Summary
YASA just announced a prototype axial-flux motor with an unprecedented 59 kW/kg peak power density. That’s roughly triple the power density of today’s top EV motors. In this post,
I take a quick engineering look at what’s behind this number — and what people usually misunderstand about axial-flux designs.
Introduction
Apologies for the slightly click-bait title — it’s the daylight savings fog talking. This post presents some quick thoughts on the new (2025-10) YASA record-breaker axial flux motor, and what everybody gets wrong about AFMS.
Recently, the UK-based axial-flux technology company YASA announced breaking their own unofficial power density record, with a peak power density of rather mind-blowing 59 kW/kg. Of course, its their claim and not independently verified, but I do believe the best of people. And, it’s peak rather than continuous.
Nonetheless, this is — frankly — insanely high. Perplexingly high.
Automotive motors typically reach 3–10 kW/kg continuous, and 15–20 kW/kg peak. A dragster running on nitromethane is about 30 kW/kg. The freaking space shuttle turbopump 153 kW/kg, so at least they haven’t yet beaten THAT.
The highest I have even been able to go – in simulations – was about 50 kW/kg and that was active mass only, and that required magic technology to manufacture hollow Litz wires with coolant running inside them. Would have reached a very high efficiency of around 96-97 % though.
So obviously, I had to take a pick at the record, based on what little published information was available, and see what this monster of a machine might have eaten.
What YASA Does
YASA stands for Yokeless and Segmented Armature, a nowadays-common type of an axial flux electric motor. At its simplest, it consists of a single stator sandwiched between two rotors. Importantly, the stator has no back-iron or yoke. Instead, it consists of separate (hence the word segmented) pole-pieces or teeth, with a coil around each of them (or every second one).
To make this work, the main flux path through the stator is (ideally) purely axial, i.e. there is no significant circumferential (or azimuthal) flux component.
To pick on some nits, you can make a YASA-style machine with a completely ironless stator, also. Many PCB-stator designs go this route. Granted, your stator can’t be very long/thick; i.e. its axial
dimension has to be rather small, lest your flux densities drop very low. But, for lowish-torque-density applications it can make sense. Besides, you rid yourself of the cogging torque component (which
stemps from the interaction of the PM flux and the stator iron) altogether, which can be a plus, plus you obviously have more azimuthal space for your winding conductors now. But I digress.
What Everybody Gets Wrong
Now, there’s been loooots of talk and lots of hype over axial flux machines in the recent years, with lots of written on the topic. So far, I’ve yet to read a piece of them that would actually
properly explain their benefits, to my tastes at least.
What is often quoted about the yokeless-stator machines is their, well, yokelessness. This is supposed to rid you of the mass of the yoke, and the iron losses in the yoke. And it does – THERE IS NO YOKE there, duh.
But, this is not the main or the most significant benefit, in my honest opinion.
The second thing often touted about AFMs in general is their high average airgap diameter, compared to radial flux machines of a similar OD. This is also something that I kind of disagree with. The textbook design approach of an axial flux machine sets its inner active diameter to around 58 percent of its outer diameter, giving an average diameter of 79.9 percent of the outer active diameter.
This is highish, yes, but not outlandish. A PM machine with a few more poles than usual, and either good cooling or a large diameter would easily reach similar numbers. Furthermore, an AFM must accommodate for the end-turns beyond the outer radius of the core (unless a more three-dimensional flux path with soft-magnetic pole shoes is used), further eating away this benefit.
What Actually Works
What really sets axial-flux topologies apart, in my opinion, is how you can scale their slot “depth” (now the axial dimension of the slot) without incurring any nonlinear penalty anywhere else. In radial-flux machines, deepening a slot means decreasing the airgap diameter which decreases the torque available per unit copper loss, for instance. In an axial topology, increasing the slot dimension only makes the machine correspondingly longer.
What this allows you to do is easily reach a very high electric loading – the number of Ampere-turns per meter of airgap circumference. As the torque produced is essentially the product of the electric
loading and the PM-induced airgap flux density, THIS is what gets you a high torque.
Talking about the electric loading also brings us to the true (again, in my opinion) defining benefit of YASA-style machines. The thing is, the electric loading can’t be arbitrarily high, either. Increase it,
and you increase saturation and decrease torque-per-Ampere, leakage fluxes, decrease the power factor, and risk demagnetizing your magnets.
What the YASA trick – having the flux flow through the stator axially rather than azimuthally – essentially splits your machine into two.
