs uniformly. It has split in exactly one place: the separated heavy rare earths, where a licence queue produces a fivefold price gap that no arbitrage can close because there is no exchange to close it on. In copper, the gap that opened in 2025 has largely closed. In silver, the gap that opened twice was a lease-rate spike in a market holding more than a year of demand in two vaulting systems.

The bloc that is forming is defined by law rather than geography: buyers who are legally forbidden from defecting to the cheapest supplier — the Department of War and its contractors from 1 January 2027, European projects taking CRMA strategic statusDesignation under the European Union's Critical Raw Materials Act, the regulation cited earlier in this brief that sets the Union's targets for extracting, processing and recycling its own strategic raw materials. A designated project is one Brussels has selected as counting toward those targets., Japanese firms buying JOGMEC-backed output. Everyone else keeps buying Chinese magnets, and in the first two months of 2026 the European Union bought 4,775 tonnes of them, 28.4% more than a year earlier.

THE BEAR CASE, PUT AS WELL AS IT CAN BE

The opposing view has a strong version. Refining concentration is rising, not falling: the IEA's average top-country refining share went from 70% in 2023 to 72% in 2025, and the top three refiners now hold 86% of key mineral processing against 82% in 2020. The number of mineral tariff codes under Chinese export control has tripled since 2023. The October 2025 package showed that Beijing will assert jurisdiction over a magnet made in Germany from Chinese oxide, and it was withdrawn under negotiation rather than abandoned — the suspension expires on 10 November 2026, eight weeks after this brief's cutoff. Enforcement has gone personal, with foreign nationals detained in Dalian in May 2026 and a Chinese executive placed under compulsory measures in June for mislabelling germanium. The DR Congo extended concentrate export bans to copper on 29 June 2026. And the United States has now adopted the same instrument China uses, ordering that 100% of black mass and tungsten scrap sales go to domestic buyers, with exceptions granted at an official's discretion. On this reading, both blocs are converging on administered allocation, and the copper convergence is a lull.

Most of that survives scrutiny. Two parts do not. The rising-concentration statistic is driven by nickel, graphite and manganese; rare earths are the single exception the IEA identifies, showing a "modest decline" in refining concentration from 2023 to 2025 on new US and Malaysian capacity. And a count of tariff codes measures instruments, not effect: over the same period Chinese magnet exports to the European Union rose 28.4% year on year. The black-mass order covers four Schedule B linesThe codes in the American export-classification system that state exactly which goods a rule catches. A rule written to four codes is a much narrower thing than the same rule described in words — which is why the codes, not the press release, are what to read. for one year, renewable, with case-by-case relief — it is a serious precedent and a small action.

Where the bear case wins outright is on durability. The apparatus is now built on both sides and nothing in the record suggests it will be dismantled. That is exactly why rare earths are graded structural here. The disagreement is about scope, not direction: this is a bifurcation of the heavy rare-earth and magnet chain, not of industrial metals generally.

THE FOUR ARGUMENTS WORTH HAVING

SECTION 14 Verdict by metal, and what follows from it

Rare earths — structural. High confidence A standing licence regime that survived the suspension, an administered domestic price, a technology export ban, a quota that stopped being published, a processing share above 90%, and a fivefold regional price gap. The consequence chain runs three steps and the third one is where the money is. Controls raise ex-China prices; higher ex-China prices plus procurement mandates fund rival separation and magnet capacity; that capacity, once qualified, erodes the chokepoint that created it. The asset that appreciates is therefore the qualified, permitted, operating plant rather than the oxide, and the scarcity that persists longest is of people who can commission one. There is a second-order effect too: because states are withdrawing capacity rather than metal from the price-discovery system, and because China stopped publishing quotas, the price is more headline-sensitive than it was, which makes it a better instrument for policy and a worse guide to physical balance.

Copper — partial, and demand-driven. High confidence Two producer states forbid concentrate exports to force local smelting; nobody stockpiles; the exchange price is open; the regional gap has narrowed by three-quarters; and the processing concentration everyone worries about loses money. What follows is that copper's returns come from the ordinary place — grade, cost position and the price of a metal whose mine supply is not keeping up. The state's contribution is second order. Forced downstreaming in Kinshasa and Jakarta tightens concentrate supply for everyone else, and that is one reason treatment charges are negative. Negative charges transfer margin from smelters to miners. The policy is helping copper miners by accident.

