Ark Invest's 2025 Tesla Model

[hershey101]

The team at Ark Invest published their latest model for $TSLA outlining a case for a $3000 price target by 2025. That would be a 4.5x increase from the ~$650/share that $TSLA closed at on Friday.

I haven't gone through the entire model and validated their assumptions, but some preliminary readings tell me it's quite a bit more farfetched then the previous 2024 model they had with a ($7k pre-split) $1400 price target.

I'll update this thread with my modeling assumptions and where I differ substantially from Ark.


You can find the model here: ARKInvest/ARK-Invest-Tesla-Valuation-Model
and the corresponding post on Ark's website here: ARK’s Price Target for Tesla in 2025 is $3,000 Per Share.

Update:
I fixed some of their assumptions and was able to derive a $1200 price target.


Be careful with the monte-carlo simulations, there is a bug in the spreadsheet where only the first simulation is updated based on parameters therefore all other simulation values remain the same and your mean statistics (price targets) are never updated.

Here is my valuation assumptions:

This is what Ark's model had initially:

For historical perspective, this is what they had on the original model they released a few years ago:

 

[jeewee3000]

I support healthy debate around ARK's lofty price targets.

If I understand correctly your valuation is based on the opinion that :
- Tesla will not have reached L5 autonomy;
or
- Tesla will not have monetized L5 autonomy. (i.e. Tesla Network)


While I'm skeptical that the Tesla "robotaxi" Network will be active and be generating revenue in 2021/2022 (as some very bullish Youtubers believe), I think 4 more years in the space is enough time for the Tesla Network to activate. (My personal 'bet' is during 2022 L5 will be achieved. DOJO v1.0 should activate around september 2021 and from then on it's off to the races I believe. But hey, nobody knows).

Therefore, my valuation for TSLA around 2025 is substantially higher than $1200. It all depends on FSD of course. FSD = money printer. (since all existing Tesla vehicles can become autonomous overnight)

I won't derail your thread further with FSD-talk, but just wanted to post my observation that the huge difference in price targets between you and ARK lies in the recognition of FSD.

 

[hershey101]

I support healthy debate around ARK's lofty price targets.

If I understand correctly your valuation is based on the opinion that :
- Tesla will not have reached L5 autonomy;

This...

I've spent 4 years as a CV research engineer in self-driving. I know what to look for in the car's performance. I'll admit that Tesla has done an extremely good job with a vision-only system. Way better than what most of my colleauges thought would be possible. However, they have huge tail-end errors. Now: They can do funny math to convince regulators (who are relatively stupid) that their system is 10x safer than a human, and it might work. But from a fundamental technology perspective, they will not be there in 5 years. The tail end is EXTREMELY hard to resolve. And while I'm smart enough not to bet _against_ Musk, I'm also not going to give it a 50% probability like ARK haha.
I give it a ~5% probability of success by 2025, and if I move it out to 2030, I'd be willing to bump that % to 15%-20% as the research evolves.

I agree with you FSD = money printer (IF IT WORKS). But % of it working is NOT 50%. Well maybe 50% if you restrict it to one block in sunny CA.. But not the entire US across all weather conditions. Scaling is hard... Elon knows this, he's said it many times... Just because you are scaling a software-based solution instead of hardware (cars) doesn't mean scaling is easy... Scaling is hard period.

Another thing to note is: You don't need self-driving or obscene margins (60% like ARK's model hahaha) to get a reasonable valuation target. A simple EV + energy model going out to 2030 gives you a $3T target (~$2200 with 2.5% yearly dilution, higher on the front half, and buy-backs on the latter half). This is already substantial returns (~17.5% YoY for a decade, or 15% post-inflation, well above the 7.5% historical of $SPX). You still get free call options on: Dojo as a service, self driving (which in my model doubles the valuation to $6T) and you can easily take that strategy and turn it into 25%+ returns with agressive options strategies.