Compared to a machine with a single rotor and a single yoked stator, a YASA machine can produce the same torque with half the electric loading per airgap. It has two caps, after all. Granted, you pay
the price (both mass and material) of having two rotors, but at least on the mass-budget side this is at least partially compensated by the lack of the stator yoke.
In short, it’s not about the lack of yoke or slightly higher average airgap diameter — it’s about how axial geometry lets you scale electric loading without geometric penalties.
Having sorted that one out, let’s jump into actual analysis.
Teardown
With the theory and common misconceptions out of the way, let’s get quantitative and see how plausible that 59 kW/kg really is. Or rather, what is needed to actually reach that, based on my understanding. Since we don’t have published mechanical drawings, I estimated dimensions from press photos and typical YASA proportions. These are back-of-envelope numbers, but enough to test plausibility, in
my opinion.
What is actually known
Full disclosure time: I didn’t dig very deep. I took a look at the press release linked earlier; its numbers and pics. I didn’t dig any deeper to see if any numbers had been published earlier. I am tired, and writing blog posts only pays in terms of contributing to my marketing funnel of questionable efficiency.
Now, I began my analysis from this picture here, taken from YASA’s press release.
Assuming the gent holding the motor is not a veritable giant, I give the motor an external OD of perhaps 25 or 30 centimeters. Leaving some quite ample space for the end-turns, let’s settle for a core
diameter of 20 cm. For the overall active (axial) length ignoring the housing, let’s say we want to stay in the 70-80 mm region or thereabouts.
Now, let’s assume that the tip speed of the rotor can reach a speed of 80 meters per second, and OH BOY DOES THIS SOUND LIKE ALL KINDS OF MECHANICAL FUN. But, just for the giggles, let’s run with it.
This gives us an rpm of 7600 – not too bad.
YASA machines commonly have a concentrated winding pattern, so to ease the load on our minds, let’s go with the 10/12 scheme, or more like its multiples. We quite certainly do want a multiple, as the kind of
torque densities we are speaking about here call for limiting the armature reaction, or the stator-induced magnetic field, as much as possible to avoid saturating the machine to oblivion and killing the magnets.
Let’s go with 30 poles and 36 slots.
As for the magnets, let’s go with a rather plain N48SH. The quoted peak power is not continuous, so let’s make our life easier by assuming we start with a coldish rotor. Let’s go with 50 deg here.
Finally, for the winding, let’s assume a copper fill factor of 75 percent – a rather feasible number considering the manufacturability aspects of this topology, and still leaving a few aces up our collective
sleeves – more of that later.
First attempt
Now, let’s begin the actual design or teardown process. I will be using EMDtool here, and a 2D model with one or more slices where needed.
Now, based on the initial guesses, and throwing somewhat typical guesses for everything that’s missing, we get the following geometry
where only one axial half is shown. The total active length is 6.4 cm, and the active mass is 7 kg.
Next, we compute the power vs current density curve at our assumed rpm and a pole angle of 90 degrees, up to 45 Amps per millimeter squared…..and fall short by a factor of 3.
We only reached about 250 kW instead of 750 kW.
Not nice.
We could increase the current density even further, but we see from the curvature that the machine is getting heavily saturated even now.
Second attempt
Remember, it’s essentially the electric loading and its cousin the magnetomotive force that is eating our lunch via saturation here. We can’t really influence the MMF alone apart from by increasing the pole
count, and our electrical frequency is almost 2 kHz already.
To influence the electric loading and mmf together we can apply the YASA trick for a second time. Remember, the YASA topology halves the electric loading by doubling the number of airgaps. Since we
have an axial topology, there’s absolutely nothing to stop us from increasing the gap number even further, apart from perhaps the common sense considering ever actually building this machine.
We could just stack multiple shorter 2-rotor machines together. Or we can keep our two existing yoked rotors at the ends of the machine, but add more stators and yokeless rotors in between, all with
a purely-axial flux path. This has the benefit – at least a theoretical benefit – of saving us some rotor mass.
Let’s go with this configuration now, in axial order: a yoked rotor – a yokeless stator – a yokeless rotor – a yoked stator – a yoked rotor. Correspondingly, let us approximately halve length of each stator.
One axial half of this topology has been visualized below. Please ignore the fact that the inner (lower) rotor has a thin yoke – I was too busy to implement a new geometry template, so I simply created
a yoked rotor with a very thin yoke.
Let’s again simulate the power curve, and let our current density extend all the way to 90 Arms per mmsq.
