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

[mongo]

If:

1. GF Austin, Berlin, and Shanghai make their own cells
2. Each has an annual vehicle output of 500k vehicles
3. Each car has a 60kWh pack (average)
4. Each plant makes 30GWh of cells per year
5. Each plant makes 75MWh of cells per day

Then, with a two day buffer of finished packs (or cells therein), each factory can also be a 150MWh battery storage installation (equivilent to 50 Megapacks). Size of the buffer is dependant only on available space and inverter capacity. 5 day burn in on packs -> ~400MWh. 2 weeks worth would make the plant the largest battery installation in the world (currently 1,200 MWh).

Scale out to the TWh level planned...
Add solar...
Tesla Energy for the win(d?)

Topic switch to NN
Chess and Go are examples of games with perfect information, much easier to simulate/ train against than driving. Perfect information - Wikipedia

It’s Elon Musk vs. everyone else in the race for fully driverless cars

Musk wasn’t as dismissive of simulation programs as he was of LIDAR, but he did claim that even the most sophisticated simulations fail to capture the ultimate “weirdness” of the real world. “If the simulation fully captured the real world, well, I mean that would prove that we’re living in a simulation, I think,” he said, harkening back to his comments at Recode’s Code Conference in 2016. “It doesn’t. I wish.”

 

[canoemore]

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.

Agreed. Also, a company like VW will want to be 100% confident that a given hardware/software path will work before they pay to install the equipment in 10 million cars a year. I’ve been appreciating recently how bold a decision it was for Tesla to install their FSD hardware in every car they make before it’s a proven technology path.

Does anyone really believe that any of the legacy manufacturers (or even FAANGs) are (or are likely to be) this confident in a proposed solution? Maybe NVIDIA will sell them something they claim will do the trick given enough data, but it’s still a bold move to actually commit. I think it may be a bit of a chicken and egg situation. Impossible to achieve FSD without the required scale, impossible to scale for risk-averse players without a proven solution.

I hadn’t thought of the idea myself, but I agree with others here that the other big manufacturers would be more likely to simply license a fully proven Tesla Vision system if it becomes available. Does it make that much of a difference to them if their supplier is Tesla or NVIDIA? And given Elon’s comments about FSD not being exclusive to Tesla long-term, I wouldn’t be surprised if it’s an idea being considered internally by Tesla.

 

[Buckminster]

It's still about the batteries. Even if GM, for example, gets FSD at some point, it will cost much more to run a network of gas guzzlers. They don't have enough battery capacity to build a ton of cars. Also, they would have to pack power hungry Nvidia or equivalent cards into their vehicles as they don't have their own low cost low power boards.

Exactly, the story will unfold like this:

1. FSD combined with having 1m+ in the field gives them a several year advantage
2. They are now first movers with special powers - Coca Cola original
3. Fleet will Increase exponentially (due to batteries) giving them the only APP that can compete with Uber (car needed now)
4. Safety provides another moat, why choose an APP that is two or three 9's behind

The only way Tesla will be in any form of competition is if the entire field place their AVs on Uber and Tesla choose to sell cars rather than run them directly on the network 24/7.

 

[Nolimits]

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.

Another possibility is that they r trying to sell at higher profit margin w the fsd now before the subscription model is released?

just another thought.

 

[joh01652]

The top four headlines yesterday on Bloomberg had to do with Goldman Sachs "unprecedented" sale of $10.5 billion of stocks in a block-trade spree pre-market on Friday.
From the article: "Goldman Sachs Group Inc. liquidated $10.5 billion worth of stocks in block trades on Friday, part of an extraordinary spree of selling that erased $35 billion from the values of bellwether stocks ranging from Chinese technology giants to U.S. media conglomerates. The Wall Street bank sold $6.6 billion worth of shares of Baidu Inc., Tencent Music Entertainment Group and Vipshop Holdings Ltd. before the market opened in the U.S, according to an email to clients seen by Bloomberg News. That move was followed by the sale of $3.9 billion of shares in ViacomCBS Inc., Discovery Inc., Farfetch Ltd., iQiyi Inc. and GSX Techedu Inc., the email said."

CNBS had a similar article on their website, but it didn't mention Goldman, only Archegos Capital Management: ViacomCBS, Discovery plunge due in part to forced liquidation of Archegos Capital positions

In every article the authors make the case that something seems afoot, maybe overleveraged hedge funds like Archegos getting margin called, traders don't know why GS did this, and there is speculation there might be more big sales to come.

