The heavy rare earth bottleneck

Short versionDysprosium is about one percent of mined rare earths. On that ratio, the average neodymium magnet can hold about 1.5 weight percent dysprosium and stay in balance with supply. The grades that need it most were asking for two to seven times that.

Start with the ratio, because everything follows from it. Constantinides gives a rough production-share estimate for the four rare earths magnets actually care about, blended across bastnasite ore and ionic clays: neodymium about 15 percent of everything mined, praseodymium 5, samarium 3, dysprosium 1.² Call it 15:5:3:1. Elsewhere he puts dysprosium nearer 1.5 percent,³ so take the range rather than the point.

Now the bridge. A sintered neodymium magnet is about 31 weight percent rare earth in total.² If dysprosium is 1 percent of the rare earth that comes out of the ground, and rare earth is 31 percent of the magnet, then the average magnet can contain roughly 1.5 weight percent dysprosium and be, in his phrase, "in balance with existing supplies." Above that, the industry is drawing on a dysprosium stream the ore ratios do not support, and something has to give: price, grade, or somebody's delivery date.

Set that line against what the grades were actually asking for. This is his table, from 2015, before grain boundary diffusion had finished doing its work:³

That is the bottleneck in one picture. Not a shortage of rare earths, which there isn't; a mismatch between the ratio the earth supplies and the ratio the application wants. A magnet with no suffix is comfortably inside the budget. An SH-grade motor magnet, in 2015, wanted nearly three times its share. An EH grade wanted six or seven.

The reason this does not resolve itself is co-production. You cannot mine dysprosium. You mine an orebody and you get the whole lanthanide series in whatever proportion that rock happens to hold, and then you sell what you can. Constantinides watched the same mechanism play out one element-pair earlier: Mountain Pass in the 1980s was run for europium (color television) and cerium (glass polishing), and the magnet rare earths came out of the residue. By the mid-1990s neodymium demand had outrun its natural ratio, and the operator was stockpiling cerium and lanthanum it could not sell.⁴ Wanting more dysprosium today means wanting more of everything else that sits beside it, whether or not there is a buyer.

In practiceThe place this shows up is not a commodity desk, it's a design review. Somebody specs an EH grade because the thermal model has an uncomfortable margin and EH makes the margin go away, and the part is now carrying six or seven times its proportional claim on the scarcest thing in the magnet. Most of the time nobody at that table knows it, because the grade is a line on a drawing and the dysprosium is three tiers up the supply chain. The cheapest heavy rare earth in any program is the one a better thermal design meant you never needed.

Short versionDy and Tb have much larger anisotropy fields than neodymium, which is what buys coercivity. They also have much lower saturation polarization, which is what you pay with. Terbium is about 1.5 times better at the job and several times the price.

The relevant numbers are three rows of one table:⁵

The anisotropy field is roughly the field the crystal itself resists being turned by. More of it means the magnetization is harder to reverse, which is coercivity, which is what keeps a magnet from quietly giving up some of its output the first time the motor gets hot and the inverter asks for a large negative d-axis current. Substituting dysprosium or terbium for neodymium raises it a lot. It also drops saturation polarization by more than half, and saturation polarization is what remanence — the useful flux — is made of. That is the whole trade in two columns: you buy coercivity with remanence, and the older the grade, the more literally that is true.

Two things worth keeping straight, because they get conflated constantly. Curie temperature is a hard limit of physics, the point at which the material stops being ferromagnetic at all; nobody designs anywhere near it. Maximum working temperature is a rule of thumb about how much irreversible loss you are willing to accept in a particular magnetic circuit, and it is the number that binds. The grade suffixes are about the second one. Constantinides puts the onset of the dysprosium requirement at roughly 80 °C² — which, since a traction motor rotor has no interest in staying below 80 °C, is the sentence that connects section 1 to the automotive industry.

Terbium is more effective, and terbium is scarcer and dearer, so for decades dysprosium was the default.⁵ That default is exactly what changed when a process arrived that could spend a small amount of heavy rare earth in precisely the right place.

Short versionReversal starts at the grain boundary, so put the heavy rare earth at the grain boundary. It roughly halved the loading. It did not make the problem go away, and it does not work on thick parts.

This is covered at length on the grade chart page, and the figure below is the same one, so I will keep it short here. A sintered magnet is millions of grains. Demagnetization nucleates at defects right at the grain boundaries and sweeps inward. Alloying dysprosium through the entire grain to fix a boundary problem works, but most of the dysprosium ends up in the grain core, where it does nothing for coercivity and a great deal against remanence.

