Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Understanding The Decline In Tesla’s FSD Take Rate

This site may earn commission on affiliate links.


On August 31st, 2021, “Troy Teslike” tweeted data from his long running survey of thousands of Tesla vehicle buyers, showing a low take rate for FSD, Tesla’s “Full Self Driving” software add-on offered since late 2016. For Q2 2021, that survey shows a worldwide FSD take rate of only 11% across all Tesla models, compared to a high of 46% in Q2 2019. Rather than simply assuming this represents a major loss of confidence in FSD, it is worth examining the data and understanding the factors that led to this decline.






First, let’s look at the Q2 2019 spike in FSD take rate. In the span of two quarters, the take rate for the Model S rose from 33% to 81%, and for the Model 3, it rose from 15% to 46%.This can be explained by changes to Tesla’s self-driving packages and reduced pricing. Before February 28th, 2019 (late Q1 2019), Tesla offered Enhanced Autopilot (EAP) and FSD. EAP contained a less-polished version of all current Autopilot features, namely autosteer and auto cruise control, and some of today’s FSD features, including auto lane changes, Navigate on Autopilot, and Summon. FSD, at the time, offered no additional, working features, only the promise that Tesla would later avail buyers of autonomy features as the software improved. It stands to reason that, with no working features, FSD would have had a low take rate.  Then, on February 28th, 2019, Tesla stopped selling EAP, introduced a more basic Autopilot package priced at $3000, and began requiring buyers to purchase FSD for features formerly in EAP.  FSD, now priced at $5000, became a useful addition to the vehicle.

In the same Tesla announcement on that day, in addition to the FSD package receiving new features, Tesla offered sale prices on Autopilot and FSD for new vehicles. This was great for new car buyers and contributed to the spike in FSD take rates, but it had the potential to instantly decrease the resale value of existing owners’ vehicles. In response to backlash from existing owners, Elon Musk tweeted that Tesla would offer a discount on Autopilot and FSD for existing owners:






A table shown to Tesla’s customers at the time (and shared by Troy Teslike) clarifies the statement Elon made:






This substantial deal was a large motivator for many Tesla owners to purchase Autopilot and FSD at a large discount, and for the portion who had previously ordered and received vehicles with EAP, their vehicles were automatically upgraded to FSD at no additional cost. Given the discounts and package changes, the Q2 2019 spike in FSD purchases is understandable.

On a number of occasions over the years, Elon Musk has put forth FSD timelines that have proven to be overly optimistic. Without Elon’s high level of optimism, Tesla and SpaceX would almost certainly not be where they are today. At the same time, this has created unmet expectations among a number of potential FSD adopters. It is understandable that some pessimism surrounding FSD would have contributed to the decline in take rate following Q2 2019.

However, there are less concerning and more practical reasons as to why the take rate has tanked so much in the past two years. Price has been a key factor. Since Q2 2019, the price for FSD has gradually increased, but ultimately, it appears that the larger price difference between Autopilot and FSD, as well as the decrease in the average selling price of Tesla vehicles have been the primary causes of the take rate decline.

To give perspective on the overall price changes for FSD and Autopilot, a graph is provided below showing the FSD, Autopilot, and Enhanced Autopilot price changes over the past several years, along with a graph from Troy Teslike showing the Model S/X, Model 3, and Model Y FSD take rates in North America:

yGV2FpSATZ6gaYn7eCga1O59IbjYWfkOtmgGJUFNNuyM_M78vqs0T3UmBzuFRynDYckpyH3Up_vcmu_m-uARwgN0zEDnzfIV4VwzTIdYFzmKvvHQeg-jnxh9kjgW7AtmcRqVsMlU=s0


9s5wleoes_jcLVLMvOWxg2koCEBGz5cPVzYJB2y4fVuF4X29K6-XZU5lcKc4JG4Hn5CD4hawcqccbS5dBnrwVhsjhlVpfL1auTF7kWkabX4Db9ZYUeR4uKlBWh7KEi1IkePDBo8V=s0


While the take rates have fluctuated a great deal, the combined price of FSD and Autopilot has surprisingly not fluctuated more than $2,000 in the past five years, aside from the early 2019 discounts. Arguably, the total price of FSD and Autopilot has been less significant than the price difference between FSD and Autopilot.

In Q4 2019, Tesla changed the price of Autopilot to $0, meaning its features (traffic aware cruise control and autosteer) were now included with each vehicle. At the time, FSD was priced at $7,000, and the combined cost of FSD and Autopilot dropped by $2,000.  However, this also changed the cost difference between FSD and EAP or Autopilot from $3,000 to $7,000. For many Tesla owners, Autopilot is quite useful and is considered an essential add-on, so when it had a price tag, it was relatively easy for buyers to justify spending $3,000 or $5,000 on it. Then, an additional $3,000 for FSD was not too hard to justify.

