How magnets get recycled

Short versionRecycling is not one process. It is a choice of which stage of an existing process you hand the old material back to.

1 · Two loopsStart with the line it goes back into

A sintered neodymium iron boron magnet — NdFeB, the kind in almost everything worth recycling — gets made in the line in the figure, which runs left to right. Concentrate becomes separated oxide, oxide becomes metal, metal becomes alloy, alloy becomes powder, powder gets pressed and sintered into a block, and the block gets machined, coated and magnetized into a part.

Every recycling scheme in existence is a decision about where in that line the old material rejoins it. That is the whole taxonomy. Everything else is implementation.

If you want the forward process in detail: How magnets are made walks the same line from the other end.

1 · Two loopsThe chemistry half

The left of the line is a chemical plant. Its job is separating fifteen elements that are, from a chemist's point of view, almost identical — which is why it takes hundreds of stages of solvent extraction to do, and why the industry consolidated around a handful of places that can do it at all.

What matters here is that this half is built to reject things. Separation is a purification process. Feed it something contaminated and the contamination reports to a raffinate stream and leaves.

1 · Two loopsThe metallurgy half, and the hinge

The right of the line is a powder metallurgy shop. Its job is getting a specific alloy into a specific crystallographic alignment and holding it there through sintering. It is not built to reject anything. Whatever composition you hand it, it will faithfully turn into a magnet of that composition.

The hinge between the two halves is powder. Everything upstream of powder is about getting the chemistry right; everything downstream is about getting the physics right. So the two useful re-entry points are the two sides of that hinge — and they are what the industry means by short loop and long loop.

Short versionShort loop skips the entire chemical plant. That is its whole appeal, and the source of every problem it has.

2 · Short loopBack in as powder

Take a magnet that is already the right alloy. Break it down to powder. Put the powder back into the line at the powder stage. You have skipped mining, separation, reduction and melting, and you have kept the expensive part of the material — the alignment-capable alloy — intact.

The industrial version of this is usually hydrogen. Hydrogen gets into the lattice, the material swells and falls apart into a coarse friable powder, and the powder can be sized and blended into the normal feed. One commercial route sold as hydrogen processing of magnet scrap does exactly this, pulling magnets out of end-of-life products and returning the extracted alloy powder to the magnet supply chain.¹

Energy per kilogram, capital per kilogram and time per kilogram are all a fraction of the long route. On paper it is obviously the right answer.

2 · Short loopEverything in the feed goes into the part

Here is the catch, and it is not subtle. The powder stage has no purification step after it. Whatever arrives in that powder — every element, every oxide, every organic residue — is in the magnet you ship.

So short loop does not really have a feedstock specification. It has an alloy specification, applied to garbage. The input has to already be, within fairly tight limits, the thing you want to make.

That makes short loop a sorting and cleaning business wearing a metallurgy costume. The hydrogen step is the easy part.

2 · Short loopWhat disqualifies a feed

None of these is exotic. All of them are normal features of a real magnet in a real product.

In practiceThe two lines in that table that cost people the most money are the adhesive and the plating, and they cost it in the same way: the material assays beautifully and behaves badly. I have spent a lot of my career on magnet bonding and adhesion, which means I have spent a lot of it on the inverse problem — getting magnets to stay stuck to things. Everything that makes an assembly durable in service makes it awful to take apart cleanly, and nobody designing a rotor has ever once been asked to make the magnets easy to get back out. If a short-loop business model assumes clean separation of bonded magnets from their carriers, ask to see it done on the actual parts, at rate, not on a sample somebody prepared.

Short versionLong loop turns the magnet back into elements and feeds the front of the plant. You are not recycling a magnet; you are mining one.

3 · Long loopBack to the elements

The other option is to take the material apart completely. Oxidize it, leach it, and put the resulting rare earth solution into the same separation train that handles mined concentrate. Out the far end come separated oxides, which go on to metal, alloy, powder and magnet exactly as if they had come from a mine.