The saving is real and it is large. Constantinides puts it at roughly half: SH grades from over 4 percent dysprosium down to about 2, and the 8–11 percent grades down to 4–5.³ Go back to the table in section 1 and halve the right-hand column. SH lands near 2 percent, which is within shouting distance of the 1.5 percent balance line. EH and AH land at 4 to 5, which is still three times over. Halving a number that was six times too big leaves you three times too big. Grain boundary diffusion bought the industry a decade. It did not repeal the arithmetic.

There is also a limit that gets skipped in most summaries. Diffusion reaches as far as diffusion reaches: practically, 3 to 4 millimeters from the surface, with the concentration falling off exponentially with depth, which puts the maximum thinnest dimension of a fully treated magnet at around 6.5 millimeters.⁵ Thin rings, arcs and plates — which is most of what goes into a motor — are fine. A large block is not, and large blocks still get their coercivity the old way, out of the bulk alloy. So the heavy rare earth demand that grain boundary diffusion does NOT address is concentrated in exactly the parts that are biggest and most expensive to get wrong.

In practiceWhen somebody tells you a program has "designed out" its dysprosium exposure, the useful follow-up is about geometry, not chemistry. What is the smallest dimension of the largest magnet in the assembly? If the answer is comfortably under six millimeters, the diffusion story probably holds. If somebody is buying a thick block and quoting a diffused grade's heavy rare earth content, one of those two things is wrong, and it is usually not the block.

Short versionChina's mining quota is split in two. The light half grew 149 percent from 2018 to 2024. The heavy half sat at 19,150 tons every single year. Then, from 2025, they stopped publishing the split at all.

China issues annual production quotas for rare earth mining, split into two categories: rock-type, which is the light rare earths out of Bayan Obo and goes overwhelmingly to China Northern Rare Earth; and ion-adsorption type, which is the southern clays, which is the heavy material, and which goes to China Rare Earth Group. The two halves have had completely different histories.⁶⁷

From 2018 to 2024 the light quota went from 100,850 tons to 250,850, up 149 percent. The ion-adsorption quota was 19,150 tons in 2018 and 19,150 tons in 2024, and in all six years in between.⁶ Not slow growth. No growth. The heavy share of China's own mining quota fell from 16 percent to 7.1 percent while the total more than doubled.

There are two ways to read a number that constant. One is resource constraint: the southern clays are a finite and environmentally brutal resource, and in 2012 their remaining life was being estimated at 15 to 25 years at then-current rates,⁸ which on its own terms runs out somewhere between next year and 2037. The other is policy: a quota held deliberately flat is a supply curve that does not respond to price, which is a useful thing to own if you are the only one who owns it. I do not think you have to choose. Both are true, and they point the same direction.

The quota is not the throughput

Here is the part that makes the quota figure misleading if you stop there. China's heavy rare earth FEEDSTOCK is much larger than its heavy rare earth quota, because a great deal of it is not Chinese. Global Witness, working from Chinese customs data, put imports of heavy rare earth oxide from Myanmar at 41,700 tons in 2023 — more than double China's entire domestic heavy mining quota, in a trade worth $1.4 billion.⁹ The material is ionic clay mined in Kachin State and trucked across the border for separation.

So the world's supply of the elements that decide whether an electric vehicle motor survives a hot lap runs through an active civil war. In October 2024 the Kachin Independence Army took Chipwi and Pangwa, the towns at the center of that belt; China closed border gates; imports collapsed to 311 tons in February 2025, down 89 percent year on year.¹⁰ USGS now estimates Myanmar's output at 22,000 tons for 2025, down from 27,000 in 2024, and flags that the figure is itself derived from Chinese import data.¹¹

One detail there is worth pausing on, because it cuts against the obvious story. Between late September 2024 and March 2025, terbium oxide rose 21.9 percent in China while dysprosium oxide FELL 3.2 percent.¹⁰ The seizure moved terbium and did not move dysprosium. Anyone telling you the Kachin fighting spiked the heavy rare earths is half right, and it is worth knowing which half.

The control line was drawn on exactly this boundary

In April 2025 China put seven elements under export license: samarium, gadolinium, terbium, dysprosium, lutetium, scandium and yttrium, along with their oxides, compounds, alloys, sputtering targets, and neodymium magnets containing terbium or dysprosium.¹² Neodymium and praseodymium — the two elements that make up most of a magnet by mass, and most of its cost in ordinary times — were not on the list, and have not been since.

