ABUNDANT AND UNAVAILABLE

Why the constraint on critical materials is never the rock

Article 5 of 8  ·  Series II of III  ·  Published 11 November 2026  ·  Analysis → Forecast → Recommendations

Series II, fifth row. Each article measures one physical layer with the same three questions — concentration, criticality, substitution time — and adds a row to one comparison table. The previous row found a substitution time with no engineering answer. This one finds the same thing for a different reason, and the reason is more general.

1. The Mine That Failed Twice

Mountain Pass, in the California desert, opened in 1952 and supplied most of the world's rare earths through the Cold War. By 2002 it was defunct, because Chinese material was cheaper.

In 2008 a group of investors formed Molycorp to bring it back. Their timing looked excellent: in 2010 China restricted rare earth exports to Japan, prices spiked, and the company's shares went with them. Molycorp raised money, bought processing plants in Arizona and Estonia, and acquired a Canadian materials firm with operations inside China.

By 2013 revenues were in free fall. The company filed for bankruptcy in 2015. In 2017 the mine sold for $20.5 million to a consortium that included a Chinese-owned firm, and for a period the rare earth concentrate dug out of California was shipped to China to be processed.

Nothing about the geology changed across those sixty-five years. The rock was in the same place, in the same quantity, at the same grade. What changed was the price, and the price was set somewhere else.

In 2025 the United States Department of Defense took a different approach to the same asset. Rather than a grant or a loan, it agreed a ten-year price floor of $110 per kilogram for neodymium-praseodymium — close to double the market price at the time — structured as a contract for difference in which the government takes thirty percent of the upside. It put $400 million into preferred stock, took a ten-year warrant for up to fifteen percent of the company, and committed to buy the entire output of an expanded magnet plant for ten years, with a guaranteed minimum of $140 million a year.

That is a different kind of dependency from the four already on the table, and it is easy to misread, because it looks like an ordinary supply problem and behaves like a financial one.

2. What a Refining Constraint Actually Does

The intuitive model of a critical minerals problem is scarcity: a country holds the deposits, everyone else has none, and the map is destiny. Almost none of this category works that way.

Mining and refining are different industries

China accounted for roughly sixty-one percent of mined rare earth supply in 2024 and around ninety-one percent of global refining and processing capacity for the elements that matter. Those two numbers describe two different problems, and only the second is a chokepoint.

The pattern repeats across the category. China holds about ninety-eight percent of low-purity gallium production and produces around ninety percent of mined germanium. In graphite it has over sixty percent of flake and nearly eighty percent of synthetic output, rising to more than ninety percent of the value-added products such as spherical graphite that a battery anode actually requires. The further downstream the processing step, the higher the concentration.

Rare earths are not rare. Gallium is a by-product of aluminium refining and could in principle be recovered wherever bauxite is processed. Graphite is mined in a dozen countries. The concentration is not in the ground. It is in the tanks, the kilns and the solvent-extraction trains — and those were built, deliberately, over three decades, largely because nobody else wanted the cost or the effluent.

Running the instrument

The third cell is the point of this row. In rows one to three the substitution time was an engineering number. In row four it was the time needed for a political decision. Here it is neither: it is a financial endurance test, and the party running the test is the party being substituted.

Why the capacity was allowed to concentrate

No country decided to hand this over. It was a succession of individually reasonable decisions taken over thirty years, each of which traded a low-margin, environmentally difficult industrial step for a cheaper input.

Separation of rare earths is chemically tedious — the elements are nearly identical in behaviour and must be pulled apart across hundreds of sequential stages — and it produces radioactive and acidic waste. It earns thin margins. For a Western firm through the 1990s and 2000s, exiting it looked like discipline rather than surrender: the input got cheaper and someone else absorbed the effluent and the capital.

Which is worth stating plainly, because it determines what reversing it costs. This concentration was bought, over decades, with accepted losses and tolerated pollution. Undoing it means somebody being willing to buy it back the same way.

3. Subtheme One — The Licence Is the Instrument

The export controls of the last three years are usually described as restrictions on supply. That describes their effect imprecisely and their design not at all.

The sequence

Gallium and germanium came under control in July 2023. Graphite licensing followed in October, effective that December. Antimony was added in August 2024, effective the following month. In November 2024 a new dual-use export control regime was announced, and in December outright bans on gallium, germanium and antimony to certain destinations, with graphite tightened further. Tungsten followed in February 2025, and seven rare earth elements in April 2025.