Silver — unsupported. High confidence, in the negative No licence regime, no state buying, no processing chokepoint, no tariff when one was invited, and an investment-driven drain that the same investors can reverse. The consequence for capital is not that silver is uninvestable — it had a spectacular two years — but that the strategic-metal framing supplies no protection when the flows turn. A silver position is a position on investor allocation and on solar and electronics fabrication, both of which are visible monthly. It is not a position on statecraft, and pricing it as though it were is how people get hurt.

SECTION 15 Where the money lands as the rewiring proceeds

The chain runs from the ore to the motor, and at each step the question is who can be replaced. Mining rare earths can be replaced — Nigeria, Thailand, Brazil and Madagascar all raised output in 2025 without needing anyone's permission. Making a magnet can be replaced given oxide, capital and time. Separating dysprosium from terbium to four nines, on schedule, in a plant that customers have qualified, cannot yet be replaced at scale outside China and Malaysia. That is the hierarchy.

Layer 02 is the layer nobody can buy. The equipment that builds a separation plant is a genuine chokepoint with pricing power, it is disclosed as such by the most credible Western operator in the industry, and there is no listed way to own it. When a mandate asks where value concentrates and the honest answer is "in a supplier you cannot buy," saying so is more useful than reaching for a proxy. The same is true of most of the magnet chain: Vulcan Elements is private, Noveon is private, and Vacuumschmelze is being bought by a uranium company rather than trading on its own.

SECTION 16 The scenario ladder

No probabilities are assigned here. Assigning odds to a decision taken by a small number of people in Beijing or Washington is arithmetic dressed as insight. What can be specified is what would come first, what to watch, and what would retire each scenario.

Dates from the instruments and filings cited in Figure 11 and Section 11. Forward events only; nothing here is a forecast of outcome.

SECTION 17 The rare-earth chain, name by name

Every company below has been read from its own most recent filing — for the American names, the 10-Q, the quarterly report an issuer files with the Securities and Exchange Commission: unaudited, but made under legal liability for what it says. Where a name is graded against the criterion this brief uses for durability — a right that is hard to replicate — and appears not to have one, the defence is made explicitly rather than left to the reader.

MP Materials. The 10-Q for the quarter ended 30 June 2026⁵¹ is the most instructive document in this brief. Revenue was $108.5 million, up 89%, with price-protection income of $17.6 million on top. NdPr production reached 840 tonnes, up 41%, and NdPr sales 1,006 tonnes, up 127%. Rare-earth concentrate revenue was zero, against $11.9 million a year earlier, because the company "cease[d] all sales to customers in China" in July 2025. Cash and short-term investments stood at $1.5 billion against $1.0 billion of debt, with 2026 capital spending guided to $500–600 million.

Materials segment adjusted EBITDA was $32.505 million in the quarter, and $17.580 million of that was the Department of War's price-protection payment. Strip the floor out and the segment earned $14.9 million. Across the half, segment adjusted EBITDA of $69.237 million included $59.853 million of floor income, which is 86.4%.

That dependence is the case against MP, and it is also the case for it. The floor runs to 31 December 2035 and pays on stockpiled, affiliated and third-party designated volumes alike; the government also guarantees the 10X plant a minimum of $140 million of EBITDA, subject to annual escalation, and has an obligation to buy every magnet it makes. MP does not have a scarce ore body — Mountain Pass bastnaesiteThe mineral that carries the rare earths at deposits like Mountain Pass. It is a perfectly good ore and there is nothing unusual about having it — which is the point being made here. is good but ordinary — and magnet making is replicable. So the grade needs a defence on this brief's own criterion. The hard-to-replicate right is a permitted, operating, customer-qualified separation and metal chain inside the United States, plus a ten-year contractual price and a guaranteed buyer. It is not the rock. Section 11 shows what happens to companies that try to assemble that chain from scratch. Seadrift had a signed government contract and a bought site in 2023 and does not exist.

One more line from the quarter's earnings release⁵² shows the shape of things to come: MP has signed a long-term agreement to supply gadolinium to a new US aerospace and defence customer. Consolidated Adjusted EBITDA for the quarter was $28.5 million and the net loss was $20.3 million. A company that still loses money at the bottom line is nonetheless selling individual heavy rare earths on long-term contracts to defence buyers, which is precisely the transition the DFARS rule was written to force.

The floor may soon stop paying anything. China's association index put NdPr at $107.72–110.72/kg on 2 September 2026³⁸; MP's implied benchmark in the June quarter was well below that. If the market price settles above $110, floor income goes to zero and, once 10X reaches full capacity, MP begins paying the government 30% of the excess. A rare-earth bull market is, past a point, a transfer from MP's shareholders to the Treasury. One number moved the wrong way: REO production fell 16% year on year in the quarter, to 11,072 tonnes, against an "Upstream 60K" ambition of roughly 60,000 tonnes a year.