That is not an oversight. Restricting neodymium would hurt Chinese magnet exporters and could be mined around within a decade. Restricting dysprosium and terbium costs China's own producers comparatively little and cannot be mined around at all quickly, for every reason in sections 1, 5 and 6 of this page. The October 2025 escalation, which added an extraterritorial rule covering foreign goods with more than 0.1 percent Chinese-origin rare earth content, was suspended in November until 27 November 2026. The April measures were not suspended and remain in force.¹³

The underlying position is a gradient, and the gradient is the argument. The IEA puts China at 60 percent of mined magnet-rare-earth production in 2024, 91 percent of refined output, and 94 percent of sintered magnet production.¹⁴ Concentration rises as you move downstream — and heavy rare earths sit at the far end of it. Benchmark Mineral Intelligence puts China at 85 percent of all rare earth oxide production and 99 percent of dysprosium and terbium oxide.¹⁵ USGS, which publishes no separation-share figure for the heavies at all, gives the primary-source version of the same point: China supplied 100 percent of US terbium, holmium and lutetium compound imports over 2021–2024, and US net import reliance for heavy rare earth compounds and metals was 100 percent in every year from 2021 to 2025.¹⁶

In practiceWhen somebody quotes you a Chinese rare earth quota, the first question is which half, and the second is what year, because from the 2025 first batch onward the split has not been published — issued quietly in June 2025, with recipients reportedly told not to share the numbers.¹⁷ A peer-reviewed treatment calls this the start of a "nondisclosure era."¹⁸ For anybody building a supply model, the practical consequence is that the most load-bearing input just became unobservable, and any model that still shows a confident 2026 heavy quota is showing you an assumption wearing a number's clothes.

Short versionBefore May 2025 the answer was zero. Today it is essentially one plant in Malaysia. Everything announced and financed for the rest of the decade adds up to something like 40 percent of what China and Myanmar make now — and that is the optimistic reading.

The first commercial-scale dysprosium oxide separated outside China was produced by Lynas at Kuantan, Malaysia, in May 2025, with terbium following in June.¹⁹ Note the date. Not the first significant volume, not the first cost-competitive volume — the first. Forty years into the neodymium magnet industry, the count outside China was zero until last year.

Before the chart, the honest caveat, because it governs everything in it: there is no authoritative public figure for world dysprosium or terbium production. USGS publishes a dedicated heavy rare earths chapter and puts no world production table in it.¹⁶ Lynas, the one company outside China separating these at scale, has never published a dysprosium or terbium tonnage in any quarterly or annual report. Every headline number in this industry is a total-rare-earth-oxide number, and the heavies are a sliver inside it that nobody is obliged to disclose.

Two sourced numbers on that chart do not reconcile, and I would rather show the seam than hide it. Benchmark's 99 percent Chinese share of dysprosium and terbium implies about 40 tons a year outside China. Trade-press readings of Lynas's own reporting imply something nearer 250 to 300 tons of dysprosium oxide and 40 to 60 of terbium, which would be 7 to 9 percent. Both are 2025–26 figures. The gap is approximately the difference between "before Lynas" and "after Lynas," and neither figure is auditable, for the reason in the paragraph above. Treat the ex-China number as somewhere between negligible and under a tenth, and be suspicious of anyone who states it more precisely than that.

The projects, graded

• Lynas, Malaysia. Operating. The only one. Mt Weld is worth understanding here because it is instructive: it is not a heavy deposit. Lynas states 12,790 tons of contained dysprosium oxide, which against its reported reserve or resource works out to roughly 0.3 to 0.6 percent of total rare earth oxide.²² Mt Weld is a very large LIGHT deposit with just enough dysprosium in it to justify a circuit. That is a different business from a heavy deposit, and it is the business that got there first.
• Iluka, Eneabba, Western Australia. 725 t/yr Dy+Tb, targeted 2027, financed by a A$1.65 billion Australian government loan, construction more than half done.²¹ On tonnage and on funding certainty this is the largest credible thing on the board.
• Caremag, Lacq, France. 600 t/yr Dy+Tb, targeted late 2026, €216 million closed with Japanese and French state money.²⁰ Nearest term of the large ones.
• MP Materials, Mountain Pass. More than 200 t/yr Dy+Tb planned; heavy separation was called imminent for mid-2026 and I could find no production announcement as of today. Worth knowing why the number is that size: Mountain Pass bastnasite is heavy-poor, and MP's own SEG+ product — which is ALREADY the heavy-enriched fraction, after cerium, lanthanum and NdPr are taken out — is about 4 percent dysprosium and terbium on a total-oxide basis.²³ Against NdPr output of 2,599 tons in FY2025, full build-out puts MP's Dy+Tb to NdPr ratio around 1 to 30. The Defense Department price floor that underwrites the company covers NdPr, not the heavies.
• Energy Fuels, White Mesa, Utah. 47 t/yr in phase 1B, targeted 2027 and already slipped from late 2026; a further 368 t/yr in a phase 2 that is not financed.²¹ Section 6 is about why a uranium company is on this list.
• Aclara, Louisiana. 173 t/yr, targeted 2028, not financed — a non-binding EXIM letter of interest is not money.²¹
• Serra Verde, Brazil. Producing ionic-clay material since 2024 and frequently listed as a heavy rare earth producer. It makes a mixed rare earth carbonate and does not separate. That is feed, not supply, and the distinction is the entire subject of this page.