Interesting articles, but I don't understand the underlying mechanisms of any of it--thought sharper minds here might have some idea. In the fantasy world which I reside I'd like to think GS is going to put those sales into buying more TSLA.

 

[kbM3]

4 years for Soviets to steal and copy A bomb.

Four years to replicate the Manhattan Project was amazing, but by extrapolation it would have seemed like they should have had no trouble knocking off copies of reusable rockets in only a couple of years, right?

 

[keydiver]

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.

Uhhh...batteries?? Even if VW does successfully build out their planned 240 GWh by 2030, that is still only enough for <5 million cars @ 50 kWh per car. In 2030.

 

[ZeApelido]

Topic switch to NN
Chess and Go are examples of games with perfect information, much easier to simulate/ train against than driving. Perfect information - Wikipedia

It’s Elon Musk vs. everyone else in the race for fully driverless cars

It was great hearing Andrej Karpathy on that podcast using the same words I've been using when arguing with these Twitter clowns opining that Waymo/Cruise are gonna win because they have great simulators. Simulations are useful for varying characteristics of of data you've collected, but they only allow you to interpolate the space your real data collected has sampled. As Karpathy said, you need your real world data to cover as large of a space as possible (e.g. all the edge cases possible), so that your model can interpolate from that data and cover everything.

There is no extrapolation. It is basic brute-force model training.

Now for someone to copy it, well if they do so illegally I could see it not taking too long for them to use it. I have no idea if possible, but if they are actually able to copy the network architectures and the node weights, then they can take the backbone portions and use it with transfer learning and a little bit of their own new data to tune it for their camera setup. This will at least reduce the amount of data they would need to collect. But obviously illegal.

They could also try to learning on the input/output results coming from Tesla's system, but I imagine this would be too tedious and too black/box to be very useful.

 

[Singuy]

The top four headlines yesterday on Bloomberg had to do with Goldman Sachs "unprecedented" sale of $10.5 billion of stocks in a block-trade spree pre-market on Friday.
From the article: "Goldman Sachs Group Inc. liquidated $10.5 billion worth of stocks in block trades on Friday, part of an extraordinary spree of selling that erased $35 billion from the values of bellwether stocks ranging from Chinese technology giants to U.S. media conglomerates. The Wall Street bank sold $6.6 billion worth of shares of Baidu Inc., Tencent Music Entertainment Group and Vipshop Holdings Ltd. before the market opened in the U.S, according to an email to clients seen by Bloomberg News. That move was followed by the sale of $3.9 billion of shares in ViacomCBS Inc., Discovery Inc., Farfetch Ltd., iQiyi Inc. and GSX Techedu Inc., the email said."

CNBS had a similar article on their website, but it didn't mention Goldman, only Archegos Capital Management: ViacomCBS, Discovery plunge due in part to forced liquidation of Archegos Capital positions

In every article the authors make the case that something seems afoot, maybe overleveraged hedge funds like Archegos getting margin called, traders don't know why GS did this, and there is speculation there might be more big sales to come.

Interesting articles, but I don't understand the underlying mechanisms of any of it--thought sharper minds here might have some idea. In the fantasy world which I reside I'd like to think GS is going to put those sales into buying more TSLA.

If you look at the chart it looks like some good old fashion pump and dump. Viacom was on a tear until the massive dump which erased all its gains and then some

 

[powertoold]

Now for someone to copy it, well if they do so illegally I could see it not taking too long for them to use it. I have no idea if possible, but if they are actually able to copy the network architectures and the node weights, then they can take the backbone portions and use it with transfer learning and a little bit of their own new data to tune it for their camera setup. This will at least reduce the amount of data they would need to collect. But obviously illegal.

I remember Elon saying that the FSD chip has security features to ensure that only Tesla software runs on it. I'm not sure if these security features also reduce the likelihood of someone trying to reverse engineer the NN model.

Even if they're able to reverse engineer the model, we already know that Tesla will continually update it, because it's a game of trailing 9s, not 100%. So whoever is reverse engineering the system will essentially have to create a software workflow to automate the reverse engineering and have unit tests to figure out what changed with each update, lol. This seems like a lot of work in itself.

I can see some companies in China trying to do this, but if it were so easy to reverse engineer fsd, a lot of companies would already be doing it with current AP.

 

[Mengy]

https://twitter.com/x/status/1375932607880380417

I assume Gary is a bit off-base with this, but AI is not in my wheelhouse.

Gary sounds uninformed here in my opinion. Level 5 FSD isn't something you "just copy". The software & hardware is only half of it, you also need the data to train the AI. Look how long it's taken Tesla to get this far, it's going to take non-software companies like other automakers MUCH more effort to do the same.