However, once Autopilot became standard on all Tesla vehicles, it became much harder to justify an additional $7,000 add-on for features that were considered less essential.  Since then, the price of FSD has further increased from $7,000 to $10,000, meaning there’s now a cost difference of $10,000 between Autopilot and FSD.  This is likely a key reason as to why the take rates for FSD have fallen so significantly.

In addition, it should be noted that the take rate for FSD with Model S/X vehicles has fallen significantly less compared to the Model 3/Y. This can be attributed to the Model 3/Y targeting a more price sensitive demographic. It is much easier to justify purchasing a $10,000 add-on for an $80,000+ vehicle than it is for a $40,000+ vehicle. Of course, Tesla’s sales growth in recent years has come from the Model 3/Y, thus lowering the average sale price.

Further, the global take rates for FSD have been much lower than for North America, as seen in this graph provided by Troy Teslike:

1A84sSD5WBGRHQ3Fct5T2j03tAwdiAADasGIJXjmo2oo7cwvHSzfgJQDI6Hsin3Uu7rbyB0ZrbiKFStxi1zBhGxjV6Zbh9HFc0eTi1kGRPOZfg0EE4yPjOkzF3WBIoIuUiqbICq1=s0


Troy states that “the FSD take rate is very low in China and APAC and they now represent a larger portion of Tesla’s global deliveries which reduces the global take rate.”  Tesla has been delivering increasing numbers in China and Europe. Unfortunately, Autopilot has been crippled in Europe for regulatory reasons, and FSD development has primarily taken place on U.S. roads. All, or virtually all, of the impressive, recent “FSD Beta” YouTube videos have been on North American roads. It is no wonder that take rates are lower overseas.

One might ask whether Tesla should take any steps to increase FSD uptake.

In fact, Tesla has started offering a reasonably affordable $199 a month subscription plan for FSD, allowing owners to try it out for just one month.  In addition, Tesla’s impressive efforts in 3D vision and labeling, neural networks, and accelerated training, as demonstrated at AI Day recently, have shown that the product will speak for itself once “FSD Beta” is polished enough to be released to the public.

The drop in FSD take rate is not as alarming as it might appear. Considering the early 2019 discounts, the increases in FSD marginal cost versus Autopilot, price sensitivity combined with declines in the average sale price of Tesla vehicles, and Autopilot/FSD performance lacking overseas, the relatively low take rate is understandable. It seems likely to trend upward soon.

 
Last edited by a moderator:
With FSD available for $199 a month, it doesn't make sense for most people to pay $10K up front for the possibility of getting a Beta version. According to what I have read, if you get the monthly subscription, you can still request the FSD Beta. If you get it and you like it, you can continue with the $199 a month. If you don't want to be a Beta or don't like it, you can stop paying. So the take rate at $10K may not be a good indication of how many people will sign up to try the Beta.

Notice, the take rate for Model S and Model X owners is much higher, about 60% and holding steady, after a similar, but higher peak in 2019. Clearly, those purchasers are not so price sensitive.

My new Model S LR, which I ordered with FSD arrived Saturday, and I am currently, as I type this, having the car updated to the new software version so I can request FSD Beta. BTW, I took out my purchase sheet from my first Model S which arrived in February 2013. It has almost exactly the same purchase price (before rebates) as my new Model S. The new one is $2K more, but includes the $10K for the FSD. My old Model S was also an 85, and had most all the features available, with 19 inch tires, and the rear facing seats. According to inflation tables, the Model S should be 17% more expensive. My old Model S came with 265 miles EPA range, while the new one has 405.
 
I imagine that when FSD is determined to save lives, compared to manual driving, the Govt will put pressure on for it to become standard.

i disagree on the basis of human psychology. Humans don’t do relative risk very well. Driving is dangerous and a computer is probably safer. But people have the illusion of control on the roads and don’t want to give that up. Even if the computer crashes 100 times less than a person, it still crashes and people just don’t like the idea of being hurtled to their death by a machine with little ability to change that.

So, it’s not just that it has to be safer; it must be nearly perfect before the gov mandates it. Arguably autopilot is already safer, but the gov is considering restricting it, not mandating it. And that’s because we just don’t like the idea of a machine in control of our fate.
I think one thing that would help adoption is if the government changes the law to provide for strict liability to the car company when a vehicle crashes while under control of their hardware and software. Tesla sidesteps that with their warning, but my view is that we need to shift liability to the car companies because the know their systems lull drivers into a false sense of security.
 