So there are several problems IMO with Ark's modeling:
- They are making up crazy #s to justify obscene targets when they simply don't need to. They could have published a rational valuation model, gone to 2028 with the same #s I'm using for 2030, gotten 20%+ YoY, and not been laughed at.
- They aren't valuing over $2T+ in revenues: Solar, cloud compute.
- They are over-emphasising things that don't even deliver that much to the Tesla bottom line and yet end up leaving themselves open to redicule: eg. Insurance. This is not a HUGE business, even in ARK's own model, if you put insurance to 0 you don't take a huge hit to the pricing. Yet the way they've modeled it shows they don't understand the insurance business.

Piper Sandler is also another rediculous note: they went out to 2040 on their latest 100+ pg. research paper to come to a $1500+ valuation

It looks like we are due for another 2-3 year of lul as valuation catches up to prices, and the fake-news of fintweet leads to a vally of disappointment, IV comes down substantially, and we can repeat the same strategy we did back in 2018-2019 of scooping up cheap LEAPs with IV <45%.

 

[Buckminster]

I've had a chance to sift through the ARK model and wanted to share some of my reactions. Many of these things have been called out on Twitter, some of these comments here as well, so may not be groundbreaking new information for many here. That said, hoping that synthesizing these in to a single post may be of use to many in this community. For reference:

• ARK's Price Model: ARK’s Price Target for Tesla in 2025 is $3,000 Per Share
• ARK's GitHub Data Model (includes their old models as well): ARKInvest/ARK-Invest-Tesla-Valuation-Model

TL;DR: Tesla has immense optionality in its business and the permutations of outcomes that justify a $3,000+ valuation by 2025 are equally immense. It may never achieve autonomy nor launch a ride hail network, but it may still hit that price target if it delivers 5M+ units at strong margins and continues growing at 50% CAGR through 2030. Importantly, that growth trajectory will demand a higher enterprise multiple than what ARK has currently ascribed to Tesla in 2025. The ARK model ignores HUGE parts of the business which would just add more support levels for future valuations. TSLA is the next TSLA and this truly still is just the beginning. I look forward to seeing the Machine that Builds the Machine coming to scale and watching the markets realize the level of operating margin and operational execution that Tesla has actually achieved.

---

Below are my observations and comments on some specific assumptions or model details that I personally found interesting:

1. The model assumes MAX 25% market share in any individual segment, which seems entirely reasonable. The one aggressive assumption is the addition of M-2/A and a "Neighborhood EV" ("N-EV") which both have extremely large addressable markets (40M and 100M respectively). In other words, a lot of the extremely high price scenarios have Tesla getting a fair chunk of both of those markets.
   
   
2. Assumes M-3 ASP of $50k, M-Y ASP of $53k, and C-TRUCK of $55k. M-3 seems high, C-TRUCK seems low; modifying any of these doesn't materially change bullish outcomes given the potential distributions skewing to M-2/A or N-EV, but it does increasing some of the bearish outcomes given more of those models skew towards M-3, M-Y, and C-TRUCK.
   
   
3. The highlighted bull case scenario ($4k) assumes 5M units produced/sold in 2025 with an overall ASP of $38k. This begins to show the optionality at Tesla's disposal. Tesla may never need to produce an N-EV or M-2/A, but can keep pushing cost and price of M3/MY lower and achieve a similar margin mix and ultimately valuation.
   
   
4. Ignores Energy; Ignores Solar; Ignores Service; Ignores SaaS (incl. Dojo); Ignores Licensing; Ignores Credits; Ignores SEMI. In other words, this is a very simple model focused solely on Automotive in most of the model outcomes with Ride Hail / Robotaxi pushing the highest model outputs.
   
   
5. Assumes Tesla continues with debt/equity raises to fund factory buildouts. Arguably they may not need to given their operating leverage, the current war chest (enough cash to build another 10 GFs), and BTC. This would impact operating margin calculations, which would also boost valuations.
   
   
6. Assumes a 70% loss ratio on insurance. This is fairly high considering the argument that Tesla's would be orders of magnitude less likely to have accidents and Tesla would be much better equipped to repair their vehicles relative to a third party insurer (i.e., lower cost to settle claims). That said, this is not a material driver to valuation.
   