How many phone companies tried to copy iOS and Android during the early days of the smartphone market? How many succeeded in "catching up"?

Tesla won't be the only auotomaker with Level 5 FSD but they will be the first and they will have a huge part of the resulting market share for a very long time. Plus they will likely sell & license their FSD hardware + software to many other automakers who give up trying to do it themselves once the race is lost.

I don't think this is going to go the way Gary thinks it will....

 

[rk20]

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.

Tesla's FSD product is highly integrated and customized for Teslas. It would take a tremendous amount of effort to make it into an OEM type of product ready to be integrated into other EV manufacturers. This would be a big distraction. Why would Tesla do this instead of investing to extend their existing lead.

 

[StealthP3D]

I think I agree with Rob. Let's be honest, while Tesla is on a trajectory to possibly be the largest company in the world someday, the notion that it could happen a few months from now is simply preposterous. Elon must have mistweeted there and meant to tweet on another thread.

I find your leaping to the conclusion that "Elon must have mistweeted there and meant to tweet on another thread", preposterous (italics added). While it's certainly possible he didn't mean to tweet on that particular thread, saying he "must have mistweeted" excludes the possibility that he meant to tweet on that exact thread. And that is what's preposterous!

Personally, I have no reason to believe he actually meant to to reply to a different thread, so to exclude the most likely possibility, is preposterous and means you are on your way to creating your own personal reality, one that that is based more on what you believe than on what you observe. Elon has a long history of posting things that seem preposterous on the surface. And sometimes they don't seem so preposterous after some time passes.

I haven't heard Rob's take on this yet but it would really surprise me if he too excluded the possibility that Elon meant to post that on that thread. He probably just said it was possible that Elon accidently posted on the wrong thread, not that he must have posted on the wrong thread.

While I can't crawl inside Elon's brain, I think the most likely interpretation of that tweet is not that he meant it was likely, either that TSLA could be the most valuable company in the world in a few months, nor that FSD would be fully solved in all conditions in a few months, but that big breakthroughs were coming, possibly as soon as a few months, and when that happens people are likely to be stunned. So stunned that the price of the stock could reflect the magnitude of that stunning development (perhaps over a somewhat longer time period). You also have to account for Elon time. While it could happen "in a few months", that phrase doesn't have well-defined time limits. And, even if that phrase is defined to mean "less than a year", Elon was still only speaking to his belief that it might happen that soon. It doesn't mean he's right nor does it mean it must happen in that timeframe. And as I've pointed out above, "it" might refer to a conflating of perceived value, market cap and the time it takes the market to re-value such a change. "It" might also simply refer to FSD breakthroughs.

One thing that I think informs Elon's thinking is that he knows the difference between what is possible, however unlikely, and what is not possible. Some things are not possible. SpaceX will not be landing on Mars in 2023. Mark Spiegel will not be awarded a Nobel prize. Internal combustion engines will not become more efficient than electric motors.

While I think it seems highly unlikely, it's certainly possible that TSLA could be the most valuable company in the world within a year. To say it's impossible is preposterous. That said, I don't really think that's what Elon was saying, and, even if it was, at least not quite that literally. But I'm not so presumptuous to claim that I know what he meant or even whether he meant to post to that thread or not. I can only make informed guesses.

 

[Mengy]

I find your leaping to the conclusion that "Elon must have mistweeted there and meant to tweet on another thread", preposterous (italics added). While it's certainly possible he didn't mean to tweet on that particular thread, saying he "must have mistweeted" excludes the possibility that he meant to tweet on that exact thread. And that is what's preposterous!

Inconceivable!!!

You might be right though. There is the possibility he tweeted what he meant to and simply took it down to stave the wrath of the SEC again. I must admit it is possible Elon thinks Tesla might become the most valuable company in the world a "few months" from now (Elon Time), however unlikely that scenario might seem to me personally.

I'm not saying I'd be upset if he was correct.

 

[Chunky Jr.]

I don't think other automakers will try to copy Tesla's generalized approach. Instead, they will use a third party solution, such as Waymo, that depends heavily on high resolution maps that work in limited areas initially, and slowly expands over time.

 

[Knightshade]

Agreed. Also, a company like VW will want to be 100% confident that a given hardware/software path will work before they pay to install the equipment in 10 million cars a year.

Sure. The point is- once Tesla has L5 working (if they ever do on current HW) then they do know that path works.

