Executive summary (20-second highlights):

Something nobody argues about is that for the past two years the critical minerals’ working assumption has been that material inside the allied perimeter is qualified and material outside it is exposed, and every instrument built since 2024 encodes that shape.

Price floors reward suppliers who are not Chinese. Project Vault buys for a reserve the same logic defines. The current apparatus sorts the world into two ‘boxes’ and then finances the near one.

What it never asked is whether the near ‘box’ wants to sell. That question had no answer because nobody had needed one inside the ‘allies’, at least until now.

Ontario’s Premier Doug Ford told the Associated Press that he would begin cutting off electricity and Ontario-produced critical minerals if the U.S.-Canada trade fight escalated further, and he named two critical minerals, which is the part worth holding onto: high-grade nickel, and uranium. Everything, he said, is on the table.

A bit of background:

Section 338 is the detail a trade lawyer notices and a commodities desk often does not.

It is a discrimination provision from the interwar tariff regime, unused for most of a century, and it sits outside the modern architecture: it applies whether or not the goods comply with the continental trade agreement, and it carries no expiry date. The choice of statute is a disclosure.

Now put Ontario's two nouns against the map. Canada supplied C$28.8bn of critical minerals to the United States in 2025, roughly 57% of its total critical minerals exports, per Natural Resources Canada. Sudbury, where Vale and Glencore both operate, is one of the largest nickel-producing districts in the world, and nickel sits inside jet engines, armour plate and missile casings long before it reaches a battery. Cameco's Blind River facility in northern Ontario is described as the largest commercial uranium refinery on the planet.

Note which node each sits on, because the exposure levels are not the same.

Sudbury is an orebody, and an orebody has substitutes given enough (long) time and price. Blind River is a conversion step, and conversion steps do not: the refinery turns concentrates from several countries into uranium trioxide before they reach enrichment, and the United States has nothing comparable of its own (yet).

An eventual nickel restriction would be a cost event, where a refining restriction would be a ‘queue event’, and queues in the fuel cycle are measured in years.

Ford also has form. He surcharged Ontario electricity into three U.S. states in March 2025 and has argued since October 2025 that the Canadian constitution gives provinces jurisdiction over critical minerals. Nobody has tested that proposition against a uranium concentrate shipment under contract, which is why it is unpriced.

Back to the original point, the credential regime never asked who governed the resource/project, because allies were assumed to be allies. What we are seeing is that in the current scenario, this assumption can be challenged as easily as any other.

The premium paid for allied supply has been treated as compensation for one risk: over-dependency on a single supply. It was never priced against the risk that Canada, Australia or Zimbabwe withholds, because those governments were the solution rather than the variable. 

The more consequential fact is what Canada did on the following morning, because it is not what the Ontario Province threatened.

Canada’s Finance Minister Francois-Philippe Champagne published the countermeasure, tariffs of 15%, 25% and 50% on more than 700 US product lines covering C$27.6bn of imports, each rate matched to the corresponding US Section 338 or Section 232 rate, effective 12:01 a.m. on September 8. The targets are steel and aluminium, dairy, appliances, agricultural equipment, pulp and paper, plastics and electronics, and the tariff-item list was updated on August 26. A C$7.5bn support package for workers travels alongside it.

There is no minerals line in it. Not one item in Canada's published retaliation touches either of the two flows Ontario named the day before, and the omission is not an oversight in a 700-line schedule.

That gap is the analytically interesting part, because an import tariff and an export restriction are completely different machines with different victims.

A 50% duty on American steel entering Canada is paid by Canadian buyers. An export restriction on Ontario nickel or Blind River uranium would be paid by American ones. Canada reached for the instrument that landed on its own importers and left the other untouched.

Two things survive that:

Either way, the allied premium now carries a variable that was not there in July.

Focus on the ‘counterparty’ and ‘decline’; that is the unseen event that most are still missing.

Ivanhoe Electric announced a Preliminary Project Letter from the Export-Import Bank (EXIM) of the United States for potential debt financing of US$1.1bn for the Santa Cruz copper project in Arizona, up from US$825M in EXIM's April 2025 letter of interest.

The company states plainly that it is not a binding commitment and that board consideration is expected in spring 2027. The 2025 study put initial capital at US$1.24bn for a 23-year life producing 1.4 Mt of copper cathode by heap leach, on 6,000 acres of privately held surface, mineral and water rights.