Add the quantified announcements and you get about 2,140 t/yr of ex-China dysprosium plus terbium separation targeted by 2030, of which roughly 1,572 tons — 73 percent — is financed or under construction and about 568 tons is aspirational. Against an estimated world total near 4,000, the financed portion is around 40 percent of today's output and the full announced portion a little over half. That is not nothing. It is also, if every single project lands on its announced date at its announced capacity, still less than China and Myanmar make now.

In practiceThe diligence question that sorts this list fastest is not about the resource, the flowsheet or the offtake. It is: what is the first date on which this facility will separate a kilogram of dysprosium from anything, and what has that date been for the last three years? A project whose first-heavies date has moved right twice is telling you something the technical report is not. And a capacity figure quoted as total rare earth oxide, in a conversation about dysprosium, is nearly always doing the work of hiding the number you asked for.

Short versionThorium and uranium follow the rare earths into the same minerals. Monazite runs 5 to 10 percent thorium oxide against a US regulatory threshold of 0.05 percent. That is not a permitting inconvenience, it is a different industry — which is why a uranium company is now a heavy rare earth company.

Thorium and uranium substitute readily into the crystal sites the light and heavy rare earths occupy, so the minerals that concentrate rare earths tend to concentrate actinides too. The numbers, by mineral:²⁴

In the United States, "source material" is defined at one-twentieth of one percent — 0.05 percent, or 500 parts per million — of uranium or thorium by weight, and above it you need a license under 10 CFR Part 40, from the NRC or from an Agreement State.²⁶ Monazite at 6 percent thorium oxide is roughly 120 times that threshold. It is not a marginal call, an exemption, or a matter of good environmental practice. It is the same legal category as uranium ore.

Two things follow, and the second is the one people miss.

First, xenotime — the phosphate that actually carries the heavy rare earths, where the heaviest lanthanides live — is a uranium problem more than a thorium problem. It runs about seven times less thorium than monazite and about three times more uranium.²⁴ Both are around sixteen times over the threshold, so the conclusion survives, but if you repeat the common line that the heavy mineral is the thorium-rich one you will have the mechanism backward. And in practice xenotime is mined blended with monazite anyway, so the feed carries monazite's thorium regardless.

Second, and this is the load-bearing one: ion-adsorption clays are the exception. Their rare earths sit as loosely bound ions on clay surfaces rather than locked in a thorium-bearing crystal lattice, so they are simultaneously heavy-enriched and low-radioactivity. Put that together with section 4 and a lot of things line up at once. The world's heavy rare earth supply comes from ionic clay, in southern China and Kachin State, because ionic clay is the one heavy-enriched resource that does not come with an actinide problem attached. It is not only that China got there first. It is that the deposit type China and Myanmar have is the one that skips the hardest regulatory step, and almost everybody else's heavy resource is monazite or xenotime.

Which is why a uranium company is now a rare earth company

Energy Fuels' White Mesa Mill in Utah is a conventional uranium mill with a Utah Agreement State radioactive materials license — not, despite constant trade-press repetition, an NRC license — and an on-site 11e.(2) byproduct disposal cell. It has been processing monazite since 2021, produced its first kilogram of 99.9 percent dysprosium oxide in August 2025, and announced roughly 30 kilograms of dysprosium and 1 kilogram of terbium from US monazite in March 2026.²⁷ The company is refreshingly direct about the position: "Uranium and thorium are problems for most REE companies — but we want them!"

