How electric motors work

Short versionTwo mechanisms, and only two. A rotor carrying its own field lines up with the stator's. A rotor of shaped iron twists to give flux the shortest steel path. Everything after this is a choice about which one you use and how you pay for it.

1 · Alignment torqueA compass needle in a rotating field

Put a magnet on a shaft and surround it with a field you can rotate. The rotor's own field wants to line up with the applied one, and any angle between them makes torque. Turn the field and the rotor follows.

δ is the load angle — how far the rotor trails the field. Torque peaks at 90° and collapses past it, which is what "pulling out of step" means. This is alignment torque. You get it from a permanent magnet or from a rotor coil carrying current, and from nothing else.

1 · Reluctance torqueIron alone will do it too

Take the magnet away and leave a shaped lump of iron. Flux prefers the short route through steel to the long route through air, so the rotor twists to present its easy axis to the field. (Reluctance is the magnetic circuit's version of resistance: how hard it is to push flux down a given path. Low reluctance is where flux wants to go.)

Note the 2. Reluctance torque repeats twice per revolution of the field, because the iron does not care which end points at the north pole. While it needs no magnet and no rotor current, it is weaker than alignment torque for the same size, and it only exists if the rotor has saliency — a much easier direction and a much harder one.

1 · The rotating fieldThree phases, no moving parts

The rotating field is the trick the whole industry is built on. Three windings spaced 120° apart around the stator, carrying three alternating currents (AC) 120° apart in time, add up to one field of constant size sweeping around the bore at the supply frequency.

Nothing in the stator moves. Change the frequency and the field speeds up or slows down — which is exactly what an inverter (sold as a variable-frequency drive, or VFD) does: it synthesizes AC at whatever frequency and amplitude the machine needs. Change the pole count and the field sweeps proportionally slower at the same frequency.

Two families follow from here: rotors that lock to the rotating field and turn at exactly its speed — synchronous — and rotors that have to lag behind it to work at all — asynchronous, which means induction. What follows runs roughly in the order these machines were invented.

Short versionThe oldest design, and still the cheapest way to turn a battery into rotation. A sliding contact does the switching, which is why it needs no electronics — and why it wears out.

2 · Brushed direct current · from the 1830sThe mechanical inverter

Before power electronics, the switching had to be done mechanically. A brushed direct current (DC) machine puts the windings on the rotor and the field on the stator, and lets a commutator — a split ring of copper segments with carbon brushes riding on it — reverse the current in each rotor coil twice per turn, so the rotor's field is always held at the angle that makes torque.

It is the same job an inverter does today, done by a sliding contact and actuated by the shaft itself. Driving one could not be simpler: apply a DC voltage and it runs, speed roughly proportional to voltage and torque roughly proportional to current.

It is also the only machine on this page with a consumable part. Brushes wear, they spark, they throw electrical noise and carbon dust, and they set a hard ceiling on speed. Every topology after this one is, in some sense, an attempt to do that switching without touching anything.

Magnet billFerrite arcs — strontium or barium hexaferrite — in almost every low-cost application, and wound field poles in the larger ones. Neodymium turns up only where size matters: premium cordless tools, some servo actuators.

2 · Universal motor · from the 1890sThe same machine on AC

Wire the stator field winding in series with the rotor armature and something useful happens. Reverse the supply and both the field and the armature current reverse together, so the torque does not change sign. The machine runs on AC as happily as on DC — hence universal — and it does it at speeds no line-frequency induction motor can reach, because nothing ties its speed to the supply frequency.

That is why vacuum cleaners, routers, blenders and circular saws sounded the way they did for a century.

What killed it was not a better AC motor. It was the battery. Once a tool runs off a DC pack, the universal motor's one real advantage — running straight off the mains — is worth nothing, and a brushless machine with an electronic drive wins on efficiency, life and noise. Dyson has shipped over a hundred million units of a brushless motor that spins at 125,000 rpm;¹ no commutator ever built would survive an afternoon at that speed.

Used in: corded power tools, older vacuums, small kitchen appliances. A shrinking installed base, displaced one product at a time as things go cordless.

Short versionNo magnets, no rotor connections, nothing to wear. The rotor has to run slower than the field to make any torque at all, and the price of that is heat in the part hardest to cool.