The effects were immediate and visible. Rare earth magnet exports fell 74.3 percent year on year in May 2025, to 1,239 tonnes. By December they had recovered to 5,952 tonnes, while less-processed rare earth exports ran at 4,392 tonnes, still about sixteen percent below the year's average. In early 2026 a further set of dual-use controls was applied, aimed at Japan — which holds roughly fifteen percent of advanced permanent magnet manufacturing and is the only substantial non-Chinese capacity in that step.

What the licence actually collects

Here is the part that is rarely discussed. To obtain a licence, the applicant submits the end use of the material: what it goes into, which product, for which customer, in what quantity, on what schedule.

The licensing regime is not only a valve. It is a disclosure requirement, and the disclosures arrive continuously, from every buyer, in commercially specific detail.

Considered as an intelligence instrument rather than a trade one, this is remarkable. A foreign ministry that wished to map the magnet content of another country's defence and automotive production would find that extremely hard. A licensing authority that requires end-use declarations receives the same map as a condition of doing business, updated with every application, supplied voluntarily and accurately, because an inaccurate declaration risks the licence.

And the map has an operational use beyond knowledge. A regime that knows which firms depend on which element, in what quantity and for what programme, can calibrate a future restriction with precision instead of applying a blunt embargo — hitting one sector while leaving others supplied, which is both more effective and far less likely to unify the response.

It is also why the recovery in export volumes should not be read as relaxation. Volumes returned; the licensing architecture did not go away. A valve that reopens is still a valve, and the information it gathered while closed does not expire.

4. Subtheme Two — One Market, Two Prices

The clearest evidence that this is a refining constraint rather than a mining one is that the same element now trades at two entirely different prices depending on which side of the separation capacity it sits.

In September 2026, dysprosium oxide traded around $212 to $218 per kilogram inside China and up to roughly $3,250 per kilogram in North America — a multiple near fifteen. Terbium oxide ran near $997 inside China against about $7,500 outside it. China's own rare earth price index stood at 262.3 on 21 September, a level that reads as orderly and conceals the split entirely.

A single aggregate index for a market that has separated into two is not a measure of anything. It is the average of a domestic price set by abundant capacity and an external price set by scarcity, and the two now describe different economies.

Why the gap does not close

On a textbook reading, a fifteen-fold premium is an extraordinary invitation to build separation capacity. The premium has persisted anyway, and Mountain Pass explains why.

A separation plant is a three-to-seven-year build with heavy capital cost, difficult permitting and a chemically demanding commissioning period. Its investment case rests on a price, and the price is substantially controlled by an incumbent with roughly ninety percent of installed capacity and a demonstrated willingness to use it. The incumbent does not need to act during construction. It needs only to be able to act afterwards — and everyone financing the plant knows it, which is sufficient to raise the cost of capital to a level the project cannot carry.

The barrier is not technical and it is not geological. It is that the alternative's revenue is set by the party it is intended to replace.

What the price floor is actually for

Seen this way, the Department of Defense structure stops looking like industrial subsidy and starts looking like the correct instrument for the specific problem.

—  A price floor of $110 per kilogram for ten years removes the incumbent's principal weapon. If the market price falls below the floor, the project survives, so a price war no longer works.

—  A contract for difference with thirty percent government upside means the public purse is repaid when prices are high, which makes the commitment politically survivable across a decade and several administrations.

—  A hundred percent offtake of the expanded magnet capacity — 7,000 tonnes a year, against a facility scaling from 3,000 to 10,000 tonnes — removes demand risk as well as price risk, because a new entrant in a concentrated market has no customer base.

—  The $400 million equity stake and a warrant for up to fifteen percent align the state with the outcome rather than merely funding it.

Whether this particular deal succeeds is not the interesting question. What matters is that the structure identifies the constraint correctly. Capital was never the binding problem: Molycorp raised plenty and still failed. Survivable revenue was the binding problem, and this is the first Western intervention that addresses it directly.

It also sets a template that is straightforward to copy, and the copying is where the next several years of this substrate will be decided.

5. What Most Analysis Gets Wrong

Four errors recur, and each changes a decision.

