Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

[FEERSUMENDJIN]

It does need to be cheaper. $100ish a month. It does not need to be good enough to take a nap. It needs to be good enough that Tesla owners find it a compelling value at that price.

A robotaxi trial will not move the stock. It’s a trial. Logistically, it will be a slow, expensive endeavor. It will be years before robotaxis have enough scale to move the needle financially. In that time, it will be a huge cost center.

I agree, that the potential for this as a money maker as level 2/3 is huge if they go down in price and keep the high prices for the full autonomy roll out. I would pay for the ADAS for my old Dad no problem. Tesla have defo been slowing the uptake deliberately, but once they decide the product is good for that level of release, they will go for it. At 100 bucks per month in 2.5 million cars (only 5 fold increase over current fleet) would be a billion dollars of almost pure profit per quarter. Seems attractive !

 

[Knightshade]

If they need one employee for every 50 cars, that is 20,000 people that need to be hired to deploy a million vehicles.

YM 20,000 robots
HTH!

 

[2daMoon]

Those cars have to be cleaned, maintained, charged and parked. They need people to deal with regulatory issues and develop the app and supporting infrastructure.

If they need one employee for every 50 cars, that is 20,000 people Optimus Humanoids that need to be hired built to deploy a million vehicles.

It’s probably a couple of years from start of trial to meaningful catalyst.

FTFY

Keywords to consider: Convergent, Disruptive, Technologies

 

[Usain]

A robotaxi trial will not move the stock. It’s a trial. Logistically, it will be a slow, expensive endeavor. It will be years before robotaxis have enough scale to move the needle financially. In that time, it will be a huge cost center.

I don't see how a robotaxi trial will be slow or expensive.

I'm expecting rapid progress and low incremental cost.

 

[FEERSUMENDJIN]

I don't see how a robotaxi trial will be slow or expensive.

I'm expecting rapid progress and low incremental cost.

Lots of people are poised to be part of the Robotaxi network. That is where we get the running start. Full owned Tesla Robotaxi business will take more time to establish. A really great example of a wedge hybrid implementation.

 

[Usain]

This post works much better with removal of the text block between your two statements and read as telling yourself something






Literally the entire history of regulation and scrutiny around Tesla says this is....not a reasonable take....on how that'd go.





They really are not.





Agreed- Tesla needs at least L4 though, which is what Waymo is. FSDb is not even 3. There are fundamental parts of an L3 (or even moreso L4) system that simply do not exist in FSDb. This isn't "It just needs to get a little better at what it does" this is "it needs to BOTH get a lot better at what it does, plus have multiple additional capabilities and features added that don't exist at all yet in the system.


Most people SUPER optimistic on this stuff just can't be bothered to understand all the parts of self driving and think L4 is just really good L2. It's fundamentally not.


None of which is to say those things can't potentially be added-- some easier than others-- but they're not things you go from "not having at all" "to L4 safe and functional" in months. And at least one (a complete OEDR) it's possible will require more hardware than exists on cars today (more on that in a second)





That depends on what it turns out the actual needed HW is for L4. As that does not exist yet on a Tesla nobody actually knows yet.

Remember when Tesla was sure HW2 was enough? It wasn't. Remember when they were sure HW2.0 cameras were enough? They weren't.

It's possible existing fleet HW isn't enough either. Or it might be. But nobody knows that yet- therefore we don't know if Tesla needs new HW or not for RTs.





If by safety drivers you mean actual drivers, sure. They could do that today.

it won't be a robotaxi, and it'll never be anything but the human driving the thing though.

Doing that wouldn't make much sense economically however-- uber and lyft generally lose money.

Assuming HW4 is good enough and FSD progress accelearates, what can Waymo do that Tesla will be unable to accomplish in relatively short order?

 

[Usain]

People said that about Uber but it's a huge cost center.

With Uber, the human driver takes up to 75% of the revenue.

 

[Knightshade]

Assuming HW4 is good enough and FSD progress accelearates, what can Waymo do that Tesla will be unable to accomplish in relatively short order?

The word assuming is doing a lot of work in that sentence.

Apart from which the first part being true would strand like 4+ million Tesla owners who all paid for FSD but have no upgrade path to get HW4 computer or cameras in their existing vehicles.