3 · Induction · 1888The squirrel cage

The rotating field made a second kind of machine possible, and it is the one that electrified the twentieth century. The rotor is a stack of steel laminations with conducting bars around the rim, shorted together by a ring at each end — the squirrel cage. Aluminum or copper, cast or fabricated. No magnets, no windings to connect, no brushes, nothing to wear out.

The rotating field sweeps past those bars. A changing field through a closed loop drives a current (Faraday's law), and a current sitting in a magnetic field feels a force. The rotor gets dragged around after the field — the same way a magnet swept under a table drags a coin sitting on top of it, and for the same reason.

Magnet billNone. The most widely built motor in history contains no permanent magnet at all.

3 · Induction · 1888It has to slip

Keep dragging the coin and it never quite catches up. If it did, it would stop feeling anything. Same here: if the rotor ever caught the field, the bars would see no changing flux, no current would flow, and there would be no torque. So it never catches. The gap is slip, usually a few percent at rated load.

Slip is not a defect to be engineered away. It IS the operating principle. But you pay for it: rotor current means rotor heating — current squared times resistance, written I²R — inside a spinning part you cannot easily get a coolant to. Roughly, rotor loss is the slip fraction of the power crossing the airgap.

The torque–speed curve that falls out of this is why these motors start themselves straight across the line, with no drive at all: torque is there from standstill, rises to a breakdown value, then falls off a cliff near synchronous speed.

3 · Induction · 1888What it buys and what it costs

The stator also has to supply magnetizing current to build the field in the first place, because nothing in the rotor brings its own. That current does no work but still has to be carried, which is why induction machines run at a lagging power factor and why they get relatively worse at part load.

• No magnets, no rotor connections, no rotor position sensor needed
• Starts on a fixed-frequency supply; tolerates abuse, heat and overload
• Cheapest torque per dollar in the world, at scale
• Rotor loss is inherent to the principle, so the efficiency ceiling is real
• Lagging power factor; poor part-load efficiency
• Lower torque density than a comparable magnet machine

Used in: pumps, fans, compressors, conveyors, machine tools — and still in traction. BMW put an induction machine on the front axle of its sixth-generation drive,² and Tesla remains the main manufacturer still using them in light vehicles.³

Short versionMake the rotor's field out of current instead of material. You can turn it off, which no magnet lets you do. You pay in rotor copper, rotor heat, and one more thing to break.

4 · Externally excited synchronous · 1890sMake the rotor field with current

Lock a rotor to the rotating field instead of letting it slip and you have a synchronous machine. The original way to give that rotor a field of its own — and it long predates any magnet strong enough to do the job — is to wind DC coils onto the rotor poles and feed them through slip rings and brushes.

The modern name is the externally excited synchronous machine, or EESM. (You will also see wound-rotor synchronous, current-excited, or separately excited. Same machine.) Newer designs replace the slip rings with a rotating transformer, so the excitation crosses to the rotor with no contact at all — ZF has been showing one since 2023.⁶

The useful part is that the field becomes a control variable. Weaken it at high speed for free, strengthen it at low speed for peak torque, and switch it off entirely when the machine is coasting — which kills both drag losses and the risk of the machine generating into a fault. A magnet cannot be switched off. That is the whole pitch.

Magnet billNone. This is the topology vehicle manufacturers — original equipment manufacturers, or OEMs — have actually put into volume production to get rare earths out of the drivetrain. BMW has used it since its fifth-generation drive,⁴ and the Neue Klasse rear axle is the second generation of it.²

4 · Externally excited synchronous · 1890sWhat you give up

• No magnets at all; the field is controllable and switchable
• Excellent high-speed and constant-power behavior
• Inherently safe when unpowered — no voltage at the terminals
• Copper on the rotor means rotor I²R loss, in a place that is hard to cool
• Heavier and larger than a magnet machine of the same output
• The excitation path adds parts: slip rings and brushes, or a rotary transformer and its own converter

While the revival is real and shipping — BMW's first Neue Klasse vehicle entered series production in late 2025⁵ and Renault is targeting 2027 at Cléon⁷ — it is worth being clear about what kind of win it is. Nobody chose this because it makes a better motor. They chose it because the magnet is a supply-chain problem, and this is the topology that solves that problem today rather than in a press release.