Treated this way, a pile of old magnets is just an ore — and a spectacularly high-grade one, with no thorium and no overburden. That framing is worth holding on to, because it explains both the appeal and the cost structure.

3 · Long loopYou are paying for tolerance

What the long route buys you is that the feed no longer has to be clean. The nickel, the iron, the carbon, the mixed grades — all the things that disqualify a short-loop feed — report to residue streams and leave. Separation does not care that the neodymium arrived in a rotor rather than in bastnaesite.

That tolerance is the product. It is also the bill: you are running the most capital-intensive, reagent-hungry, permit-hungry part of the entire rare earth industry, and you are running it on a feed that somebody had to go and collect first.

• Takes mixed, plated, bonded, assembled and unidentified feed
• Output is separated oxide, which anyone can buy and anyone can price
• Recovers the heavy elements — dysprosium and terbium — as separate products
• Capital and permitting are on the scale of a chemical plant, because it is one
• Every stage has a yield, and the yields multiply
• You have thrown away the alignment-capable alloy you already had, and will pay to make it again

3 · Long loopIt is a spectrum, not a switch

The two loops are the ends of a range, and the middle is populated. One Japanese producer has been recycling its own magnet powder since 2012 by three named routes at once: back to oxide, back to alloy, and back to sintered block.² That is the whole spectrum inside one company, chosen per stream according to how clean the stream is.

Which is the right way to think about it. Cleanliness of feed selects the loop, and a serious operator runs more than one because the feed is not all the same.

While the short loop is the one that gets the press, in tonnage terms the long loop is where most recycled rare earth actually moves, because most available feed is not clean.

Short versionEverything inside the factory fence is a solved problem, has been for decades, and is not what anybody means when they say they are getting into magnet recycling.

4 · Post-industrialInside the fence

Making a sintered magnet is a subtractive business. A pressed block goes into the furnace and a smaller, squarer, coated part comes out, and the difference is scrap. Depending on the geometry, a startlingly large fraction of the alloy that enters a magnet plant does not leave it inside a magnet.

So magnet plants have always had a pile of their own material to deal with, and they have always dealt with it, because it is expensive and it is sitting right there.

4 · Post-industrialFour piles, four answers

Grinding swarf — the slurry off the machining operations, magnet solids mixed with cutting fluid. Fine, wet, oxidizing and self-heating when it dries. The classic answer is to calcine it and put the oxide into separation.

Green scrap — pressed but unsintered. Still loose powder held together by nothing much. The cleanest material in the building and it goes straight back into the powder stream.

Sintered rejects and offcuts — full-density alloy, known grade, no coating. The ideal short-loop feed, and it is already inside the fence.

Uncoated blocks and returns — customer returns and obsolete inventory of known provenance. Short loop if the grade records survived; long loop if they did not.

4 · Post-industrialThis is not new, and it is not a market

It is worth being blunt about how established this is. A magnet maker recycling its own swarf is not an emerging business; it is housekeeping with a spreadsheet attached. The Japanese producer above has run it since 2012.² A US magnet plant that started up recently was producing commercial-grade dysprosium and neodymium-praseodymium oxide from the fine scrap off its own machining lines within months of opening.³ In Europe, a magnet maker's swarf now gets shipped to a dedicated recycler and comes back as mixed rare earth oxide for reintegration into that same maker's production.⁴

And that last arrangement is the tell. Post-industrial feed is nearly always spoken for. It has a known owner, a known composition, and a known value, and the owner is not confused about any of them. There is no pile of orphaned swarf waiting for a startup to discover it.

Two consequences: post-industrial volume scales with magnet production, not with magnets in service; and it is structurally limited to a fraction of what the plant makes.

Short versionPost-consumer is where the tonnage is, where the interest is, and where nearly every entrant gets into trouble — for reasons that have nothing to do with their chemistry.