  • Like
Reactions: pilotSteve
The simple beauty of TACC is that your can rely on the distance to stay safe within your settings. We trust it. It does not lull us into lack of control.
Of course you must supervise, near all automatic operations of any process need eventual human supervision. More so when life and death machines like autos are involved. We once had AP try to steer us into a left turning vehicle at a light head-on several updates ago. Scared us big time. My hand was on the wheel and it was necessary right away.
Mandates are for dictating policy, America may not usually go for that. Think how long it took for the seat belt to become "mandatory".
 
The simple beauty of TACC is that your can rely on the distance to stay safe within your settings. We trust it. It does not lull us into lack of control.
Of course you must supervise, near all automatic operations of any process need eventual human supervision. More so when life and death machines like autos are involved. We once had AP try to steer us into a left turning vehicle at a light head-on several updates ago. Scared us big time. My hand was on the wheel and it was necessary right away.
Mandates are for dictating policy, America may not usually go for that. Think how long it took for the seat belt to become "mandatory".
So wait, you trust AP even though it tried to steer you into an oncoming car? I only trust it on wide open highways and in stop and go traffic when you aren't going fast enough to do much damage. I would never, ever trust it on roads with turning cars or traffic lights, and certainly not in New Jersey or NYC. The NJ turnpike, NYS Thruway, Garden State Pkwy do NOT qualify as wide open roads. I-90 in South Dakota is wide open.
 
  • Like
Reactions: dyau
The simple beauty of TACC is that your can rely on the distance to stay safe within your settings. We trust it. It does not lull us into lack of control.

Elon has previously mentioned that the majority of crashes while on Autopilot happen to experienced Tesla drivers, not new drivers. This indeed confirms that we do get complacent. New AP drivers are terrified and pay close attention. Those of us that have been driving it for a long time realize that it's pretty damn good, and take the opportunity to look away, check our phone, or generally do what we're not supposed to be doing. And that's when AP sends you into a wall.

It's the experienced drivers who crash. And they crash because they are lulled into a false sense of security. Thus, that's why the system needs to be 99.999999% reliable before the system should be allowed to have full control.
 
Once FSD is out of beta, the take rate will probably go up.
I think it's more nerve-wracking to have to babysit the thing, watching it ever so vigilantly knowing it's going to stuff it up and you will have to intervene.
Standard AP is proven (ok, mostly, as long as you're not near a firetruck, and in that case, you should be taking over control anyway!)
 
Elon has previously mentioned that the majority of crashes while on Autopilot happen to experienced Tesla drivers, not new drivers. This indeed confirms that we do get complacent. New AP drivers are terrified and pay close attention. Those of us that have been driving it for a long time realize that it's pretty damn good, and take the opportunity to look away, check our phone, or generally do what we're not supposed to be doing. And that's when AP sends you into a wall.

It's the experienced drivers who crash. And they crash because they are lulled into a false sense of security. Thus, that's why the system needs to be 99.999999% reliable before the system should be allowed to have full control.
Maybe. I try to be skeptical of generalizations, but there may be something to your observation. On the other hand, having used AP1 for 6 years and having read hundreds of these posts, I am more aware of the limitations of AP, so while I might be more relaxed about it in general, I also know a good deal more about the limitations and foibles. I find myself both more willing to rely on it when I think I can, and also quite willing to disconnect and drive when in a situation I think it will not handle particularly well.
Also, I often think back to one of the very first trips I took immediately after AP was first released in November 2015. I drove up I-95 from the Boston area to Camden, Maine to visit a friend. I was yacking on the phone (hands free) to another friend as the car passed my intended exit in Portland, where I had planned to head east along US 1. The car kept on going up the highway for some miles before I noticed I had missed the exit. (In those days, there was no "nag," so you could let the car drive as long as you wanted.) Fortunately, it was a lovely day, dry and clear, and a weekday with very light traffic. Apparently there was no lane drop or other decision point for the car as I passed the exit, so it kept going merrily along. I had probably only driven with it for a few hundred miles at most by then was probably a bit over-confident in the system, but there was no adverse consequence from this situation, which reinforced my positive feelings about its performance. I guess the point is that newbies can also get over-confident if they have error-free early experiences. It takes a period of driving to experience enough "corner cases" to develop your mental inventory of what seems safe and what seems riskier.
 
  • Like
Reactions: IslandRoadster
When I tried using AP, just for demo purposed, in Brooklyn it first turned itself off completely, along with everything else when it saw something that I didn't see because it wasn't there, and one time did phantom braking, I think due to multiple overpasses. Then I tried it on a more open highway, the NYS Thruway/I-87 and it also did something unexpected in light traffic. I wasn't impressed and now only use it like I said above: bumper to bumper stop and go traffic, and truly wide open highways.
 