   
7. The outlier bull cases (e.g., 20k+ / share) have assumptions of 15M+ units sold in 2025; likelihood of achieving that is quite low based on what we know today (e.g., they have a $22k price target scenario where Tesla sells 66M units in 2025). So, take the outliers with grains of salt as they are in fact very extreme outcomes. They do however show Tesla's ultimate potential if world domination was on the roadmap, but these would be major deviations from the 5-10 year Automotive plan we have heard from Tesla today.
   
   
8. Combing through their simulation outputs and you can see that several very conservative outcomes can still reach or surpass their bear case assumptions (e.g., 0% ride hail or autonomous fleet, 40% margins, 6M units sold, $9k/unit capital efficiency, 13 enterprise multiple = $1,300/share) (e.g., 0% ride hail or autonomous, 48% margins, 5.7M units sold, $4k/unit capital efficiency, 13 enterprise multiple = 1,350).
   
   
9. Point made by Meet Kevin in one of his latest videos, but worth reiterating,
   
   ARK is assuming a very low terminal Enterprise Multiple. ARK is using a 13 multiple in all scenarios for Automotive and Robotaxi and 10 for insurance. This multiple is effectively lower than any average Enterprise Multiple as at 12/31/2020 for Large Cap US Companies (telcos and utilities are both ~14; source: EV/EBITDA (Enterprise Multiple) by Sector/Industry 1995 – 2021) and would be roughly the average across all large caps in 2017/2018/2019. Pushing these multiples to more reasonable growth company multiples (20+), which considering Tesla's growth should continue through 2030 - would be reasonable, and you could very quickly more than 2x all the share price outputs. This again confirms how much margin of error Tesla has to still achieve these valuations. For instance, it may never hit 40% margins (say it plateaus at 30%), but still reach the expected ($3k/share) valuations if it hits 6M units in 2025 and is still on track for 50% CAGR through 2030, thus demanding a higher multiple. All while ignoring Ride Hailing and Robotaxi entirely.
   
   
10. Ending with a tangent, I absolutely love the fact that ARK is willing to publish not only their research notes for free, but equally make available their underlying data model. This is completely against the grain in the investment world and truly shows how disruptive they are to their own industry sector. The model is not only simple (arguably oversimplified), but extremely flexible and allows you to play with whichever variables you believe are the most relevant.

Overall, ARK has put out a model that some will scream is unrealistic (because of the emphasis on Robotaxi at higher valuations), but what has actually spelled out a fairly reasonable and conservative view on the Tesla Automotive business. As many here know, Energy and Solar will be major tailwinds to the Tesla business. It comes across, to me anyways, as actually a very subtle way for them to call out that Tesla actually has the potential to 10x in 5 years if they execute across multiple facets of their business, whereas "simply" executing across a single facet (Automotive) will allow them to grow in to a 3-5x in 5 years.

 

[Discoducky]

I like ARK and invest in nearly all of their ETFs, but I just don't get why they are not modelilng energy or solar. Seems silly to leave it out. My thesis includes these and actually overtakes auto in 2026 when Tesla should reach saturation of selling cars as opposed to selling solar and batteries for the grid.

 

[hershey101]

I like ARK and invest in nearly all of their ETFs, but I just don't get why they are not modelilng energy or solar. Seems silly to leave it out. My thesis includes these and actually overtakes auto in 2026 when Tesla should reach saturation of selling cars as opposed to selling solar and batteries for the grid.

View attachment 647031

Agree 100%

The key business here isn't solar (hardware) as much as the dynamic grid management software/intelligence stack on top. And yes it's going to be a HUGE business.

I like modeling my companies based on what the CEO has said publicly. In the case of Tesla this is: 20M cars in 2030, Energy storage/solar business as big as vehicles (though I only model 50% of automotive revenue for 2030). And then his other ideas: Dojo as a service, Self-Driving etc. as free call options on $TSLA stonk.
I still get insane amounts of potential growth from here through 2030. Elon's downplayed the impact of insurance on the long-term business (he's never claimed it'll be huge but just a nice to have). Elon also publicly stated ride-sharing w/o robotaxies is something they aren't considering when ARK explicitly asked him on Twitter. YET: it's in the model.