5 · Post-consumerOutside the fence

Post-consumer means magnets that left the factory inside a product and are coming back inside a worn-out one: drive motors, hard drives, appliance motors, e-bike hubs, wind generators, power tools.

This is the stream everybody wants, and the arithmetic for wanting it is sound. Magnets made over the last twenty years are a large and growing above-ground stock, they are geographically distributed where the consumers are rather than where the geology is, and unlike a mine nobody needs a permit to own them.

5 · Post-consumerThe chemistry is not the problem

Almost everyone who sets out to do this has a solid background in chemistry, and it shows: the flowsheets are good. Leaching a mixed magnet feed, knocking out the base metals and running a separation to individual oxides is well-understood work, done at demonstration scale in several places.⁵

Given a reliable input stream of known composition, there are many good ways to go from there, and the differences between them are matters of taste, cost and site. That part of the problem is not where these companies die.

In practiceWhen I do technical diligence on a recycling company, the flowsheet is usually the part of the data room I spend the least time in. It is normally fine. What I want is the input side: a signed agreement with somebody who physically holds the material, a stated price for it, an assay of an actual delivered parcel rather than a literature composition, and a plain answer to what happens to that agreement when the price of neodymium-praseodymium doubles and the holder discovers he has options. A pitch deck with forty slides of chemistry and one slide of feedstock is telling you exactly where its author's attention has been — and the absence is the finding.

5 · Post-consumerHarvesting is the business

The hard part is getting a metric ton of magnet into the building at a price that leaves room for everything after it. Call it harvesting: locating end-of-life product, acquiring it against competing bids from ordinary metal recyclers, transporting it, and physically opening it to get the magnets out.

Every one of those steps is a logistics and labor problem, not a technical one — and they are all upstream of the thing the company is actually good at. The next section puts numbers on why that ordering matters so much.

Short versionMeasured as ore, a car is a terrible deposit and even a hard drive is only an ordinary one. What decides whether either is worth harvesting is not grade at all.

6 · EconomicsGrade, applied to products

Mining people judge a deposit by grade: how much of the thing you want per metric ton of the stuff you have to move. It is a useful discipline to apply to consumer goods, because it converts an intuition into a number you can argue with.

The chart uses grams of neodymium-praseodymium per metric ton of whole product, with an operating mine drawn in for scale — the Mountain Pass reserve at about six percent total rare earth oxide, of which about sixteen percent is neodymium-praseodymium, so roughly 9,400 grams a metric ton.⁶

6 · EconomicsA car is not a rich resource

Take the obvious example. A Tesla Model 3 weighs about 1,760 kilograms.⁷ European policy work assumes about 2 kilograms of NdFeB per electric vehicle, which is at the high end of the published range but a reasonable round number.⁸

The arithmetic2 kg magnet ÷ 1,760 kg car = 0.11% magnet by mass
× ~30% neodymium-praseodymium in the alloy⁹
= about 340 g of NdPr per metric ton of car
against 9,400 g/t at the mine — the car is roughly 28 times leaner than the ore

About four and a half pounds of magnet in two tons of car. The figure people usually quote in conversation is ten pounds, which is roughly double any published estimate I can find — and it would not rescue the number: at 4 kg the car is still fourteen times leaner than the ore. As a deposit you would not get out of the elevator with it either way.

6 · EconomicsNothing in the house beats a mine

Now run it on the densest thing consumer goods have to offer. A 3.5-inch hard disk drive carries about 12 grams of voice coil motor magnet in a drive that weighs 400 to 630 grams, which works out around 7,000 grams of NdPr a metric ton.¹⁰ That is the best post-consumer feed on the chart — and it still lands a little under the grade of an operating mine.

Below it the floor falls away: the electric car at 340, and then the conventional car, where a teardown of four high-volume US models found NdFeB only in the speakers and in none of the ancillary motors. Call that one 13 grams a metric ton, about a seven-hundredth of mine grade.¹¹ The scrapyard is not a rare earth deposit, and never was.

