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

[BBone]

Wow Loup mentions going to $300B in gross profit in 10 years (if you believe Elon's projections which they cautioned). Taking a 20 P/E on that gives you $60 trillion MV and about 120X current valuations! (DISCLAIMER: NOT INVESTMENT ADVICE, RUN YOUR OWN NUMBERS

Oops math on a sleepy brain. $6 trillion MV. The 120X still is good math

 

[BBone]

The article said 10M cars @ 30K/car = 300B
(operate up to 10M of their own robotaxis, which he thinks can generate a gross profit of about $30k per car per year)

Your're right Chunky. I've replied with a correction to the $6T. And yes I'd be happy with 10% of that as well!

 

[CyberDutchie]

Morgan Stanley’s take on Tesla Autonomous Day:

Tesla’s autonomy investor presentations conveyed impressive technological leadership but, in our view, left big questions around time-to-market and did not change our views on the impediments to removing the human driver in a commercial service at scale. Attention returns to a challenging 1Q.

Nough said...

 

[Causalien]

and I thought short were smart lol
View attachment 399729

I am starting to understand this new dangerous reality we live in. Where anonymous ppl who've never done any chip design can sound like an expert to the normal public just because of repetition and popularity.

What sounded like the most normal statement (and didn't even need to be stated to those of us who worked in the field) can be used as the straw that will break the camel's back.

 

[mblakele]

and I thought short were smart lol
View attachment 399729

It sounds like we can expect FUD that the FSD hardware isn't all that great, because "it's only 14nm and modern fabs can do better". Like most FUD that contains a kernel of truth, and if Tesla were a fab company it'd be damning. For example if Samsung couldn't do better than 14nm and didn't have a roadmap to get there, that'd be a problem for Samsung. But Tesla isn't a fab company.

All Tesla needs is an FSD computer that's fast enough, cheap enough, small enough, and doesn't use too much power. It doesn't matter if it's 14nm or 45nm. Today's FSD computer fits the requirements perfectly. Paying for a smaller process size would probably be a waste of money, and wouldn't necessarily improve FSD performance.

 

[SOULPEDL]

and I thought short were smart lol
View attachment 399729

Architecture and transistor line widths (nanometer dimension) are 2 different things. At intel we called it a shrink appropriately to use a smaller line width. It was done mostly to reduce power consumption. Tesla seems to have created its own architecture for FSD. No doubt the shrink will have improvements, but payoff is in lower power "especially critical in city driving." 72 watts ads up sitting at red lights. Ya the comments are useless, but their the spin would be expected.
And some die use 5nm now... wow that's tiny! Didnt know.

 

[powertoold]

Actually, AP2 surpassed AP1 for many owners in early 2018 already, which was probably in development a few months before general release - which was about a year after the November 2016 FSD demo video. It conclusively surpassed AP1 by most accounts in mid 2018.

Nope. But I guess it's subjective. To think that the FSD demo from 2 years ago wasn't disingenuous is blind. I'll leave it at that.

 

[RobStark]

The video is impressive, but I would like to see Tesla handling a complex urban environment.
I mean... driving in my city in Italy is way more complex than what was shown in that video.
If they really want to have robo taxi on the streets next year they should already be able to drive in those kind of complex environments.

After Tesla makes sure FSD works in its main US markets Elon said on twitter they will turn their attention to Norway.

 

[Causalien]

I now gets why Elon was saying Tesla cars are appreciating assets now.

Imagine Tesla Network as an version of Airbnb where you can rent penthouses for half the price of a Motel room.

Then the builder was selling these penthouses for the price of a trailer, and they are in such huge rental demand, that it’s basically guaranteed to pay for itself in a year.

Then, only two possible outcome from there:

1. The builder will stop selling these penthouses.
2. The builder will keep selling these penthouses at correct price point that ROI is on par with other rental properties.

This theory probably needs to be tamed. FSD will not be availabel everywhere at once for regulatory reasons. So the amount of income from such is limited at first and gradual. And tesla cars will only increase in price at those places.