I think another factor, which is hard to capture with data, is that the people buying Teslas now are not, in large part, fan-boys like those of us that comment in the forums. They're car buyers, who just want a good car. Many FSD purchasers before were people who wanted to help the mission, to push electrification and the concept of FSD forward. We considered it an investment.

Now, the average car buyer doesn't want to invest in Tesla. They just want a nice car and they want something for their money. When there's *actual* full self-driving, they'll buy it. But for now, they're just not going to loan Tesla money. Early FSD adopters were willing to make that loan.

I agree with you.
There is also the risk that it will not become available at all within the ownership period and then having to sell the car with FSD but only getting a smaller portion of what was paid (and not used).
Also the price of the car and the price of FSD both increased a lot, probably making it too expensive for most people who are already getting into a more expensive purchase than they would have made for an ICE car.
 
Lull into false sense of security & distracted vs. those who are supervising the computer.
Now some features like TACC have never once made an error in our MY. Yes supervision is required anyway. Yes we do trust.
AP always makes some errors. Again, yes supervision is required.
Today the level of AP is not going into the back for a nap and wake up at the destination. Not yet.
When the car folds up into a briefcase let me know too please.
 
And I’m sure Tesla would love to still be able to charge for standard autopilot. The fact is, many other companies have implemented auto lane keeping and auto lane change in vehicles that cost substantially less.
The auto lane keeping/adaptive cruise control/vehicle sensing system in my $23 Hyundai works as well as AP. That suite of functionality is quite common now.
 
  • Like
Reactions: glide
I think the only logical conclusion is that Tesla doesn't want us to buy FSD ......yet.

I don't have a crystal ball but based on this data they would sell roughly 4X the product at 5-6K. For software which costs 0 dollars to reproduce this is all the math you need. At 500K cars/year that is roughly 1/2 Billion dollars a year being left on the table.

So why don't they want us to buy it? A. they fear that too many FSD cars will expose its limitations and create liability or brand damage. B. They want to sell it to us for 20K when it really works well.
 
  • Like
Reactions: IslandRoadster
Well my crystal ball tells me they already know that A, FSD will have issues and limitations, and B, it will never fully get there. As I've posted elsewhere, AI or FSD will be able to do only what you tell it to do, or program it to do. Even the "deep learning" and "neural networks" that are being used are just programming. They don't know everything, nor do they learn. They will add to the knowledge base of what they've encountered but that doesn't mean they know how to react. They will learn not to crash into an emergency vehicle, but didn't it take human intervention to learn that?

AI and FSD will always be "just around the corner, in two weeks".
 
The numbers in the crowd-sourced data may not be perfectly accurate, but I believe they tell a true story in that the actual take rate for FSD is low, and that the take rate has decreased over time.

The reasons? I think the primary reason is that it doesn't actually work. No one is going to buy something, especially something that's $10K or $200/month that doesn't actually work. That's why I haven't bought it, haven't considered it, and am waiting on a new Model X that doesn't have it.

Someone above contemplated that the take rate may go up once FSD is "out of beta". This made me laugh.

"Beta", in software development, refers to a stage of development of a nearly-finished product, where the product is released to a limited number of users, who are not affiliated with the development of the product, and who have agreed to test the product in a non-production environment, and submit feedback to the developers for additional refinements prior to release.

This is not what Tesla uses the "Beta" label for. Tesla has not adhered to anything about the rules of Beta software. FSD, EAP, even the auto wipers are released to everyone, they are being used in a production environment, and Tesla is neither soliciting nor accepting feedback from the use of those features. That is not Beta. That is Released software.

Tesla's use of "Beta" = "sh!t that doesn't work yet". It's their excuse for half-baked, worthless, hyped software.

I don't have FSD on my two Model 3's, I have EAP. EAP is crap. It doesn't actually work, and never has. If I don't intervene at least once every 10 minutes, it would at best miss freeway exits or at worst crash the vehicle. But that's apparently OK, because it's "Beta". 🙄 If you can't make EAP work, then by definition you cannot make FSD work because FSD is a superset of EAP.

You want the take rate to go up? Put out something that works.
 
In October 2020 Elon claimed FSD will be finished by end of 2020, and he'll have a fleet of robo-taxis on the road by the end of 2020 (so 2-3 months later). Ok, so let's give him some margin for COVID delays, instead of 2-3 months, let's give him a year. Sounds fair, no?