Let’s read the last part again. The project's principal de-risking feature is the private land, which removes the federal permitting queue. The federal money is the part that has not arrived. Trump endorsed financing Santa Cruz on August 7, and seventeen days later the bank produced a letter.

The same morning, USA Rare Earth announced completion of the upsized US$1.55bn capitalisation of the vehicle that will buy 100% of Serra Verde's Phase 1 production: US$750M of Department of War equity, US$250M above the amount originally contemplated, plus a forward purchase contract for not less than US$300M over five years. The third leg, a senior secured revolving facility of up to US$500M, arrived as a commitment letter from an unnamed institutional bank.

We covered the mechanics of that third leg last week: on August 21, the parties amended Clause 2.2(b) of the April offtake, redefining the required government support as the support the government confirms it has provided. The stockholder vote was held on Friday, August 28. 

Set the week's instruments on a ladder rather than a light switch, because the market is paying for the switch.

Five instruments across four days, and the only line where cash has moved and the counterparty is not a state sits at the bottom. It is a private round. The DRC rail concession (we will cover it in a bit) adds a sixth, and it is still a letter of interest.

This matters because of what the credential is worth. We named the mechanism in CW30: Section 6 of EO 14415 carves out Project Vault and any foreign project financed, guaranteed or insured by EXIM or the DFC, together with any company or project receiving grants, financing, loans or equity from State, War, Commerce or Energy. Money is the passport. A project holding one of those instruments sits inside the compliant perimeter regardless of where its material comes from, which is why the market began paying for the instrument rather than for the orebody. 

Which is where the ladder matters, and where we would rather be precise than dramatic. Section 6 names financing, guarantees and insurance. A letter stating in its own text that it is not a commitment to finance is none of the three, and we are not in a position to say it qualifies its holder today. The defensible claim is the narrow one: rung two and rung four are being paid for as though they were the same document.

The Democratic Republic of Congo (DRC) President Félix Tshisekedi and Angolan President João Lourenço signed a 30-year concession with Mota-Engil for the Dilolo to Sakania railway, a 1,004.5 km linking Kolwezi, Tenke and Lubumbashi to the Angolan border and onward to the Atlantic at Lobito. The convention is valued at about US$1.258bn, with a 10% state stake and a 7.5% gross-revenue royalty.

This is the piece of the Lobito Corridor the U.S. has been trying to close for two years. The DFC issued a letter of interest of up to US$1bn for the Congolese section in December 2025, on top of a US$553M loan signed for the Angolan leg. Total U.S. commitment across the corridor has been put at roughly US$4bn, and the stated purpose throughout has been to give Congolese and Zambian copper and cobalt a western route that avoids Chinese-financed logistics.

Mota-Engil has had China Communications Construction Company, a Chinese state-owned builder, as a large minority shareholder since 2021, reported at close to a third of the company with board representation. 

Pay careful attention to that: the corridor built to route around Chinese logistics will be run for thirty years by a company with a Chinese state enterprise on its share register. Section 6 asks who financed the project; it does not ask who owns the sponsor.

The practical exposure is narrower and more interesting. One operator now runs track on both sides of the Angola-Congo frontier, which concentrates pricing power over Congolese export logistics in a single company, and DRC has not said how that will be regulated. 

If you have exposure to Congolese copper or cobalt, the freight rate has become a bilateral negotiation with a counterparty that has its own view on where the cargo should go. Not a fair game.

Since CW29 we have treated the price floor as an instrument rather than a subsidy: a mechanism deciding which shareholder captures the premium. Until Wednesday it was a policy position. Lynas Rare Earths then published audited full-year numbers, and the instrument is now a line in a set of accounts.

For the year to June 30, 2026, Lynas reported revenue of A$977.9M against A$556.5M, up 76%, EBITDA of A$386.0M, up 282%, and net profit after tax of A$222.4M against A$8.0M, on ready-for-sale oxide output up 25% to 13,089 t and NdPr sales up 12% to 7,337 t. Closing cash stands at A$1,209.1M.

Take the EBITDA line rather than the profit line. Net profit rose about twenty-seven times, but it rose off A$8.0M, which is a rounding error on a company this size, so the multiple is arithmetic rather than magnitude.

The number to sit with is the realised price. Lynas achieved an average selling price of A$80.7/kg REO across all rare earth products, with the June quarter at A$98.2/kg, which it attributes to three things: 

That is a producer stating, in an audited disclosure, that off-index pricing contributed to a record realised price. It is not a disclosure that isolates how much. The accounts give no split between the three drivers, and anyone attributing the year to the floors is supplying a number the company did not.