My faith in Ark Invest dropped dramatically after this publication. It was already down substantially after the new hires at Ark started making a fool of themselves on twitter. Cathie, no doubt, is great. But the rest of them are novices. Only researcher from Ark I respect other than Cathie is the girl who does Biotech (Tasha or something). And that may only be because I have a very naiive understanding of biotech.

 

[ZeApelido]

This...

I've spent 4 years as a CV research engineer in self-driving. I know what to look for in the car's performance. I'll admit that Tesla has done an extremely good job with a vision-only system. Way better than what most of my colleauges thought would be possible. However, they have huge tail-end errors. Now: They can do funny math to convince regulators (who are relatively stupid) that their system is 10x safer than a human, and it might work. But from a fundamental technology perspective, they will not be there in 5 years. The tail end is EXTREMELY hard to resolve. And while I'm smart enough not to bet _against_ Musk, I'm also not going to give it a 50% probability like ARK haha.
I give it a ~5% probability of success by 2025, and if I move it out to 2030, I'd be willing to bump that % to 15%-20% as the research evolves.

I agree with you FSD = money printer (IF IT WORKS). But % of it working is NOT 50%. Well maybe 50% if you restrict it to one block in sunny CA.. But not the entire US across all weather conditions. Scaling is hard... Elon knows this, he's said it many times... Just because you are scaling a software-based solution instead of hardware (cars) doesn't mean scaling is easy... Scaling is hard period.

Another thing to note is: You don't need self-driving or obscene margins (60% like ARK's model hahaha) to get a reasonable valuation target. A simple EV + energy model going out to 2030 gives you a $3T target (~$2200 with 2.5% yearly dilution, higher on the front half, and buy-backs on the latter half). This is already substantial returns (~17.5% YoY for a decade, or 15% post-inflation, well above the 7.5% historical of $SPX). You still get free call options on: Dojo as a service, self driving (which in my model doubles the valuation to $6T) and you can easily take that strategy and turn it into 25%+ returns with agressive options strategies.

So there are several problems IMO with Ark's modeling:
- They are making up crazy #s to justify obscene targets when they simply don't need to. They could have published a rational valuation model, gone to 2028 with the same #s I'm using for 2030, gotten 20%+ YoY, and not been laughed at.
- They aren't valuing over $2T+ in revenues: Solar, cloud compute.
- They are over-emphasising things that don't even deliver that much to the Tesla bottom line and yet end up leaving themselves open to redicule: eg. Insurance. This is not a HUGE business, even in ARK's own model, if you put insurance to 0 you don't take a huge hit to the pricing. Yet the way they've modeled it shows they don't understand the insurance business.

Piper Sandler is also another rediculous note: they went out to 2040 on their latest 100+ pg. research paper to come to a $1500+ valuation

It looks like we are due for another 2-3 year of lul as valuation catches up to prices, and the fake-news of fintweet leads to a vally of disappointment, IV comes down substantially, and we can repeat the same strategy we did back in 2018-2019 of scooping up cheap LEAPs with IV <45%.

Good post.

What are your thoughts on the value of Tesla's potential high throughput acquisition edge case / fat tail data? To me this, combined with focus on NN systems, allows the best chance to scale learning to conquer all the edge cases.

Do you believe the fat tail simply cannot be solved with camera sensors? Or the compute power is too limiting? Where do you think the bottleneck will be?

 

[Buckminster]

ARK have to keep their target relatively low. Otherwise folk would buy TSLA rather than ARKW

 

[Mengy]

I like ARK and invest in nearly all of their ETFs, but I just don't get why they are not modelilng energy or solar. Seems silly to leave it out. My thesis includes these and actually overtakes auto in 2026 when Tesla should reach saturation of selling cars as opposed to selling solar and batteries for the grid.

That's funny, my own Tesla valuation model also forecasts Tesla Energy revenue overtaking their auto segment in 2026 as well. My model forecasts this as a result of battery production versus auto production, 2026 is when I think they will have enough GWh available for energy to allow this.