(additionally apart from which- I don't really care what Waymo can or can't do- I don't invest in Waymo- I care about what Tesla can or can not do)

 

[Krugerrand]

Q1 delivery estimates are plummeting.

Good. Hope they estimate to zero. Hope the estimates stay at zero. I’d support all estimators going on an extended vacation. Forever.

 

[Usain]

The word assuming is doing a lot of work in that sentence.

Apart from which the first part being true would strand like 4+ million Tesla owners who all paid for FSD but have no upgrade path to get HW4 computer or cameras in their existing vehicles.

(additionally apart from which- I don't really care what Waymo can or can't do- I don't invest in Waymo- I care about what Tesla can or can not do)

Yeah, an investment thesis always includes assumptions.

So I said, "Assuming HW4 is good enough and FSD progress accelearates, what can Waymo do that Tesla will be unable to accomplish in relatively short order?"

We could also assume HW3 is good enough. There is reason to believe that it could be. We really just want to assume that hardware limitations will not cause a new local maximum.

The reason I compare to Waymo is because Waymo has at least proven that a robotaxi is possible. I'm not seeing any technical reason that Tesla can't do what Waymo is doing, but Tesla will do it a lot cheaper. Then it's just a matter of scaling up for Tesla to dominate any market it decides to enter.

 

[stef]

https://twitter.com/x/status/1769470759775195226

Q1 delivery estimates are plummeting.

Very hard to believe consensus is still 475k after they tanked the stock 20% since the meeting with IR ...

 

[Knightshade]

Yeah, an investment thesis always includes assumptions.

So I said, "Assuming HW4 is good enough and FSD progress accelearates, what can Waymo do that Tesla will be unable to accomplish in relatively short order?"

We could also assume HW3 is good enough. There is reason to believe that it could be. We really just want to assume that hardware limitations will not cause a new local maximum.

Ok... but WHY do you assume that?

Tesla has been wrong on the needed HW before. Multiple times. On multiple aspects of the hardware (driving computer, media computer, cameras, etc).

Why are you assuming THIS TIME IS DIFFERENT when they, themselves, haven't achieved the thing yet with any set of HW? (nor has anyone else-- other than companies using FAR MORE hardware and even then only in limited geofenced areas).

The reason I compare to Waymo is because Waymo has at least proven that a robotaxi is possible. I'm not seeing any technical reason that Tesla can't do what Waymo is doing

Waymo, again, is using vastly more hardware-- and still only able to do it in a very few geofenced, HD mapped, places.

That's not, at all, Teslas approach.... so the phrase "I'm not seeing any technical reason that Tesla can't do what Waymo is doing" makes no sense

Tesla is not trying to do what Waymo is doing.

They're trying something very different in fact. Different hardware, different software, different ODDs, different nearly everything.

The fact Waymo is doing what it's doing tells us almost nothing useful about what Tesla is doing, nor how likely it is what Tesla is trying to do will work out.

 

[Krugerrand]

Very hard to believe consensus is still 475k after they tanked the stock 20% since the meeting with IR ...

Nah. That just leaves them with another ‘plausible’ narrative to short it more.

 

[larmor]

Very hard to believe consensus is still 475k after they tanked the stock 20% since the meeting with IR ...

Consensus should really be 80% of 475k: 380k...

 

[Drax7]

When fsd proves 10 safer than a human, car insurance cost will likely come down.
The car injury lawyers may feel the effect, and that may further reduce car insurance premiums.
Stay positive, something good always happens.

 

[Usain]

Ok... but WHY do you assume that?

Tesla has been wrong on the needed HW before. Multiple times. On multiple aspects of the hardware (driving computer, media computer, cameras, etc).

Why are you assuming THIS TIME IS DIFFERENT when they, themselves, haven't achieved the thing yet with any set of HW? (nor has anyone else-- other than companies using FAR MORE hardware and even then only in limited geofenced areas).







Waymo, again, is using vastly more hardware-- and still only able to do it in a very few geofenced, HD mapped, places.

That's not, at all, Teslas approach.... so the phrase "I'm not seeing any technical reason that Tesla can't do what Waymo is doing" makes no sense

Tesla is not trying to do what Waymo is doing.