The company supplied a different one, and it is worth reading twice. The same results put the average China domestic NdPr price at US$55.0/kg in June 2025 and US$100.8/kg in June 2026. The benchmark the floor was written to replace roughly doubled underneath it.

Which changes what the instrument is worth: if a floor is an option, and an option is worth the distance between the strike and the market.

The US$110/kg NdPr floor was written against a market near US$55/kg. It now sits a single-digit percentage above the June 2026 China print.

Now the other side of the same accounts.

The Malaysian heavy rare earth expansion carries a capital cost of A$294M against A$180M guided, a 63% overrun the company ties to equipment sourced outside China. We reported the shock in CW30 when it was guidance. It has now cleared an audit, in the same statement as the record realised price.

So the instrument is measurable in both directions. It lifts the price a Western producer receives, and it inflates the capital cost of the circuit that receives it, and both trace to the same export-control regime.

Anyone modelling an ex-China separation project off floor pricing alone is capturing half of a two-sided instrument, and modelling it off a floor the market has already reached is capturing less than that.

The market sold the stock on operations: a consensus miss, ore quality at Mt Weld, shares down on the day. The disclosure that matters in twelve months is the pricing note.

Copper printed records in three of four sessions last week.

In the house unit, COMEX September reached US$14,942/t (US$6.7775/lb) on Wednesday before retreating just over 3% to US$14,473/t (US$6.5645/lb) on Thursday. In London, the official cash price reached US$14,525/t at Wednesday's close, up US$355 on the week.

Underneath, the warehouse story ran both ways inside five days: stock arrived, the premium for immediate metal collapsed, then 51,400 t were booked out again and the premium rebuilt.

CRU, which projected a 639,000 t surplus for 2026, now describes the market as ‘at best balanced’.

In CW34 we broke down the August spread collapse as a position-roll unwind rather than scarcity resolving, and set the falsifier: on-warrant stock resuming a sustained drawdown while the COMEX to LME spread holds above US$400/t would flip the read toward physical tightness.

Curiously enough, both appeared within four sessions.

We are not retiring the thesis yet: one week of cancellations through a single prompt date is not a sustained drawdown, and the spread compressed again on Thursday. September will be an important month to settle the case and analysis.

What tempered it was demand. Chinese industrial profits rose 11.2% year on year in July, down from 15.1% in June and the weakest this year, per the National Bureau of Statistics. A record set on tariff geography met a demand print that does not support it, and the price gave back 3%.

Lithium did the same thing on a different unmade decision.

The most active carbonate contract on the Guangzhou Futures Exchange closed at 152,500 yuan/t on Wednesday, up 11% from July's lows and 29% in 2026. BMI raised its 2026 average forecast for Chinese carbonate to US$20,100/t against a 2025 average of US$10,502, while saying it views current levels as beyond what fundamentals alone justify. Spot has averaged US$22,941/t so far this year, so the upgraded number implies a trough of US$14,500 in 2028.

We are highlighting this section as it is the most consequential.

CW29 put the support instrument, rather than the mineral, at the centre of macro exposure: two portfolios holding the same mineral through different mechanisms are not holding the same risk. This week supplies the inverse face of that, and it is the half that has portfolio arithmetic attached.

The ex-China premium has been bought as a jurisdiction premium.

Own the tonne outside China and the political risk goes away, which is why an Australian oxide, a Brazilian carbonate and a Canadian nickel cathode have been treated as the same trade with different postcodes.

What a holder is exposed to is whichever government's instrument sits between him and the tonne, and as you just read, this week those governments started pointing instruments at each other.

Diversifying across four minerals that all clear through the same instrument is not diversification, because the instrument is the correlated object.

A book holding a U.S.-floored NdPr producer, a U.S.-credentialed copper developer and a U.S.-backed African corridor holds one position in three costumes, and it settles on the U.S.' calendar.

A book holding one mineral through a commercial floor, a multilateral facility, a private round and a sovereign offtake is more diversified than it looks, because those four fail on different news/events.

The market appears to treat allied supply as jurisdictionally safe, federal sponsorship as a funded guarantee, and the copper build inside American warehouses as physical scarcity. Each is priced in the obvious direction, and each has support in this week's data.