And since they put 10 million cars a year on the road they can duplicate the size of Teslas data set in a year or less at that point (no, they won't all be EVs... but they weren't all gonna be EVs anyway and the data they gather is just as good for this purpose)

As others have noted- most of the high level "design" work on the software (birds eye views, training against video, etc) is publicly available. It's the data set that's the hugely valuable proprietary thing.


There'd still be significant work involved, but it should take less work and less time than it took Tesla... (since they will be collecting data from a much larger fleet even in year 1- and they won't be going down dead end local maximum paths as Tesla did for years before switching to current methods).


Now- Gary is nuts, and has no idea how any of this works, so his "just copy it" notion is misguided- as is his "everyone will have L5 in 2025" notion...

But someone suggesting someone else will have it 5 years after Tesla has L5 is totally reasonable if "someone" is a company that has a large annual fleet sales they can use for data, and they have Tesla to use as a guide for the general how to, AND they are willing to go all in on doing it... (that last one might be tough for legacy... but maybe if Deiss push-evs thing has been working out he could talk the board into it? Certainly been good for VW stock price so far anyway)


Keep in mind- many believe Tesla will have L5 -this year-

The current (general) sensor suite has been on Tesla cars less than 4.5 years right now- the MOST current even less. Meaning Tesla will have done it starting from NOTHING in about 5 years.

Imagine how fast they'd have done it if they already knew the right solution at the start and had a fleet 10-20 times bigger when data gathering began.

 

[powertoold]

Imagine how fast they'd have done it if they already knew the right solution at the start and had a fleet 10-20 times bigger when data gathering began.

I wonder if there'll ever be a documentary series on fsd development at Tesla. It'd be great to know their thought process and approach for all the other "local maximums" they encountered. There's probably some stories to be told when the AP team slept overnight at the office, lol.

 

[elasalle]

Tesla Publishes Patent: 'Estimating object properties using visual image data' for Enhancing Autonomous Driving Systems

Tesla has published a patent 'Estimating object properties using visual image data.' The disclosed invention helps to receive data based on an image captured with a vehicle camera in order to partially identify the distance of an object from a vehicle.

www.tesmanian.com

 

[tschmidty]

The top four headlines yesterday on Bloomberg had to do with Goldman Sachs "unprecedented" sale of $10.5 billion of stocks in a block-trade spree pre-market on Friday.
From the article: "Goldman Sachs Group Inc. liquidated $10.5 billion worth of stocks in block trades on Friday, part of an extraordinary spree of selling that erased $35 billion from the values of bellwether stocks ranging from Chinese technology giants to U.S. media conglomerates. The Wall Street bank sold $6.6 billion worth of shares of Baidu Inc., Tencent Music Entertainment Group and Vipshop Holdings Ltd. before the market opened in the U.S, according to an email to clients seen by Bloomberg News. That move was followed by the sale of $3.9 billion of shares in ViacomCBS Inc., Discovery Inc., Farfetch Ltd., iQiyi Inc. and GSX Techedu Inc., the email said."

CNBS had a similar article on their website, but it didn't mention Goldman, only Archegos Capital Management: ViacomCBS, Discovery plunge due in part to forced liquidation of Archegos Capital positions

In every article the authors make the case that something seems afoot, maybe overleveraged hedge funds like Archegos getting margin called, traders don't know why GS did this, and there is speculation there might be more big sales to come.

Interesting articles, but I don't understand the underlying mechanisms of any of it--thought sharper minds here might have some idea. In the fantasy world which I reside I'd like to think GS is going to put those sales into buying more TSLA.

Stay on topic, man! This thread is now all about FSD, RoboTaxis and stealing AI tech. Try to keep up.

Haven't seen the comical FUD posted here yet from Road and Track. Won't link, but this is the level of hard hitting journalism we expect ;"their report on a 13-minute video posted to YouTube".

Tesla's 'Full Self-Driving' Beta Called 'Laughably Bad and Potentially Dangerous' ​

 

[Runarbt]

Tesla's FSD product is highly integrated and customized for Teslas. It would take a tremendous amount of effort to make it into an OEM type of product ready to be integrated into other EV manufacturers. This would be a big distraction. Why would Tesla do this instead of investing to extend their existing lead.

Sure, the cars will have to be equipped with the needed sensors/hardware suit.
And restrict to EVs with a certain range only, and robotaxi-functionality can only be utilized on Teslas ride-hailing network.

Lets say they sell a complete package, hardware/software - at say $20k a pop - 2000% markup?

I as a shareholder would be pissed if they didnt.

Tesla will be battery/production restrained for decades to come.
If they can ramp sensor-suite production, without restricting their own car production, this is virtually free money.