—  Counting mines rather than refineries. A mine outside China does not reduce dependence if the concentrate is shipped to China for separation, which is what Mountain Pass did for years after it reopened. The question is never where the rock is; it is where the solvent-extraction train is.

—  Reading restored volumes as restored access. Magnet exports recovered through late 2025, and the licensing regime that reduced them remains fully in place along with everything it has learned. A valve that has been reopened is a different thing from a valve that was never installed.

—  Quoting a single price for a market that has two. The aggregate index looked stable through 2026 while dysprosium traded at a fifteen-fold premium outside China. Any model, contract or hedge built on the aggregate is pricing a market the buyer cannot access.

—  Treating new projects as capacity. Announced capacity in this category has a long history of not arriving, and the reason is almost never engineering. It is that the project became uneconomic before commissioning. Capacity should be counted when it is operating and contracted, not when it is announced and financed.

6. Base, Stress and Extreme

Three ways the next decade runs, so the forecasts below have a frame.

Base — a second, expensive supply chain

Western capacity is built slowly under government price guarantees, the template spreading from the United States to at least one other major economy. A structurally higher-cost non-Chinese supply chain emerges for defence and selected industrial uses, while consumer-facing manufacturing continues buying at the world price. The two-tier market becomes permanent rather than transitional, and the premium is carried as a security cost. This is the most probable path and it is already visible in outline.

Stress — calibrated denial

The licensing regime is used with the precision the end-use data makes possible: specific firms or programmes are denied while the broader market stays supplied. The response does not unify, because most buyers are unaffected and have no reason to bear a cost on behalf of those who are. Production stops in narrow, high-consequence places — a defence programme, a single vehicle platform — and the wider market registers it as a company problem rather than a systemic one.

Extreme — the alternative fails again

A change of administration or fiscal circumstance weakens the price guarantees before non-Chinese capacity reaches scale, prices fall, and the projects built between 2025 and 2030 follow Molycorp. The material loss is smaller than the credibility loss: a second failure within twenty years would establish that Western states cannot sustain an industrial commitment for the ten to fifteen years this category requires, and the cost of financing any third attempt would price that in permanently.

7. Forecast — One Year, to end-2027

Direction. Ex-China dysprosium or terbium oxide prices remain at least three times the Chinese domestic price at the end of 2027. Probability 0.70. Confidence: Medium-High.

The current multiples are roughly fifteen and seven and a half, so a threshold of three allows for very substantial convergence and still resolves yes. Closing the gap requires operating non-Chinese heavy separation capacity at scale, and heavy rare earth separation is the step where the concentration is most nearly total. Nothing under construction reaches that scale within the horizon.

Second-order effect. A persistent premium of that size makes substitution engineering economic in places it previously was not — reduced-dysprosium magnet designs, different motor topologies, recovery from end-of-life assemblies. The most durable response to this constraint may turn out to be design change rather than supply, and it is driven by the price gap rather than by policy.

What weakens it. The premium partly reflects thin trading in a small ex-China market rather than a stable clearing price. A negotiated settlement that broadens licence approvals could compress it quickly, and reported ex-China prices for illiquid elements are less reliable than the Chinese domestic series.

8. Forecast — Three Years, to 2029

Direction. At least one further G7 government establishes a price floor, contract for difference or guaranteed-offtake mechanism for a domestically refined critical mineral. Probability 0.65. Confidence: Medium.

The American structure is public, legible and directly copyable, and the diagnosis behind it — that the binding constraint is revenue rather than capital — travels to any jurisdiction with the same dependency. The European Union, Japan and Australia each have declared critical minerals strategies and none has yet adopted a price guarantee of this kind. Early 2026 controls aimed specifically at Japan sharpen that country's case considerably.

Second-order effect. Price floors in several jurisdictions at once create a guaranteed-revenue tier that sits outside the market entirely. That is a substantial change in how industrial capacity of this type gets financed, and the mechanism will not stay confined to critical minerals if it is seen to work.

What weakens it. Price floors are fiscally awkward and politically exposed: they pay out precisely when prices are low and the public case is weakest. Governments may prefer stockpiles, grants or joint procurement, all of which are easier to announce and none of which addresses the endurance problem.

9. Forecast — Five Years, to 2031

Direction. China's share of global rare earth separation capacity remains above 75 percent at the end of 2031. Probability 0.70. Confidence: Medium.

