How Should Companies Set Magnet Selection and Inventory Strategy When Rare-Earth Prices Are Volatile?

How engineering and procurement should split the magnet decision between a physics-based specification and a sourcing policy when rare-earth input costs are unstable.

Rare-earth magnet pricing has often behaved less like a smooth commodity curve and more like a sequence of steps. Public mineral-commodity summaries record repeated sharp moves in rare-earth oxide prices, associated with export controls, production-quota changes, and demand shifts in large NdFeB-using sectors such as traction motors and wind turbines. Some of those moves are large relative to a converter's typical margin and unfold inside a quarter; extreme episodes have been larger still, and they do not arrive on a schedule. How quickly an oxide move appears in a finished-magnet price depends on the contract, the supplier's raw-material inventory, and the grade. A lag of weeks to a few months is a common commercial pattern. It is not a physical constant, and it should not be written into a policy as if it were one.

For a design engineer, the practical consequence is that the grade locked at prototype approval may not be the grade the company can sustain at ramp. For procurement, a purchase order written against a fixed annual forecast is exposed to an input price the buyer does not set and, in a tight market, to an allocation the buyer cannot see.

The decision in front of most teams is not "which magnet grade is best." It is: how much magnetic performance should be designed in as margin, and how much should be held as inventory or contract flexibility instead? Those two levers draw on the same budget. Teams that treat them as separate decisions usually over-specify the magnet, over-stock it, or both.

The real decision is how to split performance margin against supply buffer, not which grade is cheapest today

Most cost-reduction programs start by asking the magnet supplier for a lower grade. That lever is visible, and it is usually the wrong first move, because a grade change is the one that is hardest to reverse. Once demagnetization margin has been given up for a saving that disappears when prices mean-revert, the requalification cost remains.

A grade change is also easy to misread physically. In sintered NdFeB, the heavy rare earths most often added to raise high-temperature coercivity — dysprosium and terbium — generally trade remanence for coercivity when they are alloyed through the bulk. A "lower" grade is therefore not automatically weaker in torque. It may offer similar or higher room-temperature flux and less resistance to demagnetization when hot and under peak field. Torque capability and demagnetization margin have to be checked separately. Both also depend on the magnetic circuit, not on the grade name alone.

A more durable framing separates the magnet decision into two axes that share a budget but should not share an owner:

• Performance axis (engineering-owned). What magnetic performance must the design have to meet torque, efficiency, and demagnetization limits across the full operating envelope, including peak current and fault, not only steady load? This floor is set by the load line, the temperature-dependent BH curve, and the customer's reliability requirements. It is not set by the oxide price.
• Supply axis (procurement-owned). Given that floor, how should the company buy — one qualified grade with a time buffer, two qualified grades inside a range above the floor, the same nominal grade from two qualified sources, or contracted volume with a pricing mechanism that separates oxide cost from conversion?

The engineering value of the split is that the performance axis becomes a range with a published floor, and the supply axis becomes a decision inside that range. When dysprosium or terbium quotations spike, procurement can move inside the approved range instead of opening a design change. When prices fall, the design does not need to be renegotiated; the buyer purchases at whichever qualified point in the window is commercially better.

What this requires of the specification: state a minimum demagnetization criterion and the other magnetic limits the assembly must meet, and state a qualified range above that floor, with qualification evidence that covers the whole range. That is what turns a price shock into a purchasing choice. It does not turn an unmet floor into a purchasing choice.

Four ways of covering the same requirement differ in engineering consequence and commercial exposure

The comparison that matters in this article is not a ranking of NdFeB against SmCo or ferrite. Substituting another magnet family can be rational when the performance floor allows it, but it changes the magnetic circuit, the mechanical design, and often the drive — not only the purchase order. That is a different project. The table below compares four ways of covering the same functional requirement, each with a different exposure. It is qualitative on purpose. Prices, lead times, and margins have to come from the buyer's own quotations and tests. No universal coefficient converts an oxide index into a safe grade.

Two observations follow directly from the table.

First, the graded window and dual-source strategies are complements, not substitutes. A window with only one supplier behind it turns a price problem into an allocation problem. Dual source on a single grade turns an allocation problem back into a price problem. Running both usually buys more resilience per unit of inventory than either alone. Neither removes the need for the performance floor.

Second, designing for lower heavy-rare-earth dependence is the only strategy in the table that changes the material exposure itself, so it belongs at the design stage rather than as a purchasing tactic after tooling is frozen. It is also the strategy most often misapplied, typically on a duty whose magnet temperature, or whose section thickness, the lean grade cannot support. Grain-boundary diffusion concentrates dysprosium or terbium near grain boundaries and can raise coercivity with less total heavy rare earth than bulk alloying. It does not repeal the knee of the curve, and it does not treat an arbitrarily thick section uniformly.

What this means for procurement: the four strategies have different reversal times. Moving inside an already qualified window is a purchasing cycle. Qualifying a second source is a longer effort and can run to months where coating or process audits are required. Changing grade architecture is a design cycle. Inventory covers the fast layer. Qualification covers the slow layer. One does not replace the other.