Short versionShaped iron and nothing else. Two families: one locks to a rotating field and is now a real product, and one switches its coils a pair at a time and has been about to break through for forty years.

5 · Synchronous reluctance · 1923Saliency without magnets

Go back to the shaped lump of iron and make it properly salient. Cut the rotor into nested curved air slots — flux barriers — so flux runs easily along one axis and is blocked along the other. Feed it from a rotating field and it locks to that field and turns synchronously, running on the sin 2δ term alone. This is the synchronous reluctance machine, or SynRM.

You get the induction motor's rotor cost with none of its rotor loss: no bars, no rotor current, no slip heating. What you give up is line starting. It needs an inverter and a rotor position signal, full stop.

That stopped mattering once efficiency regulation started biting. The International Efficiency classes — IE1 through IE5, set for motors by the International Electrotechnical Commission in IEC 60034-30 — put floors under motor efficiency that an induction motor cannot clear at the top of the range. In the European Union, IE4 has been mandatory from 75 to 200 kW since July 2023.¹⁰ That regulation, not any breakthrough, is what turned a 1923 paper into a catalog item.

Magnet billNone. ABB sells the range from 0.75 to 450 kW at IE5, claiming roughly 40% lower losses than an IE3 induction motor,⁸ and is already marketing an IE6 class ahead of its formal adoption.⁹

5 · Permanent-magnet-assisted SynRM · 2000sA little ferrite in the barriers

Pure reluctance machines have a weak power factor, because every bit of flux has to be paid for with stator current. Drop cheap ferrite into the flux barriers, oriented to oppose the flux on the hard axis, and the power factor improves, torque rises, and the inverter gets smaller — for a magnet costing a small fraction of neodymium and containing no rare earth whatsoever. That is the permanent-magnet-assisted synchronous reluctance machine, PM-assisted SynRM.

It is an honest compromise, and it marks exactly where the magnet-free approach runs out of physics. KSB sells the SuPremE as a magnetless synchronous reluctance motor; its own datasheet notes that the 0.55 and 0.75 kW sizes at 1500 rpm are built with permanent magnets,¹¹ because pure reluctance cannot make the efficiency class in a small frame. That is not a criticism of KSB. It is the physics telling you where the line is, and the company has been straightforward about where it drew it.

Magnet billFerrite only. Cheap, abundant, no rare earth — and roughly a tenth the energy product of sintered neodymium, so it takes volume. Magnet-free versus magnet-light is a good-better-best question here, not a yes-or-no one.

5 · Switched reluctance · 1970sSimplest rotor ever built

Now take away the rotating field as well. A switched reluctance machine, or SRM, has salient teeth on both members — commonly six stator poles and four rotor poles — with concentrated coils wound on the stator teeth. Energize one pair, the nearest rotor pair swings into line, switch to the next pair before it arrives, repeat.

The principle is as old as the earliest electric motors. The practical machine had to wait for power electronics quick enough to do the switching, which is why it dates in any useful sense from the 1970s.

The rotor is a bare stack of laminations. No magnets, no conductors, no insulation, nothing to demagnetize and nothing to melt. It tolerates heat and fault conditions better than anything else on this page, and it is the cheapest rotor you can make.

The bill comes due in every other column. Torque arrives in pulses as each pole pair aligns — like a big single-cylinder engine rather than a turbine — so torque ripple and acoustic noise are intrinsic, not incidental.¹² The drive is not a standard three-phase inverter. And it needs precise rotor position feedback to switch at the right instant, every instant.

5 · Switched reluctance · 1970sPerpetually the motor of the future

So how much volume production is there, after forty years of advocacy? Essentially none. That absence is the finding — more telling than any of the engineering arguments made on either side.

Turntide raised $225 million in 2021 on the thesis that switched reluctance would replace the industrial induction motor.¹⁵ Its technology pages today lead with axial flux motors, and switched reluctance survives in a single heating, ventilation and air conditioning (HVAC) line.¹⁴ It has published nothing describing a change of strategy; the product pages simply changed.

