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

[Captkerosene]

He was superb when he was younger.I had a seminar with him 45 years ago and he was brilliant.
I am obviously well past my peak, but Warren Buffet is borderline senile and has been for a decade or so. Primarily I think it is because he never tried to consider climate change as a risk. It is indeed sad to see him fade.

Primarily I think it is because he never tried to consider climate change as a risk.

Buffett is in the insurance business. He's made tens of Billions betting on things like climate change. He has the numbers for hurricanes, fires, droughts and all other kind on natural disaster at his fingertips. Following the last big wave of hurricanes in Florida he made a killing betting that they wouldn't continue (he was right). He has said that there has not been an increase in severe weather (I forget the exact quote) in recent times. He's considered it and rejected it.

 

[OrthoSurg]

You are both assuming the ethical and fully legal kind of "copy and replicate the methodology" performing a clean, independent implementation of it, but perhaps Gary was referring to a more literal "copy", i.e. reverse engineer the FSD CPU and board, copy the trained NN topology and weight set and replicate it. Mind you, such "copy" would be illegal, violating copyrights and other Tesla IP.
However, technically such copy would be possible as long as they intend to use equivalent sensor suite and hardware to execute the NN. That way they could bypass the need for large data collection and training. However, it would be a "static" copy of the given NN as trained at the moment, next time the Tesla downloads a new weight set via OTA update, they would have to repeat the copy and update their own NN other wise they start falling behind.

Gary Black might think it was easy to solve Enigma during WW2. Just get your hands on one and that’s it, everyone can now decrypt the messages. Maybe he thinks there is an Alan Turring working at every automakers

 

[Freder Reeve]

Currently in the US, all but one used M3P and one used 2021 SR+ have FSD turned on. And all the used Y's have FSD.
Just something unusual.

 

[Mardak]

The thing that Tesla has in their pocket at this point is they will already have 3 million+ plus RoboTaxi capable vehicles on the road by the end of 2022 when the next competitor has 100s.

When Tesla proves their FSD approach works with fleet vision data collection from their ~2 million vehicles, a determined traditional auto manufacturer like VW could copy the approach and have multiple millions of vehicles within a year with similar sensors and compute to catch up. Add in someone like Mobileye with vision experience and dedication but lacking the high quality fleet data, it seems possible there will be plenty of partnerships when these companies realize this existential threat.

As others have pointed out, compute is constantly improving reducing the turnaround time to train neural networks that are also improving in architecture, techniques and efficiency; so there is some benefit in being a fast follower. So the big question is if competition will react to robotaxis like they did for electric vehicles. Maybe autonomy companies will be more convincing than battery companies in getting legacy auto to change their ways?

Elon Musk has mentioned winner-takes-a-quarter, so he thinks there will be at least 3 other competitors. But that tweet also highlighted "vast amounts of manufacturing" which again could be a potential advantage for existing large auto makers.

https://twitter.com/x/status/1373492611231535111

 

[Nolimits]

Currently in the US, all but one used M3P and one used 2021 SR+ have FSD turned on. And all the used Y's have FSD.
Just something unusual.

they do that frequently for used vehicles. Turn on fsd- easy up charge and cash inflows.

- Think about it. Someone trades in their model 3 for model Y. Tesla gives $35k for gently used 3. Standard up charge resale 40k. But turn on FSD and sell for 45k. Easy extra 5k for Tesla which cost them NOTHING. Buyers feel it’s a deal bc fsd is worth 10k. Win-win.

- Software is a profit cash machine!

 

[powertoold]

determined traditional auto manufacturer like VW could copy the approach and have multiple millions of vehicles with similar sensors and compute to catch up. Add in someone like Mobileye with experience and dedication but lacking the high quality fleet data, it seems possible there will be plenty of partnerships when these companies realize this existential threat.

Let's say Tesla's approach works, they're the first, and by that time, they have 1 million robotaxis operational in the US. People download the app, get used to the service, have all their settings, music profiles, etc. in the cars. The price per mile is low.

Let's say Tesla is making 2 million cars a year at that point.

Meanwhile, a competitor is trying to copy the approach. How long would it take them to make 2 million copies a year, and also have existing Tesla customers switch over? Are they also going to price their services lower than Tesla?

It seems like quite a challenge to bite into Tesla's pie.

 

[printf42]

When Tesla proves their FSD approach works with fleet vision data collection from their ~2 million vehicles, a determined traditional auto manufacturer like VW could copy the approach and have multiple millions of vehicles within a year with similar sensors and compute to catch up. Add in someone like Mobileye with vision experience and dedication but lacking the high quality fleet data, it seems possible there will be plenty of partnerships when these companies realize this existential threat.