So, if FSD will really be finished by end of 2021, and FSD will be able to drive people without a driver behind the wheel, then I'm sure the FSD take rate will skyrocket! Heck, if I can summon my car anywhere within its battery range without supervision (i.e. Tesla takes responsibility for any damage it causes), I'm upgrading the Tesla I have which Elon sold me as FSD capable, canceling my Taycan order and trading the other Tesla (with only AP1) for another Tesla with FSD. Of course if FSD is not finished in 6 times the time Elon predicted, well, then I'm paying Tesla nothing at all and buying that Taycan instead. I suspect the FSD take rate will continue to tank too - how long can a buy cry wolf (hint, Elon's been crying wolf 5 years now) before people stop believing it.

So, if you are a believer in what Elon says, FSD take rates are about to take off, go buy TSLA stock now and maybe even some Tesla cars, they will be worth over $200K each next year (again, if you believe Elon). If however you think Elon is full of hot air and hype, well, the FSD take rate should be no surprise to you.
 
"Understanding the decline in Tesla's FSD take rate"

They decided to charge $10k for it, when many owners are hardly even aware of what Autopilot is - or even that THAT is included with the car.

The only people ticking the $10k FSD box are people that don't need to look at price tags.

I bought in at $5k for EAP + $3k for the promise of something more later & being part of its development. $5k for EAP was a reasonable addition for all it offered (which also meant AP itself, which wasn't standard yet). $3k was a stretch at the time to buy something that didn't really exist, but I REALLY wanted to be at that bleeding edge when it did happen.

Today, you look around non-Tesla platforms of discussion, and you'd be lucky to find any comments from people that actually understand AP, let alone FSD.

I think Elon's "rightness" is half true. He doesn't overestimate it, but underestimates how long it'll take to be perfect enough for general consumption or "Robotaxis" (I say a couple years, others say decade+, Elon thinks in months) - and is completely ignoring the snowballing problem of the information vacuum that Tesla's lack of messaging or advertising is causing. I don't even think Tesla realizes how many owners aren't aware of what AP/FSD is. That's scary, and I think the problem is sneaking up on him... but it'll be ignored as long as Tesla keeps its 6-month waiting list up and sales keep amplifying. We'll see, I guess.

tl;dr: FSD is severely misunderstood and the take rate is likely indeed very, very low.
 
  • Like
Reactions: pilotSteve
tl;dr: FSD is severely misunderstood and the take rate is likely indeed very, very low.
I think Elon knows he's selling empty promises and justifies it to himself as the cost of some bigger mission. Some will argue that Elon really thought AP2.0 was capable of summon from New York to L.A. - there by the way he didn't miss the time estimate, the AP2.0 hardware was never capable of it. However, I doubt Elon really didn't know that things like redundancy (which AP2.0 did not have) are necessary for self driving by itself across a continent, or while you're sleeping in the back seat. Want an even more direct evidence? Elon sold P85D as a 700hp car, Tesla advertised it as 691hp motor power, but it took them 2 years to admit the battery had a pyro fuse which would blow if the motors produced more than ~480hp . He claims to be an architect and an engineer, so obviously he can grasp a concept of a fuse, no? If your argument is that "but Tesla only said motors are capable, not the battery", then the argument for FSD is the same - your floor mats are FSD capable, your windshield, seat, doors, screens, etc. The car will never drive itself, but hey, it's mostly Full Self Driving capable.
 
I think a lot like Elon's described his process (but definitely not as ... motivated 🥱😴 that guy is superhuman) - seeing the fundamentals of a system as building blocks, seeing what can be possible given the physical limits of something. So no, I don't think he's ever straight lied - knowing the truth but selling more than it. But I do also know that Tesla management themselves often watches Elon's twitter and gets direction (and motivation) from there, insane as though it may be. He tweets as he thinks, then just kinda... makes sure it happens as close to his promises as possible. These promises of "2 weeks" are probably seen only later at Tesla among the AP team - like "whoa. what? crap, okay we'll try".

So it kinda made sense with AP2.0's capabilities and setup, that it should be able to. It can already see more than a human can, at all angles at all times, so why couldn't it? Look at how much processing power it has! It has all the inputs, why couldn't it?

The convergence with implementation is the hardest part. If only roads were void of other drivers...

You know, back in 2012 I was pretty sure, given the building blocks, that a legitimate electric fighter jet wouldn't be too far off. I really still do want to see an electric F-16. 😂 Even now, working in the industry, along side people designing inverters and motors and battery systems... I still don't think it's too far off. Would I be surprised by the limitations when the thing starts to become more fully formed? Yeah, probably. But not outside the realm of possibility if you really push what's possible ;)