What this framing may be underweighting:

Where each of those would show up in a model, and what settles it:

What would change this read, each item dated and checkable:

Cyclic Materials closed a US$75M growth financing led by accounts advised by T. Rowe Price, taking total equity to US$237M, to break ground in Q4 2026 on a rare earth recycling campus at McBee, South Carolina, sized for 2,000 t/yr of feedstock and 600 t/yr of recycled mixed oxides from 2028, on a ten-year exclusive over 100% of VACUUMSCHMELZE's magnet by-products. Private capital is building the American scrap collection network in the week the U.S. is deciding whether scrap may leave the country, and the company doing it is headquartered in Toronto with the Canada Growth Fund on its register. Comments on extending the BIS scrap rule to magnets and swarf close on November 4, and that document settles whether this build is protected or constrained.

POSCO and IDB Invest approved a facility of up to US$700M for POSCO Argentina on August 4, announced August 28, funding working capital for the first brine lithium plant and a second due by year-end, with Sal de Oro sized at 48,000 t/yr. This is a fourth instrument type alongside the price floor, the stockpile and the federal credential: multilateral working capital qualifying a borrower by environmental and social standard rather than by nationality or funding source, which opens a nationality-blind route into the same chain. Watch whether the second plant commissions on the stated year-end date.

Atomic Eagle and Niger announced a Mining Convention with the military government restoring the Madaouela uranium project, a revised permit split 60% to the company and 40% to the state, with the ICSID arbitration filed by predecessor GoviEx in December 2024 to be withdrawn within a week of signing. Madaouela's 2022 study showed a 19-year life and 50.8 Mlb of U3O8. Five days after the settlement was framed as the precedent that lets lenders underwrite Niger country risk, mutinous soldiers attacked Base 101 and the presidential palace in Niamey, and the government called in Russia's Africa Corps to retake the base. The convention priced the legal risk. It did not price the garrison. Whether Global Atomic's US$295M Dasa debt package reaches a credit committee date is the only way to tell which event the lenders believed.

Lithium Argentina and Ganfeng signed definitive agreements finalising the PPG joint venture in Salta, with Ganfeng investing US$180M through a six-year convertible note for 67% equity against Lithium Argentina's 33%, targeting 150,000 t/yr of lithium carbonate equivalent and helping restructure a US$259M convertible due January 2027. Three days later, Chile, Argentina, Bolivia and Peru signed a joint declaration on strategic minerals. One of those four watched control of a 150,000 t/yr platform pass to a single Chinese buyer in the same week it signed. The marker is whether the declaration produces any screening mechanism, because a producer bloc without one is a communique.

Sinomine secured an additional 300,000 t export quota for Zimbabwean lithium concentrate on top of an earlier 200,000 t, against a raw-export ban due on January 1, 2027 and roughly US$500M of committed processing investment. This is a dated producer-country export cliff resolving one quota at a time, in favour of the buyer the ban was written against. Harare has issued half a million tonnes of exemptions to the counterparty funding its processing. A third quota before December would settle whether the January date is a rule or an opening position.

The auditor found a US$140bn number Congress has not been shown: The Government Accountability Office reported that the National Nuclear Security Administration has not fully reported to Congress its estimated cost for uranium enrichment for national security purposes, roughly US$36.8bn through 2052 and about US$140bn through 2105 in inflation-adjusted terms, omitted from the plan submitted in August 2024. Existing inventory is anticipated to last only into the early 2040s. Defence enrichment and the commercial SWU shortfall of 2028 to 2032 draw on the same centrifuge base, and a cost estimate the legislature has not seen is a schedule it cannot hold.

Congo has started selling traceability rather than having it imposed: Entreprise Generale du Cobalt has filled its cobalt export quotas and is pursuing direct arrangements with battery makers on the basis of traceable artisanal supply. Section 3 of EO 14415 will require an indented bill of materials tracing every component to raw-material origin, with guidance due around mid-January 2027. That requirement has arrived as a product a producer country intends to sell before it has arrived as a compliance cost buyers must absorb.

Wyoming got thirty days: The US Forest Service opened a 30-day comment period on the Bear Lodge rare earth project, an open-pit mine proposed by Rare Element Resources at Bull Hill inside the Black Hills National Forest, under the administration's expedited permitting process. Local officials who support the project objected to the window. Louisiana's 90-day report on permanently expediting permits, which names heavy rare earth extraction explicitly, falls due on November 10. Between them, they answer whether expedited federal process compresses a timeline or relocates the objection.