As others have pointed out, compute is constantly improving reducing the turnaround time to train neural networks that are also improving in architecture, techniques and efficiency; so there is some benefit in being a fast follower. So the big question is if competition will react to robotaxis like they did for electric vehicles. Maybe autonomy companies will be more convincing than battery companies?

Elon Musk has mentioned winner-takes-a-quarter, so he thinks there will be at least 3 other competitors. But that tweet also highlighted "vast amounts of manufacturing" which again could be a potential advantage for existing large auto makers.

https://twitter.com/x/status/1373492611231535111

He says winner takes a quarter, but I think he means unit share, he didn’t say how much of profits share the winner would take.

It could be similar to smartphones, iPhone takes modest unit share but the lion’s share of profits.

Reason is, even if others came up with comparable solutions, they would be 5 years behind on hardware so the unit cost would be nowhere near Tesla, but they still need to compete with Tesla on pricing so margin would be low.

 

[Singuy]

I believe Gary has the right idea but not the right execution. Once Tesla shows the world FSD is solved using NN, they will offer their Dojo up for training(as said on the conference call). They may even provide FSD computers for competitors as the ip itself isn't very valuable and any competent chip company can make such a chip in which NN and large memory bandwidth to feed the NN would become the main focus. Tesla will provide the entire software stack and data collection, essentially becoming AWS for car training. This is where Nvidia hope to be and why they have such a high valuation. A move like this would naturally suck valuation away from Nvidia or Intel's mobile eye into Tesla as Legacy could care less how FSD is provided, just as long as it's a package they can buy from any company and sell it to the end user with a small margin(like their entire business model).

Currently even though Tesla haven't solved FSD, Legacy auto makers have a hard time buying from Nvidia or other companies in scale due to the prohibited cost. Nvidia's L5 TRAINING platform is almost 10k, and this doesn't include the 8 or so lidars needed. Not to mention an entire cooling solution that need to dissipate about 3kw of power. Even so FSD isn't solved using this 30k-40k hardware package.

Tesla's version cost about 200 bucks which they can easily just sell it to Legacy, and then charge them a licensing fee for training for the entire life of the car. Basically FSD subscription fee for the entire auto sector.

No auto company will copy this as they wouldn't even know where to start. Nvidia, Mobile eye, and Waymo might ditch their approach asap trying to catch up, which is a totally different strategy as they mainly focused on localization using Lidar and HD maps.

This is another episode of focusing on the wrong thing. Having this FSD advantage has nothing to do with ride hailing at all. Elon Musk is not even thinking about...he's thinking about how to give everyone AI at a price.

 

[insaneoctane]

Anyone want to stipulate the value of licensing FSD? It would likely reduce literally by a small percentage every quarter. ie 2022Q1=$15K,2022Q2=$14.75K,...,2025Q1=$12K

Anyone want to take a stab? Big legacy auto mfg aren’t going to solve this, they will buy it. Why not from Tesla? $$$$

 

[Freder Reeve]

they do that frequently for used vehicles.
- Software is a profit cash machine

yeah we're all aware of the obvious here, what's not common is for tesla to make people shell out extra for nearly every single used 3 (and now Y) in inventory at quarter end. Usually they leave a lot more without FSD to make whatever they can while unloading more product (LR3s are piling up, now they all have fsd)
Prices have been far from regular for the past week+ as well, used cars have been 2-3k over what they usually are, leading people towards just buying new and not bothering with 2-3 year old car to just save 2-3k vs a brand new.
Point being, they seem to be in no push at all to unload their used inventory at EoQ.

 

[buttershrimp]

"Stay in your lane". Personally, I despise the phrase but it applies here.

Also, part of intelligence is being aware of what you don't know.

Totally agree, I'm sure Gary knows a lot about markets and Tesla, but kind of amazing he hasn't taken time to understand the crown jewel of Tesla and FSD, especially given his ability to understand nuts and bolts of financials.

Moments like this make me appreciate Elon... even though he makes occasional false assumptions, when it comes time to make it happen, and he recruits a Karpathy, Straubel or Franz.

 

[Paul_SF]

When Tesla proves their FSD approach works with fleet vision data collection from their ~2 million vehicles, a determined traditional auto manufacturer like VW could copy the approach and have multiple millions of vehicles within a year with similar sensors and compute to catch up. Add in someone like Mobileye with vision experience and dedication but lacking the high quality fleet data, it seems possible there will be plenty of partnerships when these companies realize this existential threat.

As others have pointed out, compute is constantly improving reducing the turnaround time to train neural networks that are also improving in architecture, techniques and efficiency; so there is some benefit in being a fast follower. So the big question is if competition will react to robotaxis like they did for electric vehicles. Maybe autonomy companies will be more convincing than battery companies in getting legacy auto to change their ways?