In fairness to the switched reluctance camp — and this is respectful, but firm, disagreement rather than dismissal — the objections are being attacked directly. Honda's venture arm invested in Enedym in late 2025,¹⁶ and modern control with wide-bandgap switches goes straight at the ripple and the noise. But it remains a bet that control electronics can solve a problem built into the geometry, and it has been that same bet for a long time. Switched reluctance has been the motor of the future for about as long as fusion has been the power plant of the future.

Used in: some appliance and HVAC drives, mining and traction niches where robustness beats refinement, and a long list of promising demonstrators.

Short versionA motor that counts. It holds position with no feedback at all, which is its virtue, and it has no idea when it has failed, which is its vice.

6 · Steppers · 1920s–1950sA motor that counts

Take the reluctance idea and cut fine teeth into both members instead of a few big poles. Energize the phases in sequence and the rotor advances one tooth pitch at a time. That is a variable reluctance stepper — no magnet, just teeth — and it is the older of the two forms.

Add a magnet and you get the hybrid stepper, which is what almost everyone means by "stepper" now: an axially magnetized permanent magnet sandwiched between two toothed steel cups, offset by half a tooth, so the magnet biases every tooth and roughly doubles the torque. Typically 200 discrete steps per turn, and far more with microstepping.

Its virtue is that it holds position with no feedback whatsoever. Count the pulses and you know where the shaft is. That is why steppers ran printers, plotters, machine tools, and every desktop 3D printer ever built.

Its vice is the same sentence read backwards. With no feedback, a missed step is silent — the machine carries on confidently wrong — and the motor draws full current standing still, so it is least efficient exactly where it spends most of its life.

Also in this family: shaded-pole induction motors, where a shorted copper ring around part of each pole delays the flux enough to fake a rotating field (cheap, inefficient, and still in millions of fans), and hysteresis motors, which make torque out of the lag in the rotor material's own magnetization loop.

Short versionThe rotor brings its own field, so nothing has to be spent making one. Highest torque density in production — and the entire rare-earth exposure of the drivetrain sits here.

7 · Surface permanent magnet · 1960s onwardMagnets on the outside

Everything so far either paid for its rotor field with current or did without one. Solid-state switching in the 1960s made it possible to commutate a machine electronically instead of mechanically, and sintered neodymium-iron-boron (NdFeB) arriving in the mid-1980s made the resulting magnet rotors genuinely strong rather than merely convenient.

Glue arc-shaped magnets to the rotor surface and the rotor brings its own field. That is the surface permanent magnet machine, SPM. Nothing has to supply magnetizing current, and no rotor current means no rotor heating, so the rotor runs cool and the machine beats an induction motor of the same frame nearly everywhere on the map.

While that sounds like a free lunch, there is a catch in the geometry. The magnets carry the flux, and their permeability is close to that of air, so the rotor looks magnetically identical in every direction. No saliency means no reluctance torque. All of the torque has to come from the magnets, and you have to buy all of it.

Magnet billThe highest of any topology for a given torque. The magnets sit right in the airgap where the field is strongest, and there is no second torque mechanism to share the load.

In practiceMagnet bonding and adhesion have taken up a good deal of my working life, so the thing I will tell you about a surface-magnet rotor is this: the magnet is almost never what fails. The bond line is. It gets thermally cycled, it takes shear from every acceleration and every regen event, and it is asked to hold a hard, brittle block against centrifugal load for fifteen years. Specify the adhesive as carefully as you specify the grade. Most people do not.

Push that far enough and retention stops being a detail and becomes the entire design problem. Formula 1 power units are the extreme case, and — this surprises people — they are surface-magnet machines. The MGU‑K runs somewhere between 50,000 and 100,000 rpm, and the MGU‑H, until the 2026 regulations deleted it, ran well over 100,000 rpm on the turbocharger shaft.²⁹ I have worked on both, and essentially everything interesting about that work is under NDA, so take this as the shape of the problem rather than an account of anybody's solution. At those speeds the magnet is actively trying to leave. The retaining sleeve, the interference fit and the bond line become the machine. For what it is worth, I have built rotors that approached a million rpm. The magnetics were not the hard part.

7 · Surface permanent magnet · 1960s onwardBrushless DC, or permanent magnet synchronous? It is the drive, NOT the motor

Two names describe the same class of machine driven two different ways, and what separates them is the back-EMF — the electromotive force, the voltage the spinning magnet rotor generates back into the stator winding.