From roughly ninety percent today, falling below seventy-five would require non-Chinese separation capacity to roughly triple within five years. Every plant involved is a three-to-seven-year build, most are not yet under construction, and commissioning a separation train is chemically demanding work with a long record of schedule slippage. The forecast is arithmetic rather than pessimism.

Second-order effect. If the share holds above seventy-five percent while Western governments have spent five years and substantial money reducing it, the political conclusion drawn will not be that the effort was too small. It will be that dependency in this category is not reversible by industrial policy — which pushes the response toward design substitution, recycling and stockpiling instead.

What weakens it. Capacity share is measured inconsistently: installed against operating, and light against heavy separation, give materially different answers. A generous count that includes commissioned-but-idle plants could cross the threshold without the underlying dependency changing.

10. Forecast — Ten Years, to 2036

Direction. At least one non-Chinese rare earth separation or magnet facility commissioned between 2025 and 2031 ceases operations or enters insolvency by the end of 2036. Probability 0.60. Confidence: Medium.

This is the Molycorp forecast, and it is deliberately uncomfortable. A wave of new capacity is being built in a market whose price is set elsewhere, under guarantees that in most cases have not yet been legislated with the durability the American one has. Ten years is long enough for a change of government, a fiscal squeeze, or a period of low prices to arrive in at least one jurisdiction. One failure out of many projects is a modest claim; the probability would be considerably higher without the price floors.

Second-order effect. The manner of a failure matters more than the fact of it. A project that fails after its guarantee lapses proves the guarantee was working, which strengthens the instrument. A project that fails while guaranteed proves something worse about the instrument itself — and the two will be reported identically.

What weakens it. Defence offtake is unusually durable, since it rests on requirements rather than on commercial demand, and a facility whose entire output is contracted to a defence ministry is insulated from the mechanism that killed Molycorp. If most new capacity ends up in that category, the forecast should fail.

11. Signals to Watch

Five observable items, each published, each moving before the outcomes above.

—  The spread between Chinese domestic and ex-China prices for dysprosium and terbium, tracked as a ratio rather than a level. The ratio is the measure of the constraint; the level is mostly noise.

—  Monthly magnet and less-processed rare earth export volumes, read against the licensing regime rather than as a demand indicator. Volume recovery with the architecture intact is a different state from volume recovery after repeal.

—  Which destinations and sectors licences are granted to, where that is observable. Selective approval is the calibrated-denial scenario beginning, and it appears in the pattern of approvals long before it appears in aggregate volumes.

—  Whether any further government adopts a price floor or contract for difference, and whether it is legislated or merely announced. The legal durability of the guarantee is the whole variable.

—  Operating rather than announced non-Chinese separation capacity, counted as tonnes actually separated. Announced capacity in this category has historically overstated delivered capacity by a wide margin.

Read together: the first two measure the constraint, the third measures its precision, and the last two measure whether the response is durable. A dependency of this kind eases only when operating capacity rises while the price spread narrows. Either alone can move for reasons that have nothing to do with the underlying position.

12. Recommendations — Individuals

This substrate reaches most people through the cost and availability of finished goods rather than through anything visible, and the honest practical content is modest.

Immediate — 30 days

Recognise that this is a defence and industrial constraint before it is a consumer one. Magnets are a small fraction of the cost of a car or an appliance, so a price shock in this category shows up as a modest increase spread across many products, not as an empty shelf. Reading it as a shortage story will mislead you about both the timing and the shape.

Build — 12 months

If you work in manufacturing, engineering or procurement in any capacity, learn where magnets and specialty materials sit in what your organisation makes. This knowledge is currently held by a small number of people in most firms, it is becoming more valuable, and it is unusually transferable between employers and sectors.

Position — 3 years

Treat the reshoring of processing industries as a real and durable employment trend rather than a political slogan. A ten-year price floor with guaranteed offtake is a materially different commitment from a grant announcement, and the work it creates — chemical processing, separation, magnet manufacture — is skilled, geographically fixed and intended to last a decade by design.

Avoid. Reading export controls in this category as an imminent consumer shortage. Three years of controls have produced price effects and programme delays, and nothing has become unavailable to ordinary buyers.