Post-industrial scrap, meanwhile, is off the top of the chart, because a sintered offcut is not a product containing a magnet. It is the magnet.¹²

6 · EconomicsThe number that actually decides it

So the whole post-consumer half of the chart sits at or below mine grade, and the best of it — the hard drive — has to be opened one at a time. Manual dismantling runs at roughly twelve drives an hour, about five minutes each, which is the figure two independent groups report.¹⁰

The arithmetic12 drives/hour × 12 g = 0.144 kg of magnet per labor-hour
at $25/hour fully loaded¹³
= about $174 of labor per kilogram of magnet recovered
against a US government floor price of $110/kg for NdPr oxide¹⁴, of which a kilogram of magnet contains about 0.3 kg — call it $33 of contained value

Five times the value of the material, spent before a single chemical step has run. That is the whole post-consumer problem in one line, and it is why every credible operation in this space is really an automation company.

It is also why the commonly quoted figure for end-of-life magnet recovery is under one percent — a number worth handling carefully, since it traces back to a 2011 estimate that has been restated for fifteen years without being re-measured.¹⁵

6 · EconomicsVintage, and why it matters

One more thing grade leaves out, and it is the reason two identical-looking piles of scrap magnet are not the same purchase: old magnets and new magnets are not the same alloy.

Everything a magnet has to survive — heat, a demagnetizing field, a rotor that gets hot in traffic — used to be bought with dysprosium, mixed through the whole alloy. The highest-coercivity grades, the ones traction motors and wind generators needed, ran 8.5 to 11 percent dysprosium by mass. A US Department of Energy demand model from 2011 put the average hybrid and electric traction magnet at about 8.7 percent, and wind generators at about 4.1.¹⁶

Then dysprosium went from a few hundred dollars a kilogram to sixteen hundred at the end of 2011, and the industry spent the next decade engineering it out. Grain boundary diffusion — putting the heavy element only at the grain boundaries, where coercivity is actually decided, instead of throughout the grain — arrived commercially around then and is now standard.¹⁷

So the arrow of time runs the wrong way for a recycler: every year, the magnets coming off production lines are a slightly poorer heavy-rare-earth resource than the ones going into retirement.

Short versionVintage changes what is in a feed far more than it changes what the feed is worth. Run the numbers before you pay a premium for old material.

6 · EconomicsBut price the whole magnet, not the dysprosium

It is tempting to go from there to "old magnets are worth several times more," and the arithmetic does not support it. On the latest published US Geological Survey annual averages — dysprosium oxide at $239 a kilogram, terbium oxide at $1,010, neodymium-praseodymium oxide at $69¹⁸ — here is the contained rare earth in a kilogram of magnet.¹⁹

The arithmetic2008 grade, 8% Dy: $22 dysprosium + $16 NdPr = $38/kg
2025 grade, 1% Dy: $3 dysprosium + $24 NdPr = $27/kg
2025 grade, 1% Tb: $12 terbium + $24 NdPr = $36/kg

Dysprosium content fell roughly eightfold. Contained value fell by about a third — because neodymium-praseodymium is a quarter to a third of the mass and dominates the total, and because in the old alloy the dysprosium was displacing NdPr rather than adding to it.

And the third line is the one that ought to stop you. Diffusion works better with terbium than with dysprosium, so a lot of modern magnets use it — and terbium is four times the price. A modern terbium-diffused magnet is worth about what a 2008 dysprosium-loaded one is. Buy by the decade and you can be wrong in either direction.