Lynas. The quarter to 30 June 2026⁵⁰ produced record gross sales revenue of A$288.9 million, up 70%, on 2,941 tonnes of rare-earth oxide sold at a record average of A$98.2 per kilogram — a 63% price improvement year on year that the company attributes to NdPr pricing, a heavier mix of heavy rare earths, and "increased premiums over the market index." That last phrase is the thesis in a company's own words: Lynas is increasingly selling outside the Chinese benchmark, to customers who pay for the fact that it is not Chinese. Cash stood at A$1,209.1 million. Dysprosium and terbium output was 19 tonnes for the quarter.

The record quarter has problems underneath it. Production is volatile — NdPr went 2,003 tonnes, then 1,404, then 1,996, then 1,857 across the four quarters of FY2026, with power outages at Kalgoorlie, kiln maintenance in Kuantan and a water-recycling plant and ore-quality problems at Mt Weld each taking a turn. And the American project failed. The $96 million four-year offtake with a $110/kg floor that replaced Seadrift is a good outcome for a company that no longer has to build a Texan plant; it is a poor outcome for anyone who counted that plant in the 2027 ex-China supply balance.

Energy Fuels. The 10-Q for the June 2026 quarter⁵³ reports total revenues of $25.1 million, of which $25.0 million was uranium concentrates; the Rare Earth Elements segment recorded no revenue in the quarter, the half, or the comparable prior periods. White Mesa has made NdPr at commercial scale and dysprosium and terbium oxide at pilot scale, and is piloting gadolinium. The strategy is acquisition: Australian Strategic Materials closed on 28 August 2026 for US$243.4 million⁵⁴, bringing the Dubbo project and a Korean metallisation plant; Vacuumschmelze is under a merger agreement signed 23 June 2026 for about $718 million in cash plus 65.85 million shares, implying roughly $1.9 billion of equity value, targeted to close in the first quarter of 2027, backed by a $250 million Goldman Sachs term-loan commitment.

Energy Fuels is graded neutral. If VAC closes, Energy Fuels becomes the only listed Western company owning mine, separation, metallisation and magnet manufacture — a genuinely advantaged position. Until it does, this is a uranium business with a rare-earth pilot plant and a large acquisition to fund. The gap between those two descriptions is the whole investment question, and it resolves on a date.

USA Rare Earth. The June 2026 10-Q⁵⁵ discloses the most striking single transaction in this study. On 3 June 2026 the company signed a $277.0 million Direct Funding Agreement and a $1.30 billion Loan Guarantee Agreement with the Department of Commerce under the CHIPS Act. As conditions precedent it issued the government 16,132,790 shares valued at $451.4 million and a warrant over 17,600,584 shares, exercisable at $17.17, initially valued at $430.9 million. That is $882.3 million of securities — 3.19 times the grant — handed over before a dollar was drawn, and the filing states that as of 30 June 2026 the company "had not received any direct funding award disbursements, nor had it received formal approval for any disbursement requests."

The underlying business is early. Half-year revenue was $11.5 million, almost all of it from Less Common Metals in Europe; the net loss was $80.0 million; the Stillwater magnet plant "has recently been commissioned and has commenced commercial production; however, we have not begun generating revenue from sintered neodymium-iron-boron magnets"; Round Top is an exploration-stage property whose accelerated mining plan anticipates commercial production in late 2028⁵⁶. Against that sits $1.53 billion of unrestricted cash, raised largely in a $1.50 billion January 2026 equity issue. The company is being valued for a chain it has not yet sold anything into, on terms under which the state now owns a large call on the upside.

Ramaco Resources. Its June 2026 10-Q⁵⁷ is unambiguous: "The Company has not recorded any revenues from the Rare Earths and Critical Minerals segment," and the Brook Mine "is currently an exploration stage property" whose estimates are inferred mineral resources with "no certainty that any part of the inferred mineral resources estimated will be converted into higher confidence mineral resources and eventually mineral reserves." Consolidated half-year revenue of $266.4 million came from metallurgical coal and was down from $287.6 million. The filing also discloses a securities class action alleging materially false or misleading statements concerning "the development and active mining status" of the Brook Mine project over a class period from 31 July 2025 to 23 October 2025 — allegations, not findings, and unresolved. An amended annual report on Form 10-K/A was filed on 24 July 2026.