If the winding is arranged and the magnets shaped so the back-EMF is roughly trapezoidal, the drive can switch current in six blocks per electrical cycle using only three coarse rotor position signals. That is what the market calls a brushless DC motor, or BLDC. (An unfortunate name, since there is no DC anywhere in the machine. What is meant is that it behaves at its terminals like the brushed DC motor it replaced.)

If the back-EMF is sinusoidal and the drive controls current continuously with a fine position signal, it is a permanent magnet synchronous machine, PMSM, under field-oriented control. Same magnets, same iron, more expensive inverter and sensor — and in exchange, less torque ripple, less noise, and better behavior at speed.

Used in: drones, e-bikes, pumps, fans, cordless tools, hard drives, appliance direct drives. By unit count, the most common motor built today.

7 · Interior permanent magnet · 1990sBury the magnets

Take the flux-barrier rotor from the synchronous reluctance machine and fill some of the barriers with neodymium instead of ferrite. That is the interior permanent magnet machine, IPM — magnets buried inside the rotor, most often as a V under each pole. Three things change at once.

First, steel holds the magnets rather than adhesive alone, which makes the rotor far more forgiving at speed than a bare surface-magnet one. (It is not the answer at the real extremes — those go back to surface magnets inside a high-strength sleeve — but across traction duty it removes a whole class of retention problem.) Second, the steel between and around the magnets shields them from the worst of the stator's reverse field. Third, and most usefully, the rotor is salient again.

The two axes are worth naming. The direct axis (d) runs through the center of a pole, where the path is blocked by magnet material. The quadrature axis (q) runs between poles, through clear steel. The difference between them is reluctance torque you did not have to buy.

7 · Interior permanent magnet · 1990sTwo torques, added

The magnet term peaks at 90°; the reluctance term peaks at 135°. Add them and the total peaks somewhere in between, at a current angle the drive has to find and hold. That is what maximum torque per amp control (MTPA) is doing every millisecond.

The practical consequence is a smaller magnet bill for the same torque. In a well-designed traction IPM a meaningful share of the output — often a quarter to a third — is reluctance torque, which is to say torque you did not buy any neodymium to get.²⁷

Magnet billRoughly 1 to 2 kg of sintered NdFeB in a typical light-vehicle traction motor,¹⁷¹⁸ on the order of 7 to 8 grams per kilowatt.¹⁹ Toyota has gone the other way with the Prius, taking the magnet from about 1.2 kg in 2004 to roughly half that by 2017.¹³

In practiceI got my start in magnetics working out how to erase hard drives and floppy disks, which is an unusual route to an instinct for what a kilogram and a half of neodymium actually amounts to. But here is the conversion that has stuck with me: at roughly two grams of magnet per drive, one traction rotor is about 750 hard drives worth of neodymium.¹⁸ Whether that sounds like a lot depends entirely on which end of the supply chain you are standing at.

7 · Interior permanent magnet · 1990sField weakening and the speed range

A permanent magnet's flux does not switch off. As speed rises, back-EMF rises with it until it hits the inverter's voltage ceiling and the machine can go no faster — unless the drive spends current opposing the magnet.

That is field weakening, and saliency makes it cheap: in an interior permanent magnet machine, the same current that weakens the field also produces reluctance torque, so the money you spend buying speed is not entirely wasted. The result is the wide constant-power speed range traction needs, from a standing start to motorway speed on one gear ratio.

• Highest torque and power density in production
• No rotor losses; best efficiency across a wide operating map
• Mechanically robust rotor; wide constant-power range
• Rare-earth content, with the price and policy exposure that carries
• Needs an inverter and accurate rotor position — it can never start across the line
• Generates whenever it turns, powered or not

7 · The magnet's limitsWhere a magnet machine actually fails

The failure mode that matters is demagnetization, and the thing to understand about it is that the limit is practical, not theoretical. A magnet's Curie temperature is a hard limit of physics — cross it and the material is not magnetic at all. Nobody designs anywhere near it. The limit that binds is the maximum working temperature, which depends on the grade, the shape, and how much reverse field the circuit throws at it, and which is a rule of thumb rather than a law.