Why this works. Knowing the mechanism is a price and a licence, not a shortage, tells you what to expect: slightly more expensive goods, some delayed products, and a genuine expansion of industrial employment in a handful of places.

13. Recommendations — Business

For anyone manufacturing physical products, this is the substrate where exposure is most often invisible, because it sits three or four tiers below the supplier you actually deal with.

Immediate — 60 days

Trace your specialty materials to the refining step, not the supplier. Which of your components contain rare earth magnets, gallium, germanium or specialty graphite, and where is that material separated or refined? Most firms can answer the first question and almost none can answer the second, because the relevant step is usually two or three tiers upstream of their direct supplier.

Then establish what your organisation has already disclosed in licence applications, directly or through suppliers. This is a live intelligence question, it is typically held in procurement as a compliance file, and it has never been reviewed as a description of your production that now sits with a foreign authority.

Build — 12 months

Qualify a second source at the refining step rather than at the supplier step. Two suppliers drawing on the same separation plant are one supplier with additional paperwork, which is exactly the structure that made the 2025 controls effective so quickly.

And get design substitution onto the engineering roadmap while the price signal justifies it. Reduced-dysprosium magnet chemistries, alternative motor topologies and recovery from end-of-life assemblies are all live engineering options that a fifteen-fold price premium now funds comfortably. This is the response with the longest payback and it is the one most often deferred, because it competes for engineering time rather than for procurement budget.

Position — 3 years

Assume the two-tier price is structural and price your long-dated contracts accordingly. A contract written on a global price for a material you can only buy at the ex-China price is transferring margin away from whoever wrote it, quietly, for its full term.

And in any exposed category, treat announced capacity as unbuilt until it is operating. Building a procurement plan around a plant that has been financed but not commissioned is the specific error that the last cycle punished, and the projects that failed had financing too.

Avoid. Treating a non-Chinese mine in your supply chain as diversification. Ask where the concentrate is separated. For a long period, material mined in California was refined in China, and buyers recorded it as domestic supply.

Why this works. The two corrections here — tracing to the refining step and qualifying a second source there — address the exposure that the controls of the last three years actually exploited, which is that most firms did not know where their materials were processed.

14. Recommendations — Capital

The structural point is that revenue certainty, not capital availability, determines outcomes in this category, and the market has historically priced the wrong one of the two.

Immediate — this quarter

For holdings exposed to specialty materials, establish which refining step each depends on and whether any long-dated customer contract is priced on a global index the company cannot actually buy at. That gap appears in margins several quarters after the price divergence that caused it, and it is not usually disclosed as a materials exposure.

Build — 12 months

Distinguish sharply between projects with a price guarantee and projects without one. This is the single variable with the strongest historical relationship to survival in this category, and it is a contractual fact rather than a judgement. A project without one is exposed to a price its competitor sets, whatever its ore grade, management or capital structure.

Read the guarantee itself rather than the announcement. Duration, whether it is legislated or administrative, what happens on a change of government, and whether offtake is included alongside price. The American structure runs ten years with equity participation and full offtake; a two-year administrative commitment with no offtake is a different instrument wearing the same name.

Position — 3 years

The durable observation is that this constraint is being addressed by guaranteeing revenue rather than by funding construction, and that instrument — long-dated public price support for strategically necessary but commercially unviable industrial capacity — will not stay confined to critical minerals if it is judged to have worked. Where it spreads next is the question worth holding. This is an observation about a policy mechanism, not a recommendation about any instrument.

Avoid. Pricing this as a commodity cycle. The ex-China premium is not a shortage clearing itself; it is two markets with different supply curves, and it persists for as long as the separation capacity stays where it is.

Why this works. Price floors, offtake terms and capacity commissioning dates are all public, contractual and slow-moving, and they bear on survival far more directly than the resource estimates that receive most of the attention.

15. What Would Change Our Mind

Each forecast carries its own weakening condition. Three developments would undermine the article as a whole.

—  The price gap closes without new capacity. If ex-China prices converge on Chinese domestic prices through licence liberalisation alone, the constraint is a policy choice that can be reversed by negotiation, not the structural feature we describe.

—  A non-Chinese separation project reaches scale without a price guarantee. Our central claim is that survivable revenue is the binding constraint. One unguaranteed project reaching commercial scale and holding it through a price trough would falsify that directly.