We’re getting closer! We’re almost reaching the power target, and it visually it seems we are not saturating as heavily despite having doubled the current density.
Third attempt
Now, let’s take things further and stack two similar machines back-to-back. We could simply add more yokeless rotors and stators in the middle, but I’d again have to work on some EMDtool templates
to be able to simulate this.
So two machines it is – let us just reduce the stator length, and increase the number of axial symmetry sectors to 4, and simulate away.
That did the trick! The active mass is now 11.9 kg, and assuming the housing is a cylindrical envelope around the motor, of 5 mm thick aluminum, we get 13.3 kg for the total mass. A bit over the YASA
number, but we could certainly optimize this further.
So, there you have it, a re-design of the YASA record-breaker. Took me maybe 60 or 90 minutes to design. I don’t see where all the fuzz is. Now off you go and just build it.
I’m kidding, obviously.
But, in essence, it seems that stacking multiple yokeless stages is the only clear path to that power density — and even then, thermal management becomes the ultimate bottleneck.
Nice picz
To just show off, let’s run a multi-slice model, with solid conductors modelled, and include some nice plots.
In terms of performance, we get 62 kW of DC losses in the winding, 25 kW of AC, and an efficiency of 84 %. That number includes a whopping 34 kW of losses in the all-solid polegaps – a better design would either segment them or use non-conductive material there. On the positive side, the magnet losses are only 1 kW or so in total, thanks to a gnarly 2 mm segmentation.
Thoughts on cooling
Now, the elephant in the room – the 90 elephants per square millimiter, to be be exact. This is quite high indeed. Perhaps absurdly high if you come from a more-industrial background and are used
to things running for 30 years continuously at 5 Amps per mmsq.
Now, YASA does not explicitly state how long the machine could run at its peak current density. They do estimate the continuous power being around half of the peak number. Naiively, this would bring
the continuous current density down to 45 A/mmsq, or a bit higher when you realize the rotor would also run hotter, bringing the PM flux down. Let’s call it 55 A/mmsq.
This is high. Granted, further optimizing the design could bring it down somewhat, but unless I have completely missed the mark somehow, that should be order of magnitude. Unless their tip speed is way
higher, which seems even more fun than our assumed 80 m/s.
To cool stuff in this neighbourhood, I assume a flooded stator, with oil running inside a plastic or composite enclosure. This is a common high-end solution, and often the second-best cooling approach after hollow conductors.
My first idea would be to inject oil at the OD of every other slot, let it run radially
inwards, and circle back through every second slot. The YASA topology allows each stator tooth to be wound separately, so I envision its well within the realm of possibility to leave a small gap between each
conductor, allowing them to be cooled from three sides each.
This could well cool the active parts of the conductors. The end-windings at the OD could be more difficult, unless we use suitable nozzles for injecting the coolant (I think jet impingement is the term) with
a high enough level of turbulence.
Or, if the cross-sectional area of the end-windings is simply larger than that in the active parts, reducing the loss density in those areas. Everything is possible in the era of 3D-printed copper, and even before there
was an aircraft starter-generator concept that used custom laser-cut copper to achieve the same trick.
Now, this still leaves the peak power somewhat problematic. Assuming we can extract the losses that 55 A/mmsq incurs, the peak-power 90 A/mmsq still leaves a metric frak-ton to dumped (mostly) into the
specific heat of copper. Unless I did an error in my quick calculations, that would translate in a temperature derivative of 30 degs Celcius per second.
On Simulations vs Reality
These results were simulated, quickly, in EMDtool. “Only a simulation” gets thrown around a lot, and most often in a negative context.
Being a simulations-oriented person myself, I find simulated results get crapped on far too often and for reasons not that well justified. A well-set-up electromagnetic simulation is rather accurate.
Full stop.
Where folks go wrong is either by setting up simulations that do not actually capture reality too well. PWM effects and eddy currents are a common example.
Alternatively, they do actually set up a very good simulation, that then cannot be built, or that cannot be built within the budget.
Summary
Recently, the company YASA claimed to have achieved a peak power density of 59 kW / kg of total mass. This post evaluated the claim using a single press photo together with some guesstimated dimensions. Based on the analysis, the claim does indeed seem plausible, but really impressive nonetheless. Based on the results, stacking multiple stators and rotors on the same shaft seems to be the most likely approach behind the results.
Check out EMDtool - Electric Motor Design toolbox for Matlab.
Need help with electric motor design or design software? Let's get in touch - satisfaction guaranteed!