A short circuit, a stalled overload or a badly timed fault puts a large reverse field across the magnet at the worst possible moment, which is when it is hot. Intrinsic coercivity — the reverse field a magnet takes before its magnetization flips for good — falls as temperature rises. The margin you designed at room temperature is NOT the margin you have at the top of the duty cycle.

That is why traction grades carry the heavy rare earths dysprosium and terbium, on the order of 50 to 200 grams per vehicle and falling,¹⁷ and why grain boundary diffusion is a commercial issue rather than a process detail: it concentrates the heavy rare earth at the grain boundaries, which is where coercivity is actually set, instead of alloying it through the bulk. Same protection, much less heavy rare earth.²⁰

The second hazard is uncontrolled generation. A magnet machine dragged above base speed with a failed inverter pushes current back into the DC link whether anyone wants it to or not. Magnet-free machines just switch their field off.

Background: what coercivity and the knee of the curve mean in detail is in Reading a B‑H curve — B is flux density, H is applied field strength.

Short versionA magnet you can rewrite while the motor is running. It gets you the wound-field machine's switchable field without paying excitation loss every second you use it — and the price is a magnet soft enough to change, which is the exact opposite of what the last forty years of magnet development has been for.

8 · Variable flux · 2000sA magnet you can rewrite

Every machine in the last section took the magnet as fixed. It is magnetized once at the factory, and after that the design works around a flux it cannot change — spending current to oppose it at speed, and generating into a fault whether anyone wants it to or not.

A variable flux machine refuses that premise. Send a short, large current pulse through the stator winding and you do not merely push against the rotor's magnet. You REWRITE it, leaving it at a new magnetization state that it then holds, for free, until you rewrite it again. Hence the other name the idea goes by: the memory motor.

Line it up against the two machines either side of it and the appeal is obvious. A permanent magnet machine has a field it cannot turn down. A wound-field machine can turn its field down, but pays rotor copper loss for every second it holds the setting. This one pays once, at the moment of the change, and then holds the new setting at no running cost at all.

8 · Variable flux · 2000sThe magnet has to be soft enough to change

Which is where it gets interesting, and where it has mostly stayed stuck.

The magnet needs coercivity inside a narrow window. High enough that the stator's ordinary reaction field — and a fault, and a hot day — leave it alone. Low enough that the inverter can actually produce a field big enough to rewrite it. Miss on the high side and you have an ordinary permanent magnet machine carrying an oversized inverter for nothing. Miss on the low side and the motor reprograms itself the first time something goes wrong.

Sintered neodymium is hopeless here, for the best possible reason: an entire industry has spent forty years making it harder to demagnetize. AlNiCo is the classic candidate, and the joke writes itself — its famously low coercivity, the flaw that made old magnets horseshoe-shaped and lost it the market first to ferrite and then to rare earths, is the exact property this topology wants. Several designs carry two magnets in one rotor: a fixed, high-coercivity one holding baseline flux, and a low-coercivity one that gets rewritten.

• Wide constant-power range without burning current to hold it there
• Field can be written down to nearly nothing: no drag loss while coasting, and no generating into a fault
• Much less rare earth than the equivalent interior permanent magnet machine, and sometimes none
• The inverter must deliver a magnetizing pulse it uses a few times an hour, and gets sized for it permanently
• The property that makes the magnet writable makes it vulnerable — a fault can rewrite it for you
• Control is harder: the machine's own parameters move depending on what was last written
• AlNiCo wants to be long and thin, which fights nearly everything else you want a rotor to be

Magnet billLow, and that is the whole point — but low in a way that leans on a material nobody has been optimizing for decades. The rare-earth content falls. The AlNiCo brings cobalt, which has a supply story of its own.

In practiceI have spent a career reducing oxides to metal, casting it into alloy, pulverizing, pressing, sintering, grinding, gluing and testing what came out — and essentially all of that effort, across the whole industry, has been pointed at making magnets harder to demagnetize. Variable flux asks for the opposite, inside a narrow window, with tight repeatability, from a material that has to survive being rewritten thousands of times without drifting.

That is not a reason to dismiss it. The machine-side argument is genuinely good, and this is the only topology on this page that gets the wound-field benefit without the wound-field bill. But when I look at one, my questions are metallurgical rather than electrical: what is the coercivity distribution across the magnet, not just its average; how far does the written state drift with temperature; and how many write cycles before the material stops landing where it is put. A motor design that assumes a perfectly behaved magnet is a motor design that has not met one.