The regulator has now said the quiet part in writing. In a September 2025 technical assistance letter to Utah's environmental regulator, the NRC advised that where ore is processed for both its source material AND other constituents such as rare earths, the wastes are generally 11e.(2) byproduct material — and that rare-earth-only processing does not qualify.²⁸ Read that carefully, because it is the moat in one sentence: the company that is ALSO recovering uranium gets a disposal pathway that the company recovering only rare earths does not.

Two things keep me from overselling that moat. It binds on CRACKING monazite, not on separating rare earths; an operation like Ucore's, taking already-cracked mixed oxide, never touches source material and is a real US capability that simply depends on somebody else having done the radioactive step. And "nowhere in the US to dispose of thorium" is too strong — licensed disposal exists at Clive, Utah and Andrews, Texas. The advantage is an on-site cell versus case-by-case third-party acceptance plus transport of licensed material, which is a large commercial difference and not an absolute one.

Elsewhere, the same constraint shows up as lost decades rather than lost margin. Lynas spent fourteen years fighting over the residue from ore assaying 750 ppm thorium oxide before receiving a ten-year Malaysian license in March 2026, conditional on ceasing water-leach-purification residue generation by 2031, treating existing residue below 1 becquerel per gram, and building no new disposal facility.²⁹ India holds something like 13.15 million tons of monazite and cannot commercially exploit it, because monazite is a prescribed substance under the atomic energy framework and sits with the Department of Atomic Energy rather than with miners.³⁰ The largest untouched heavy-rare-earth-bearing resource in the friendly world is locked up by its thorium, and has been for sixty years.

In fairness — and this is a real objection, not a rhetorical one — thorium is a weak alpha emitter with a half-life measured in billions of years, and there is a respectable industry argument that the regulatory burden is disproportionate to the hazard. I looked for a citable authority making that case and did not find one; the mining industry's own briefings accept the stringency as legitimate. The Bayan Obo tailings, for their part, measure around 276–321 mg/kg thorium-232 — BELOW the US source-material threshold — with a maximum dose rate of about 1.15 mSv/yr against natural background nearer 2.4.³¹ The problem there is volume and containment, not concentration. Treat that as a caution against both the "it's basically uranium" framing and the "it's basically nothing" framing.

In practiceIn diligence on any rare earth project, the question that separates a real plan from a deck is: where does the thorium go, whose license permits it to go there, and for how many years is that arrangement secured? Answers involving future legislation, a facility that does not exist yet, or the word "minimal" are the same answer. I have seen flowsheets that are genuinely elegant right up to the point where a stream leaves the page and the arrow has no destination.

Short versionHigh dysprosium prices make unlikely resources look interesting, and the price has been high enough, more than once, to be a fair argument. The problem with the coal route is not the price deck. It is that ten parts per million is ten parts per million.

First, the fair part, because the easy debunk gets this wrong. Dysprosium has repeatedly reached prices at which marginal resources genuinely pencil. It went from $91/kg in January 2009 to $2,377/kg in August 2011, a 26-fold rise in 31 months.³² And the relevant price today is not the Chinese one. Export licensing has split a formerly single world price in two: in September 2026, dysprosium oxide was assessed around $212–218/kg inside China and around $3,250/kg in reported North American transactions, a gap of roughly fifteen times; terbium about $990 against $7,500.³³ A US producer sells into the second market, not the first.

Which is why the usual criticism of Ramaco's Brook Mine in Wyoming — that its price deck is fantasy — does not hold up. The July 2025 preliminary economic assessment assumes $850/kg dysprosium oxide against a USGS 2025 estimate of $239/kg, and 3.6 times the published price looks aggressive until you notice that Benchmark had dysprosium trading at 4.4 times the China price ex-China in 2025, forecast to reach 8.3 times by 2027.¹⁵³⁴ That is about $1,052/kg — ABOVE their assumption. On price, the deck is conservative.

The problem is the grade, and it is not close. Ramaco publishes a total rare earth oxide grade of about 500 ppm and a distribution in which dysprosium is 2.1 percent of the basket and terbium 0.4; it does not publish a per-element table, citing proprietary information.³⁴ Multiply it out and you get roughly 10.5 ppm dysprosium oxide and 2 ppm terbium in run-of-mine material. That arithmetic checks against their own numbers: 2.62 million tons a year of feed at 10.5 ppm is 27.5 tons of contained dysprosium oxide, and the PEA projects 25 short tons, implying about 83 percent recovery. Internally consistent.