A workable policy fixes the performance floor and does not trade away the margins that price cannot buy back

There is no universal stock level or grade pair. For many industrial NdFeB duties whose magnet temperature — not the housing sensor — falls in a roughly 80–150 °C band, a workable policy has the following shape. Outside that band the same logic applies, but the floor is either easier to hold or much harder. The band is an illustrative scope, not a survey result, and it is not a rating of any particular product.

Performance floor. Set the floor from a demagnetization check at maximum magnet temperature, combined with the peak applied field the duty can produce, including overload or fault if the product must survive it. Include thermal cycling and the tolerance stack-up of the assembly, especially air gap and magnet placement. Express the floor as engineering limits, not as a grade name: a minimum intrinsic coercivity at a stated temperature, measured by a stated method; a requirement that the load line remain clear of the knee under that same condition, with the clearance defined; and a minimum remanence or flux if torque or efficiency needs it. Intrinsic coercivity alone is not a working point. A working point is a (B, H) condition on a load line. Review the floor when the thermal model, the current waveform, the fault requirement, or the mechanical stack-up changes — not when the oxide price moves.

Qualified range above the floor. Qualify one or two grades that meet the floor, and qualify the assembly at the low end under the worst-case thermal and field condition, not only at the grade used for the first prototype. The width of that range is a commercial asset only after the low end has been shown to work. A wide unqualified range is not room to manoeuvre. It is an open nonconformance.

Inventory as a time buffer, sized in weeks of demand, not as a price bet. The buffer insures against allocation and replenishment delay. It does not insure against a thermally marginal design. A practical heuristic is to size it to the shorter of (a) the time required to qualify an alternative already inside the approved range and (b) the duration of the supply events the team has actually seen in that grade and region. For some teams that lands between about one and two quarters of consumption on grades with a single qualified source and a long requalification, and much lower — sometimes near zero beyond ordinary pipeline stock — where three qualified sources can deliver inside the window. Treat those durations as planning heuristics to be replaced by the team's own lead times, not as optima.

Contract structure. Where volume justifies the negotiation, prefer a mechanism that separates a published rare-earth oxide index from a conversion charge covering manufacturing, coating, and tolerance control, rather than a single flat price that hides both. The conversion portion is usually less volatile than the oxide portion. It is not inherently fixed: energy, yield, labour, and coating inputs still move. Indexation is a contractual option a supplier may or may not accept. It is not a market law.

Specify performance, not a frozen composition. State the magnetic and coating requirements on the drawing and the purchase specification. A requirement of the form "Hcj not less than a stated value at 120 °C, by a stated method, and no irreversible loss beyond a stated limit on the specified load line" allows a supplier to offer a heavy-rare-earth-lean route without a drawing change. A requirement written as a single commercial grade name, or as a fixed dysprosium or terbium content, does not. If torque matters, also state the flux or remanence limit, so a high-coercivity, low-remanence offer cannot pass the paper specification and miss the shaft. Elevated-temperature coercivity should cite the measurement method. A room-temperature grade table does not imply it.

What this means for the design record: an assembly tested only at the top of its range is qualified for that sample, in the price environment that produced it. It is not qualified for the bottom of the window.

These strategies fail in predictable places, and some applications should not use them at all

A resilience policy applied to the wrong duty spends money without buying protection. The limits below are structural. They are not a catalogue of supplier defects.

Inventory cannot buy back demagnetization margin. If the magnet temperature and the applied field put the load line near or beyond the knee, the remedy is a different grade, a different load line, or a different thermal design. Stock does not move the knee.

Reversible temperature loss and irreversible demagnetization are different problems. Remanence falls as temperature rises and largely recovers on cooling if the knee was not crossed. That reversible drop changes torque and efficiency, and a larger warehouse does not remove it. Irreversible loss occurs when the operating point passes the knee and remains after cooling. A graded window has to be closed against irreversible loss. Sizing the machine only for the reversible drop understates the risk of the cheaper grade.

Eddy-current heating can put the magnet well above the housing temperature. In high-speed machines, and in canned or sleeved assemblies where a conductive containment wall sees a time-varying field, losses in the sleeve and in the magnet raise magnet temperature above the bulk housing reading. A window validated against the housing sensor is not validated. This is a common reason a lower grade fails in the machine after passing a bench check at the wrong temperature.

Peak field, not only temperature, sets the demagnetization case. Armature reaction, overload, and fault current can demagnetize a magnet that was comfortable at no-load temperature. A window that ignores the current waveform is open on the field axis even if the thermal axis was checked.

Coating, particle generation, and sealing can forbid a grade that magnetics would allow. In canned or hermetically sealed machines, coating integrity and debris matter independently of flux. A lower-cost magnet with a different coating stack may be inadmissible on cleanliness or corrosion grounds. That constraint should be on the drawing before procurement is asked to shop the window. This article does not rank coating systems; admissibility is a requirement of the machine, to be taken from the buyer's own cleanliness and corrosion specification.

Air-gap and placement stack-up spend margin the paper design thought it had. Permeance and torque both move with the as-built gap and with magnet position. If first-article builds show a wider gap distribution than the model assumed, the magnetic margin may already be consumed by assembly variation. Measure the as-built gap, and the magnet temperature, before treating the window as real.