Short versionSame physics, rotated ninety degrees. The airgap moves to the outer rim where torque is cheapest to make, and manufacturing gets much harder.

9 · Axial flux · 2020s in volumeTurn the machine on its side

Every machine so far sends flux radially across a cylindrical airgap. Rotate the geometry ninety degrees and the flux crosses axially, between flat discs — a pancake instead of a can. The idea is old. Building one in volume is not.

Two things improve immediately. The airgap now sits at the largest radius the machine has, instead of being spread along a cylinder, and torque scales with radius squared rather than radius. And the magnets sit on a flat disc face, which is a far better shape for getting flux across a short gap than an arc glued to a curve.

The usual arrangement is a stator sandwiched between two rotor discs, so the magnet faces work against both sides of the stator and the axial pull on the bearings cancels. (If it does not cancel, you find out quickly.)

9 · Axial flux · 2020s in volumePower density you can measure, manufacturing you cannot fake

While the power density is real and measurable — axial machines reach torque and power densities radial machines struggle to match, in a package a few centimeters thick — almost nothing about building one is standard, and that is where the risk lives.

The stator core cannot be a simple stack of punched laminations. The magnets are large flat blocks under enormous axial attraction during assembly, which is a handling problem long before it is a magnetics problem. And the gap has to be held flat across a spinning disc. When Mercedes-Benz started series production in Berlin in June 2026, it described a process of 98 steps: 65 used for the first time at Mercedes-Benz, and 35 new worldwide.²¹ Read that number again. It is the real story of this topology.

In practiceI have spent time around some of the strongest magnets ever assembled, and the part nobody warns you about is not the physics. It is the logistics. Once a block is magnetized, every fixture, every crate, every tool and every person near it becomes part of the design problem, and a surprising amount of engineering goes into simply getting the thing into and out of the building. Any time a pitch deck treats magnetized assembly as a step rather than as a program, I take that as a signal about how much of it has actually been done.

It is also worth saying that radial flux is not broken. This wheel is being re-invented because of what it buys in packaging, not because the old one was failing.

Magnet billHigh, and rising with power density. Axial flux is a MORE magnet-intensive answer, not a way around the magnet — which is worth sitting with, given that it is the topology that reached a production line while several magnet-free programs slipped.

Short versionThe topology gets picked on supply-chain and regulatory exposure far more often than on torque.

10 · The mapTorque density against magnet content

Plot the families against what they cost in rare earth and what they give back per kilogram. The interesting part of this map is what is not on it: there is nothing in the upper left. Nothing delivers magnet-machine density without magnets, and nothing ever has.

The magnet-free options are all real and all shipping. They are also all bigger, heavier, or harder to drive than the machine they replace. That is the trade, and it is being made for supply-chain reasons rather than engineering ones.

10 · What decides itThe choice is rarely about torque

Electric motor systems used an estimated 53% of the world's electricity in 2023,²³ so a percentage point of efficiency is a national-scale quantity. At the industrial end it is regulation, not performance, that has been reshaping this market.¹⁰

At the traction end, permanent magnet machines are still about 95% of the 2025 light-vehicle motor market,³ and the forecasts have them staying dominant with LESS rare earth in each one rather than being displaced.²² Run the arithmetic and you see why the number matters: at 1.5 kg per vehicle, 20 million electric vehicles a year is about 30,000 metric tons of sintered NdFeB — for traction motors alone, before wind, robotics or consumer electronics take their share.²⁸

So what should you actually price? Not the magnet. Price the combination of export licensing on heavy rare earths²⁴ and the floor price now underwriting Western neodymium capacity²⁵ — a policy-set cost, not a market one, on a component that is a few percent of the motor's mass and a large share of its risk. It is worth noticing that the companies signing long-term magnet supply agreements²⁶ and the companies engineering the magnet out are frequently the same companies. Mostly they are hedging, and mostly that is rational.

And work it backwards, not forwards. A vehicle launching in 2030 is choosing its topology now, because the rotor line, the inverter and the magnet supply agreement all have to be committed years ahead of the first unit. The decision that is due this year is not which motor is best. It is which supply chain you are willing to be married to in 2030.