Now compare it to things. Published Powder River Basin fly ash runs 5 to 15 ppm dysprosium.³⁵ Browns Range xenotime ore in Australia runs around 840 ppm dysprosium oxide — eighty times more. The fly ash paper's own benchmark for a good rare earth ore is 1 to 2 percent total rare earths, which is 10,000 to 20,000 ppm. Brook Mine's dysprosium grade is indistinguishable from ordinary Powder River Basin fly ash. That is the whole finding, and it comes from two independent published sources rather than from an opinion.

Twenty-two point seven tons a year of dysprosium oxide is about 0.6 percent of estimated world output, for $533 to $580 million of capital and 2.62 million tons a year of mining. And in Ramaco's own model, dysprosium and terbium together generate about $33 million a year of revenue against roughly $143 million of steady-state EBITDA. The heavies are not what makes this project work even on its own numbers. Scandium, gallium and germanium are — which is awkward, since Wolfpack Research put world scandium oxide demand at about $25 million a year, and Ramaco's own Defense Logistics Agency offtake is 6.4 tons of scandium, roughly the entire world market in one contract.³⁶ A shareholder lawsuit was filed in April 2026; the company's position is that its critics are short sellers.

None of which makes the underlying idea worthless. Coal ash, phosphogypsum and acid mine drainage are all real rare-earth-bearing streams that somebody has already mined, ground and moved, and recovering something from a waste you are already managing is a legitimately different economic proposition from opening a mine. But the ratio between ten parts per million and a thousand does not respond to price, permitting, policy or enthusiasm. It is a hard limit of the resource, not a technological challenge, and the projects that eventually work in this space will be the ones whose economics survive the day China resumes shipping at $212 a kilogram.

In practiceThe diligence question for any of these is not "does the chemistry work." It usually does. It is: what price does this need, and what happens to it at the price we had in 2016 — which for dysprosium was about $180 a kilogram. If the answer is that the project closes, then what is on the table is a dysprosium price forecast with some engineering attached, and nobody can underwrite a dysprosium price forecast. If I could, I would not be writing web pages about it.

Short versionThe bottleneck is one percent of the rock, ninety-nine percent in one country, largely fed by a war zone, and structurally hard to replace because the good deposits are radioactive and the clean ones are already taken.

Pulling the threads together: dysprosium is about one percent of mined rare earth and the grades that need it want several times that; grain boundary diffusion halved the requirement and stopped there, and cannot help thick parts at all; China's heavy quota has not moved in eight years and is no longer published; the feedstock behind it comes substantially from Kachin State; the April 2025 export controls were drawn precisely on the heavy/light line; the first ex-China commercial dysprosium was separated sixteen months ago; and most of the announced replacements are carrying a thorium problem that adds years before it adds tons.

Five things worth watching, and one date.

• 27 November 2026. When the suspension of the October 2025 extraterritorial controls expires. The April 2025 regime continues either way, so this is about the 0.1 percent content rule coming back rather than about the heavies becoming controlled — they already are.
• Whether Caremag and Eneabba hit 2026 and 2027. Those two are 1,325 of the roughly 1,572 financed tons. If they land, the ex-China heavy position genuinely changes. If they slip two years, nothing on the chart in section 5 matters until 2029.
• Whether Lynas or anyone else ever publishes a dysprosium tonnage. The single most useful disclosure nobody is making. Until then, every market model in this field, including the chart on this page, is built on a derivation.
• Kachin. Not for the headlines but for the customs data. Chinese imports of heavy rare earth oxide from Myanmar are the best available proxy for how much heavy material is actually moving, and they are published monthly by somebody, which is more than can be said for the quota.
• Whether any Western price floor extends to the heavies. The Defense Department arrangement with MP Materials sets a floor under NdPr, which is not where the scarcity is. A floor under dysprosium and terbium would change the financing conversation for half the projects in section 5. Its absence, so far, is worth noticing — and absence, as ever, is the finding.

The last thing, which is less a prediction than a warning about a habit. Every wave of ex-China rare earth projects since 2011 has been financed at the top of a price spike and buried at the bottom of the trough that followed, and the current wave is being financed at a fifteen-fold Western premium that exists because of an export license regime. If that regime is relaxed — for any reason, including a good one — the premium goes with it, and a fair number of the tons drawn on the chart in section 5 go with the premium. That is not an argument against building them. It is an argument for knowing which price each of them needs, and asking before the money moves rather than after.