SECTION 18 The copper names

The copper names are graded on cost and jurisdiction because there is no policy moat to grade them on. That is itself the finding. The cost measures below are not like-for-like: Freeport reports site production and delivery costs before by-products, Southern Copper reports operating cash cost before by-product revenues, Teck reports net cash unit cost after by-product creditsSubtracting the revenue a mine earns from the other metals it produces alongside the main one from the cost of the main one. It can move a stated cost per pound a long way without anything at all changing underground, which is why costs reported on this basis cannot be set beside costs that are not., and Ero reports C1A widely used measure of the cash cost of producing a pound of metal at the mine, before financing, tax and capital spending. No authority defines it, so two companies' C1 figures may not be measuring quite the same thing.. They are shown with their definitions attached and should not be ranked against each other.

Southern Copper is the clearest illustration of what a copper boom does when policy is not in the way. Its second-quarter release⁵⁸ reports record net income of $1,670.0 million, up 71.6%, on a 38.9% net margin, with half-year mined copper of 461,206 tonnes down 3.8% on lower Peruvian grades — and silver by-product sales up 137.9%. Capital spending rose 79.4% to $422.8 million in the quarter. There is no floor price here, no offtake, no licence: just grade, a by-product windfall and a permitting problem at Tia Maria that eight lawsuits have not resolved.

Ero Copper is the same story at one-twentieth the size, and its management discussion for the quarter⁵⁹ shows the leverage: net income attributable to owners of $89.5 million, adjusted net income of $87.4 million, and available liquidity up $35.5 million to $181.7 million. A single-country Brazilian producer with no policy exposure earned more in a quarter than several of the graded rare-earth names have earned in their lives.

Teck is not really a copper call any longer: its 1 September 2026 release⁶⁰ confirms that the merger of equals with Anglo American, under an arrangement agreement dated 9 September 2025, will become effective eleven trading days after the remaining conditions are met, with Anglo declaring a special dividend of approximately US$4.5 billion. Anyone holding Teck for copper is about to hold Anglo Teck for a diversified portfolio. And Rio's Kennecott is the sharpest illustration in this brief of the American refining problem: refined metal output fell 49% year on year in the second quarter³¹, partly on "reduced availability of high quality copper concentrate" — in a country that has just tariffed semi-finished copper to protect a smelting industry it barely has.

SECTION 19 The silver question, and the one name that answers it

Hecla Mining's June 2026 10-Q⁶¹ reports metal sales of $323.0 million for the quarter, of which silver was $213.7 million — 66.2% — against $122.5 million of silver a year earlier. Greens Creek, Lucky Friday and Keno Hill are in Alaska, Idaho and the Yukon. Half-year metal sales were $729.8 million against $410.6 million. It is a fine business having a very good year because the silver price nearly doubled.

Hecla's leverage runs to the silver price rather than to a policy moat, and listing silver as a critical mineral does not create one. There is no floor price, no offtake, no export licence anyone needs, and no procurement rule that compels a buyer to prefer American silver. Hecla is graded neutral on this brief's criterion for the plainest of reasons: its returns are a function of the silver price and its own grades, and the silver price is a function of what investors decide to allocate. That is a real business and an honest description of it. Anyone buying silver equities for exposure to resource nationalism has bought a betaExposure that simply moves with the underlying price. Owning a silver miner for the silver price is beta: nothing company-specific is doing the work, and nothing company-specific will protect you when the price turns. to the metal and told themselves a story about the state.

SECTION 20 The reversal case, run against every name

Suppose Beijing extends the suspension in November 2026, broadens general licences through 2027, and lets Chinese oxide and magnets flow freely. Ex-China premiums compress, NdPr falls back toward the fifties, and the entire policy-floor edifice is revealed as a subsidy to businesses that cannot stand alone. That is the bear case for this whole trade, and it deserves to be run against each name individually rather than waved at.

MP Materials is hedged the wrong way round. It is protected by a falling NdPr price, not hurt by it: the price-protection agreement pays the shortfall between $110 and the benchmark on designated volumes through December 2035, so a collapse to $55 would roughly double the floor income per kilogram. The Department of War's obligation to buy every 10X magnet, with its $140 million minimum EBITDA guarantee, does not flex with the market. MP's genuine exposure runs the other way — a price sustained above $110 ends the floor income and, once 10X is at full capacity, hands 30% of the excess to the government. Energy Fuels is hedged differently and more crudely: it is a uranium producer, and its rare-earth segment currently earns nothing, so a rare-earth reversal costs it the acquisition case rather than the cash flow.