—  Design substitution moves faster than supply. If magnet chemistries that materially reduce heavy rare earth content reach volume production within five years, the whole dependency shrinks and the refining chokepoint stops being decisive regardless of who holds it.

Jurisdiction and sourcing, recorded on the fifth row. This is the most concentrated article in Series II on both counts: nearly all the material concerns China as supplier and the United States as respondent, with Japan present mainly as a target. The European, Australian and Indian positions are materially different and are not covered here. On sourcing, the Chinese domestic price series and export volumes are well documented, while ex-China prices for illiquid heavy elements are thin assessments rather than cleared market prices, and the fifteen-fold multiple should be read as an order of magnitude rather than a measurement. Capacity shares differ between sources depending on whether installed or operating capacity is counted. Each figure is labelled accordingly below.

The Series II Table — five rows

What five rows show that four could not. Twice now the substitution time has refused to be a number, and for different reasons: water because the alternative belongs to another user, refined materials because the alternative's revenue belongs to the incumbent. Put together, they describe the same defect in the instrument from two directions. Concentration and criticality are properties of the world. Substitution time, wherever an alternative exists but its viability is controlled by someone else, is a property of a relationship — and no amount of physical data will produce it.

This row also breaks the pattern of the first four in one respect worth marking. It is the only substrate so far where a deliberate intervention has been made against the constraint, at scale, with a mechanism that matches the diagnosis. Whether it works is the subject of the ten-year forecast above. That it was correctly aimed is already clear.

The next row is connectivity.

16. Bottom Line

A mine in the California desert opened in 1952, closed by 2002, reopened in 2008 on a price spike, went bankrupt in 2015, and sold in 2017 for $20.5 million to a consortium including a Chinese-owned firm. The geology never moved. The price did, and the price was set elsewhere.

That is this row in one asset. Rare earths are not rare, gallium is a by-product of aluminium, and graphite is mined in a dozen countries. China holds about sixty-one percent of mined rare earth supply and around ninety-one percent of refining — and only the second number is a chokepoint. The concentration is not in the ground; it is in the separation trains, and it was acquired over thirty years of decisions that each looked like discipline at the time.

Two things follow, and neither is about scarcity.

The export licences are a disclosure system as much as a valve. Every applicant files the end use of the material — what it goes into, for whom, in what quantity. Volumes recovered through late 2025; the architecture and everything it has learned did not go anywhere, and a regime that knows which firm depends on which element can deny with precision instead of embargoing bluntly. That is a considerably more useful instrument, and far less likely to unify a response.

And the market has separated into two. Dysprosium around $215 a kilogram inside China and up to $3,250 outside it; terbium near $997 against roughly $7,500. A fifteen-fold premium ought to summon capacity, and it has not, because a separation plant takes three to seven years to build and its revenue is controlled by the party it exists to replace. Everyone financing one knows this, which is enough.

Which is why the substitution time on this row is not a number. Building is the easy part; surviving is the constraint, and survival is decided by someone else's pricing decision.

It is also why the response that finally addresses it is a ten-year price floor of $110 a kilogram with full offtake and an equity stake, rather than another construction grant. Molycorp had capital. What it did not have was a revenue it could count on. Whether the new structure holds is a question for the 2030s, but it is aimed at the right thing, and that is more than the last attempt could claim.

For anyone with a decision in front of them: trace your materials to the refinery, not the supplier. Find out what your licence applications have already disclosed. And when you assess a project in this category, read the offtake and the price guarantee before the resource estimate, because that is the document that decides whether it is still there in ten years.

 Forecast record

Four forecasts, one per horizon, each with a threshold, a named source that will settle it, and a date fixed before the answer is known.

Correlation. The first and third share a parent cause in Chinese separation dominance and are effectively one claim measured at two horizons. The second and fourth are linked inversely rather than positively: if price guarantees spread, the fourth becomes less likely, not more. That is an unusual structure and it is worth naming, because two forecasts that move against each other cannot both be wrong for the same reason, which is a mild improvement in the information the set carries. So the honest count is two families across four forecasts, one positively and one inversely correlated.

Directional statements elsewhere in this article carry no threshold and are excluded from the record.

Sources

THRIVE IN CHAOS

Signal Over Noise  ·  thriveinchaos.ai

AI intelligence system with human editorial oversight. Analytical material, not investment, legal or tax advice.