A heavy-rare-earth-lean route fails closed, not open. If the duty is simply too hot, or the section is too thick for a diffusion treatment to protect the interior, the lean grade does not become acceptable because the oxide is expensive. The substitution is valid only where the measured demagnetization check still closes.

Inventory has a carrying cost. Obsolescence after a design change, degradation if coated magnets are stored poorly, and the capital tied up in buffer stock are real charges. For low-volume, long-life products, a smaller buffer plus a qualified second source is often the better trade. For high-volume products with one qualified source and a long requalification, the reverse is often true. Neither statement is a formula. Both are reasons to cost the buffer explicitly.

This framework addresses price volatility and ordinary allocation, not a catastrophic cutoff. No safety-stock calculation replaces knowing whether a supplier's material origin is exposed to a specific regulatory action. That question belongs in supplier qualification and in the contract, and it should be asked before the buffer is sized.

AIC Engineering supports the window by characterising the circuit, not by replacing the buyer's qualification

The policy above is usable only if the magnetic design is characterised well enough that a floor and a range can be written down and defended. That characterisation is an engineering deliverable.

The AIC Engineering team works on the magnetic circuit and the component structure: the field and thermal picture used to locate the performance floor; permanent-magnet drive and transmission assemblies in which air-gap and demagnetization margins are tight; and special motor magnet assemblies — multi-pole rings, radially oriented rings, Halbach arrays, and linear-motor magnet tracks — where a grade substitution has to be re-validated rather than assumed from a catalogue curve.

Where the analysis shows that a graded window is viable, the next constraint is how quickly the low end can be checked on real parts. The company's stated rapid-sampling route is 3–7 days. On geometries that route actually covers, it shortens the loop between "does the low-end grade hold at temperature and peak field" and "can we buy it," which is the loop that sets how much buffer stock the policy needs. It is a sampling lead time, not a qualification waiver, and a complex Halbach or large ring still needs a process check before that interval is assumed.

Permanent-magnet product inspection — dimensional, magnetic, and coating — is what makes a window auditable against the drawing. Inspection shows whether the received part matches the qualified definition. It does not replace the assembly-level demagnetization and torque check.

For teams buying across regions, regional delivery and supply arrangements matter as much as unit price, because a buffer sized in weeks assumes the replenishment lead time is known. A global supply arrangement helps only to the extent that it makes that lead time predictable.

None of this replaces the buyer's qualification authority, and none of it asserts that a particular grade, coating, or certificate is bundled with the circuit work. The useful output is a floor, a range, and parts that can be measured against both.

Teams facing a volatile magnet cost base can close the policy in four steps

1. Write the performance floor explicitly. Take worst-case magnet temperature and worst-case applied field, add thermal-cycling and tolerance margins, and state minimum coercivity, flux, and knee clearance at that condition, with the test method named. Do not state the floor as a grade name. If the load line cannot be computed yet, that calculation is the first engineering task, before any inventory policy.
2. Test the low-end candidate before opening a window. Confirm demagnetization margin at the measured magnet temperature and at peak field, not at the housing sensor, and confirm air gap and magnet position on as-built first articles. Confirm coating and cleanliness limits at the same time if the machine is sealed or canned.
3. Split ownership and publish one range. Engineering owns the floor and the qualified range. Procurement owns the buy inside that range, the contract structure, and the buffer size. Both sides should work from the same written limits. A price change is not, by itself, a reason to reopen the floor.
4. Close the loop on real parts. A window is usable only if the low end can be qualified on a timeline shorter than the supply events it is meant to absorb. Contact AIC Engineering for a free consultation and a magnetic-circuit design and rapid-sampling plan. Bring the worst-case magnet temperature, the peak-current or fault assumption, the air-gap tolerance, and the annual volume. From those, the team can say whether a qualified window is plausible and which measurements are required to close it — not a guaranteed width from the enquiry alone.

References

1. U.S. Geological Survey, Mineral Commodity Summaries: Rare Earths (annual series; the 2024 issue is linked, and later issues supersede it for current prices and trade). https://pubs.usgs.gov/periodicals/mcs2024/mcs2024-rare-earths.pdf
2. Adamas Intelligence, rare-earth market reporting (NdFeB and rare-earth oxide price tracking). https://www.adamasintel.com
3. J. M. D. Coey, Magnetism and Magnetic Materials, Cambridge University Press, 2010 — chapters on hard magnetic materials and the temperature dependence of coercivity.
4. S. Sugimoto, "Current status and recent topics of rare-earth permanent magnets," Journal of Physics D: Applied Physics, vol. 44,
5. International Electrotechnical Commission, IEC 60404 series, Magnetic materials — methods of measurement and classification for hard magnetic materials.
6. European Commission, Critical Raw Materials Act and associated critical raw materials assessments. https://single-market-economy.ec.europa.eu/sectors/raw-materials/areas-specific-interest/critical-raw-materials_en
7. H. Nakamura, "The current and future status of rare earth permanent magnets," Scripta Materialia, vol. 154, 2018 — grain-boundary diffusion and routes that reduce heavy-rare-earth content.