Lynas is the most exposed of the advantaged names. Its record A$98.2 per kilogram average selling price rests on "increased premiums over the market index" and a heavier heavy-rare-earth mix. Both are functions of scarcity outside China. Only about $96 million over four years is protected by a floor. A relaxation scenario compresses Lynas's realised price toward the index it has spent two years escaping, at the same time as its Malaysian heavy circuit, already 63% over its original budget, is spending. USA Rare Earth is the most exposed of all: no disclosed floor, no disclosed offtake, no magnet revenue, and a valuation that has to be carried by the policy environment because it is not yet carried by sales. Ramaco is unaffected in either direction, because a segment with no revenue cannot lose any.

The copper names are almost untouched by a rare-earth reversal, which is itself confirmation of the three-trials structure. Their reversal risk is a different event: no refined-copper tariff on 1 January 2027, which is this brief's base case anyway, and which the price has already largely discounted — the COMEX premium is down to eleven cents. Hecla is untouched by everything in the policy record and moves with the silver price.

SECTION 21 Alternatives considered and rejected

SECTION 22 Predictions, dated and testable

Each of these can be checked against a published document on a stated date. None is a price forecast.

APPARATUS Evidence register

TIER 1 Statutes, regulators and official announcements

TIER 1 Issuer filings and company primary documents

TIER 2 Agency analysis, price reporting and specialist reporting

TIER 3 Republished or single-source data — flagged wherever used

APPARATUS Not verified, and carrying no conclusion

• CME Group's own COMEX warehouse file. Direct retrieval of cmegroup.com/delivery_reports/Copper_Stocks.xls returned an explicit anti-scraping block. The COMEX figures used here are a third-party republication of that report and are labelled Tier 3 in Figure 9. Anyone re-checking this brief should pull the exchange file directly.
• China's 2025 and 2026 rare-earth mining and separation quotas. Not published. USGS reports the same 270,000 t estimate for Chinese mine production in both 2024 and 2025, which is what an estimate looks like when the disclosure has stopped. No claim here depends on a quota number.
• The FY2026 National Defense Stockpile Annual Materials Plan. USGS states plainly that it "was not released." The fiscal-2025 potential acquisitions are used instead and are labelled as an annual plan, not executed purchases.
• The December 2025 general export licences. USGS confirms that China "began to issue general export licenses to selected exporters." The reported identities of the first recipients were not verified against a primary document and are not used.
• Chinese magnet export volumes after February 2026. The January–February 2026 figures used here are South China Morning Post's reporting of China General Administration of Customs data, corroborated in direction by CSIS. Monthly figures for March 2026 onward circulating in trade press were not verified against a customs primary and are not used. Claims of record first-half 2026 magnet exports appear widely in trade coverage; no customs primary supporting them was obtained, so no such claim is made here.
• The DR Congo joint ministerial order of 29 June 2026. The order itself was not retrieved. Its existence and scope rest on trade reporting and are used only in the timeline and in a scenario trigger, never as the basis for a grade.
• Indonesia's May 2026 royalty pause. Reported, not verified against a ministerial instrument. It appears in Figure 8 as a cyclical measure and supports no verdict.
• SHFE-to-LME price comparison. Deliberately omitted. Shanghai prices include value-added tax and import duty, so a raw spread against LME overstates the regional gap. No SHFE–LME premium is quoted anywhere in this brief.
• Lynas's most recent full-year statutory accounts. The quarterly reports were read; the FY2026 annual report was not retrieved. Nothing here relies on audited Lynas figures beyond the quarterly disclosures cited.
• Any assessment of whether the allegations in the Ramaco securities class action have merit. The complaint is disclosed in the company's own filing and is described here as an allegation. It is unresolved and no view is expressed.

APPARATUS Scope and limits

This is an internal structural assessment of market structure and competitive position. It is not investment advice, not a recommendation, and not a solicitation. It contains no price targets, no valuation, no position sizing and no buy, sell or hold language; the verdicts — advantaged, neutral, exposed — describe structural position against the criterion set out in the thesis and nothing else.

Three limits matter to the reader. First, the evidence cutoff is 4 September 2026, and this field reprices on announcements: the single most consequential date in the brief, 10 November 2026, falls after it. Second, the company grades rest on filings through the June 2026 quarter and, for Lynas, the June 2026 quarter report; any material development after those dates is not reflected. Third, the central quantitative comparisons — mine share against processing share, exchange stock against consumption, floor income against segment earnings — are drawn from sources with different scopes and periods, and each figure states its own basis. Where two numbers could not honestly be compared, they are not compared.

AIR Research. Evidence cutoff 4 September 2026. All figures computed from the cited sources; derivations are shown in each figure's Method field.