Elon Musk has mentioned winner-takes-a-quarter, so he thinks there will be at least 3 other competitors. But that tweet also highlighted "vast amounts of manufacturing" which again could be a potential advantage for existing large auto makers.

https://twitter.com/x/status/1373492611231535111

They will have multiple millions of vehicles in a year? Like how? Buying big printers and just print them? And buy all the AA batteries in the world to power them millions of vehicles?

 

[Mardak]

Meanwhile, a competitor is trying to copy the approach. How long would it take them to make 2 million copies a year, and also have existing Tesla customers switch over? Are they also going to price their services lower than Tesla?

If VW group really made the decision to copy Tesla's approach in installing FSD-capable hardware on all vehicles, they could deploy ~10 million future-robotaxis each year. The first year focused on data collection from the this growing fleet that would collect data from roughly 50 billion miles, which is comparable to the total miles driven by all Tesla vehicles to date. With a growing fleet of robotaxi-ready vehicles in parallel with data collection to actually reach robotaxi software, it could be possible for this competitor robotaxi fleet to deploy later than Tesla's already active fleet with customers but then relatively instantly have a larger robotaxi service. This 10 million number seems to be larger than Uber and Lyft drivers combined although that is comparing robots to humans.

Practically, a robotaxi fleet consisting of EVs will be cheaper to operate than one made of ICE vehicles and that will limit the overall competitive pricing. And Tesla is the leader in manufacturing large battery EVs that will be needed in a competitive robotaxi fleet, so even if other auto makers can copy the approach for autonomy, unclear if they will also ramp up EV production to surpass Tesla. But even then, there will likely be a transition period where even non-EV robotaxi fleet can be price competitive versus human drivers, and that could then be used to gain marketshare while building EVs to remain competitive.

 

[Oveeus]

When Tesla proves their FSD approach works with fleet vision data collection from their ~2 million vehicles, a determined traditional auto manufacturer like VW could copy the approach and have multiple millions of vehicles within a year with similar sensors and compute to catch up. Add in someone like Mobileye with vision experience and dedication but lacking the high quality fleet data, it seems possible there will be plenty of partnerships when these companies realize this existential threat.

As others have pointed out, compute is constantly improving reducing the turnaround time to train neural networks that are also improving in architecture, techniques and efficiency; so there is some benefit in being a fast follower. So the big question is if competition will react to robotaxis like they did for electric vehicles. Maybe autonomy companies will be more convincing than battery companies in getting legacy auto to change their ways?

Elon Musk has mentioned winner-takes-a-quarter, so he thinks there will be at least 3 other competitors. But that tweet also highlighted "vast amounts of manufacturing" which again could be a potential advantage for existing large auto makers.

https://twitter.com/x/status/1373492611231535111

The way I see FSD is the same way as I see ride-sharing: whoever able to last to the end (lowest cost/deepest pocket) will win.

No legacy carmakers can touch Tesla because Tesla can issue a 1% dilution and it's enough to bankrupt any of them if not outright buying them.

Only risk are those from FANG who are able to raise enough capital if not more to go against Tesla.

Ultimately though... I think it'd end up similar to how the smartphone landscape is... you have Apple who dominates a few highly profitable markets and takes most of the profit from the industry with a few remaining players sharing the crumbs. Tesla is in the position to become the Apple in the robotaxi industry. As long as that is achieved, whether or not TSLA is the best in the industry is irrelevant. It's all money after all.

 

[Mardak]

Reason is, even if others came up with comparable solutions, they would be 5 years behind on hardware so the unit cost would be nowhere near Tesla, but they still need to compete with Tesla on pricing so margin would be low.

A smart competitor following Tesla would end up better than the auto manufacturers that don't follow and go out of business. And with the many trillions of dollars people are expecting from robotaxis, a low margin is still making a lot of money. Tesla (and us investors) will be rewarded from the higher profits, but there will likely be some surviving competitors that could eventually take the lead in the robotaxi market.

 

[UncaNed]

If VW group really made the decision to copy Tesla's approach in installing FSD-capable hardware on all vehicles, they could deploy ~10 million future-robotaxis each year. The first year focused on data collection from the this growing fleet that would collect data from roughly 50 billion miles, which is comparable to the total miles driven by all Tesla vehicles to date. With a growing fleet of robotaxi-ready vehicles in parallel with data collection to actually reach robotaxi software, it could be possible for this competitor robotaxi fleet to deploy later than Tesla's already active fleet with customers but then relatively instantly have a larger robotaxi service. This 10 million number seems to be larger than Uber and Lyft drivers combined although that is comparing robots to humans.