In practiceTwo cautions I would want before anyone builds a thesis on vintage. The high-dysprosium grades were always a minority of tonnage — consumer electronics, which is most of the unit volume, has always run near one percent — so the rich vintage pool is thinner than it sounds. And there were almost no electric cars in 2008, so the traction magnets that would prove the point mostly do not exist yet as scrap. The stream where this argument really bites is wind: generator magnets at around four percent dysprosium,¹⁶ in single assemblies far larger than anything in a car, with a first large cohort reaching end of life now. And then the awkward part — I can find no public source showing that anyone prices recycled magnet scrap on assayed heavy rare earth content at all. For an industry that lives on this distinction, the absence of a published payable schedule is itself the finding.²⁰

Short versionEurope has written the most specific magnet recycling law in the world, and the number everyone wants from it — the minimum recycled content — is not in it yet.

7 · RulesWhat the law actually says

The European Union's Critical Raw Materials Act sets 2030 benchmarks for Union capacity, including recycling capable of producing at least 25 percent of annual consumption of each strategic raw material.²¹ Those are capacity aspirations, not obligations on any company.

The obligations sit in two articles about magnets specifically. One requires products from magnetic resonance imaging machines to washing machines to carry a label saying whether they contain a permanent magnet and of which family, plus a data carrier giving recyclers the weight, location, composition, coatings, adhesives and removal sequence for every magnet in the product.²² Read that list again: it is a direct legislative answer to the short-loop contaminant table above.

7 · RulesPost-consumer only — which is the interesting part

The second article requires anyone placing those products on the market, where the magnets total more than 0.2 kilograms, to publish the share of neodymium, dysprosium, praseodymium, terbium, boron, samarium, nickel and cobalt in them that was recovered from post-consumer waste. Minimum shares are to be set later by delegated act, with the end of 2031 as the backstop.²³

Two details do a lot of work. Ferrite is excluded, so the cheap magnets are out of scope. And the duty counts post-consumer material only — deliberately, because counting factory swarf would let the whole industry comply on day one with the recycling it was already doing. The law has picked exactly the half of this page that is hard.

A live amendment would broaden that to pre-consumer waste as well and add export restrictions on magnet scrap, so treat the scope as a moving target and date anything you rely on.²⁴

Elsewhere: the United States has funded recycling directly but modestly, including a Defense Production Act Title III award for recovering rare earths from electronic waste.²⁵ China does not appear to run a recycling quota parallel to its mining and separation quotas — recyclers sit outside the quota but are barred from feeding mined material into it — while its export controls explicitly reach secondary resource recycling technology.²⁶

Short versionSix questions. None is about chemistry, which is the point.

8 · The screenReading a recycling pitch

Everything above collapses into a short screen. It is the same shape as the one on Is this magnet headline for real?, pointed at a company rather than a press release.

If a business can answer all six with documents, it may still fail, but it will fail for ordinary reasons. If it cannot answer the first two, nothing downstream matters.

8 · The screenThe six

1. Where does the feed come from, and who signed? Not a market-size estimate. A named holder of physical material, a contract, and a price.
2. What does it cost to open the product? In labor-hours or in capital per kilogram of magnet recovered. If the answer is a research program, that is the company.
3. Short loop or long loop, and does the feed justify the choice? A short-loop plan fed by mixed post-consumer scrap is the most common error in the sector.
4. What is in the feed that has not been assayed? Ask for an assay of a delivered parcel, not a literature composition. Ask specifically about coatings, adhesives and oxygen.
5. Which parts of the flowsheet are being reinvented, and why? Solvent extraction is a century old and works. A team rebuilding it from first principles should be able to say what was wrong with it.
6. Who has agreed to buy the output, to what specification? Mixed rare earth oxide, separated oxide, alloy and magnet are four different businesses with four different customers.

In practiceThe best answer I have heard to the first question was not a clever one. It was a recycler who had spent two years becoming the boring, reliable, pays-on-time buyer for a category of scrap that nobody else wanted, and only then started thinking about what to do with it. That is backwards from how these companies are usually built, and it is the right way round. The chemistry can be bought or hired. The input stream cannot.