They're trying something very different in fact. Different hardware, different software, different ODDs, different nearly everything.

The fact Waymo is doing what it's doing tells us almost nothing useful about what Tesla is doing, nor how likely it is what Tesla is trying to do will work out.

Why assume HW3 and/or HW4 will be sufficient for robotaxi? Because I stand to make a lot of money if my assumptions are true. "HW2 wasn't good enough" is not a reason to throw out that assumption, IMO.

Fundamentally, Tesla is going to try to do what Waymo is doing. That is, start robotaxi services in several cities and continue to expand rapidly. But unlike Waymo, Tesla can do it profitably. I'm just not seeing a technical reason why Tesla can't accomplish this.

 

[scaesare]

Even so, it seems more likely to me that the behavior is baked in at training rather than overridden at inference—simulated data may be used, other parameters may be injected,…

The windscreen cameras are ahead of the driver’s eyes, so the vehicle has visibility before the driver.

As for safety, people rationalize all sorts of inappropriate behavior.

Interesting discussion... it seems that the assumption is that the training data is only allowing the NN to watch the driving behavior of others, be it real-world video, or simulation data.

But I don't see why the training data can't include the rules of the road, distilled into some form (from the DMV booklet, etc...) the NN can ingest. This is what humans start with when they are trained. It can also be weighted (i.e.- when visibility is low, err on the side of the written rules, etc...)... this is also what humans tend to do: conservatively do the "correct" thing as uncertainty goes up.

What made me think of this was AlphaZero, the chess AI that Google acquired. It was trained by first giving it the rules of the game, then letting it play itself, developing strategy. It quickly became the dominant chess playing force on the planet.

Of course, chess rules are immutable, whereas driving rules are not, hence the weighting I mentioned above. And the driving space is much more varied than an 8x8 board, so it makes sense to train on real-world data, rather than just letting it solely drive and learn, but my point is that it starts with the rules.

So... undoubtedly FSD started with hard-coded procedural rules... but as Tesla move more towards E2E NN's, I don't see why rules of the road don't simply become another data input for the NN to ingest, rather than something hard coded for a specific rule...

 

[Knightshade]

Why assume HW3 and/or HW4 will be sufficient for robotaxi? Because I stand to make a lot of money if my assumptions are true. "HW2 wasn't good enough" is not a reason to throw out that assumption, IMO.

I mean, I stand to make a lot of money if I buy a lottery ticket I think is a winner. That doesn't mean it's a reasonable assumption.

"I'll be rich if assuming X is right" is an incredibly bad reason to assume anything lacking an otherwise solid case for the assumption.

Fundamentally, Tesla is going to try to do what Waymo is doing

No, they are not. Their approaches are deeply, fundamentally, different regarding self driving. If you can't recognize that you're going to come to a lot of bad conclusions trying to apply data from one to the other.

. I'm just not seeing a technical reason why Tesla can't accomplish this.

This seems to be a lack of technical understanding issue more than anything- where you don't even seem clear how fundamentally different Waymo and Teslas approach's are to self driving--- nor have any understanding of Teslas own approach, tech, or limitations, and are simply going on "If I'm right I get rich".

 

[Drax7]

Eventually I expect sold out bulls and shorts to start competing to buy the stock.
These initial fsd results are very compelling.

 

[Captkerosene]

I don't pooh pooh FSD per say, but rather I pooh pooh the notion that FSD will substantially contribute to revenues and earnings anytime soon. FSD is improving gradually over time and with v12 that rate of improvement seems to be accelerating, but I think it's still too expensive for most customers to seriously consider buying it while it's in beta.

Once it hits Level 5 and truly gets to hands off autonomy things change drastically, but for now while its in beta, I don't feel we can count on FSD sales to help TSLA much at all.

In my pooh pooh opinion of course.

I think the pricing model will change. All cars get FSD at time of sale. After one year you pay $50/mo to keep it. Once we reach unsupervised we kick the monthly up to $100. Adjust the new car price up to reflect what will be a big increase in demand. 12K is just too much for people to bite on. Bundle it in with the car purchase to hide how much they're actually paying for FSD.

In this model we see new car sales growth and margin improvement this year.