Practically, a robotaxi fleet consisting of EVs will be cheaper to operate than one made of ICE vehicles and that will limit the overall competitive pricing. And Tesla is the leader in manufacturing large battery EVs that will be needed in a competitive robotaxi fleet, so even if other auto makers can copy the approach for autonomy, unclear if they will also ramp up EV production to surpass Tesla. But even then, there will likely be a transition period where even non-EV robotaxi fleet can be price competitive versus human drivers, and that could then be used to gain marketshare while building EVs to remain competitive.

I know, right? And they could buy them all themselves... So no one will know if they lied about E-MPG or whatever.

 

[MC3OZ]

Copying Tesla's FSD software would be illegal, that would be a risky move for a mainstream carmaker.

It runs on Tesla FSD computer, they would also need to copy that, also the sensor suite and a lot of the rest of the way the car operates.

Without the data it would be hard to train improved versions of FSD.

That is before we get into any copy protection mechanisms Tesla might build into the hardware and software...

There is also a tendency of the Tesla car to communicate with the mothership, the copy software would need to be modified to communicate with their own mothership.

This is complex software, compiled with a compiler Tesla has written, and running on Tesla hardware,

In any court case where Tesla suspected a competitor of copying their FSD solution, the competitor would need to prove how they developed their FSD solution. e.g. where they got the data, why the hardware and software seem identical or very similar.


Given that most car markers still don't have working OTA updates, does this seem like a task they could handle?

The best way to copy, is validate their solution against the Tesla solution, and determine where the Tesla solution is superior and why.
A cheat may be to use Tesla's solution to improve their NN training, so they might need less data.
If for the same visual input the Tesla NN spits out a different result, that is a clue more data and training is needed .
How effective any attempted shortcut would be is hard to determine.

What is in Tesla's favor is this is a difficult problem to solve, and without the edge case data the competitor never knows how their solution behaves in that situation. So they may match the Tesla NN for all of the data they do have, but that doesn't help with the data they don't have.

 

[heltok]

Even if you knew how Tesla solved everything, it still would be a nightmare to try to implement it and get it up and running at an adequate level. Software is messy and you cannot just put something in the cars and start gathering data, you need tons of validation before these steps. Then you need to iterate while still validating every step. Say VW increased new salaries by 10x and hired the best they could, it would probably still take 5 years to have it in production. And they still need to release software for their car in the meanwhile, so probably they will have to incremental development on the current stack. And the current stack probably has so much technical debt that doing it that way will be super messy.

I am leaning on that the current automakers are probably 9 years behind, they are about where Model S was 2012. To catch up they can look at the blueprint and try to copy it, so that might go faster, but they also don’t have the team that Tesla has had. And getting a good team takes time, it’s not just about hiring the right individuals.

It will be super messy at all the active safety departs at the various automakers, plans will change often and release goals will be a moving target. And I think by the end of the year, Tesla will have in production something similar to what the other automakers are hoping(but most our their engineers know will be delayed) to release in 2023.

 

[Runarbt]

Even if you knew how Tesla solved everything, it still would be a nightmare to try to implement it and get it up and running at an adequate level. Software is messy and you cannot just put something in the cars and start gathering data, you need tons of validation before these steps. Then you need to iterate while still validating every step. Say VW increased new salaries by 10x and hired the best they could, it would probably still take 5 years to have it in production. And they still need to release software for their car in the meanwhile, so probably they will have to incremental development on the current stack. And the current stack probably has so much technical debt that doing it that way will be super messy.

I am leaning on that the current automakers are probably 9 years behind, they are about where Model S was 2012. To catch up they can look at the blueprint and try to copy it, so that might go faster, but they also don’t have the team that Tesla has had. And getting a good team takes time, it’s not just about hiring the right individuals.

It will be super messy at all the active safety departs at the various automakers, plans will change often and release goals will be a moving target. And I think by the end of the year, Tesla will have in production something similar to what the other automakers are hoping(but most our their engineers know will be delayed) to release in 2023.

Yes - this is why I think no one else will bother. As soon as Tesla has a working solution and license this out, everyone else will jump in and use Tesla Visual.

Why waste a decade, billions of dollars - get further behind compared to those who just license it, and maybe in the end even fail?

Does not make sense.

 

[Christine69420]

When Tesla proves their FSD approach works with fleet vision data collection from their ~2 million vehicles, a determined traditional auto manufacturer like VW could copy the approach and have multiple millions of vehicles within a year with similar sensors and compute to catch up.

VW already tried that approach with the much simpler to implement software they currently have in their cars. Which has turned into a big nightmare for both VW and the people buying their cars. They tried to copy Tesla and made a big mess.

Your idea just won't work.