Also from the demo, it would seem that places like Italy will require longer training because the overall environment is just so different. Asia as well. I don't see fsd getting release in Vietnam for a long time.

So extreme positive income at 2020 when fsd is turned on is not possible. More like a gradual increase. The next 2 years is still basically just cash flow and manufacturing. And the biggest hype being the truck.

 

[powertoold]

No, where did you get that idea. What they said was that they adapted what they already had to the new hardware, starting a few months ago. Not the same thing at all. (I thought they said December, but that could be my faulty memory chip.)

That is based on comments from the demo impressions video I posted AND also Karpathy's comment during the presentation that Tesla ONLY added path planning for sharper turns / ramps into AP about 4 months ago.

 

[SOULPEDL]

It sounds like we can expect FUD that the FSD hardware isn't all that great, because "it's only 14nm and modern fabs can do better". Like most FUD that contains a kernel of truth, and if Tesla were a fab company it'd be damning. For example if Samsung couldn't do better than 14nm and didn't have a roadmap to get there, that'd be a problem for Samsung. But Tesla isn't a fab company.

All Tesla needs is a FSD computer that's fast enough, cheap enough, small enough, and doesn't use too much power. It doesn't matter if it's 14nm or 45nm. Today's FSD computer fits the requirements perfectly. Paying for a smaller process size would probably be a waste of money, and wouldn't necessarily improve FSD performance.

Exactly, u beat me to it. My theory is the next version is just a shrink for power savings. Maybe some new bells to boot, like a Amygdala to add emotions to the car so it can cut off jerks or slow the car because it worries about climate change and all. Middle finger optional.

 

[SPadival]

Morgan Stanley’s take on Tesla Autonomous Day:

Tesla’s autonomy investor presentations conveyed impressive technological leadership but, in our view, left big questions around time-to-market and did not change our views on the impediments to removing the human driver in a commercial service at scale. Attention returns to a challenging 1Q.

Nough said...

Blasphemy!! He forgot to mention capital raise - his real masters will be shouting at him through the Chinese wall

 

[Singuy]

It sounds like we can expect FUD that the FSD hardware isn't all that great, because "it's only 14nm and modern fabs can do better". Like most FUD that contains a kernel of truth, and if Tesla were a fab company it'd be damning. For example if Samsung couldn't do better than 14nm and didn't have a roadmap to get there, that'd be a problem for Samsung. But Tesla isn't a fab company.

All Tesla needs is an FSD computer that's fast enough, cheap enough, small enough, and doesn't use too much power. It doesn't matter if it's 14nm or 45nm. Today's FSD computer fits the requirements perfectly. Paying for a smaller process size would probably be a waste of money, and wouldn't necessarily improve FSD performance.

It's incredibly retarded to talk about Fab's competitiveness when talking about ASIC. If this was an Intel Vs AMD fight, then yeah sure..go at it. But we are talking about special utilization of doing one thing and its performance associated with just this one thing. We see 144 Tops of DLA today, which is useful to Tesla and that's all they care about it. Xavier has 20Tops of DLA in their last gen system and an unknown amount of DLA Tops in the pegasus at 500watt.

Smaller process node when it brings nothing new to the table just add COST. The cost for 7nm is 2- 3x the price of a 14nm cost and has ~20% more defective rate. So Tesla using 14nm is nothing but cost savings. I would be celebrating if they could fab this chip using 22nm.

 

[SOULPEDL]

The video is impressive, but I would like to see Tesla handling a complex urban environment.
I mean... driving in my city in Italy is way more complex than what was shown in that video.
If they really want to have robo taxi on the streets next year they should already be able to drive in those kind of complex environments.

SF and NY was promised, US only at first I believe. Besides, Itally only bought 312 Model 3's so far. Got some catching up to do...

 

[Alphacrux]

and I thought short were smart lol
View attachment 399729

This guy is bat-sugar. Process node alone (“nm”) does not totally define performance. Architecture is a huge differentiator. Tesla put a convolutional neural network (CNN) directly into hardware instead of a general purpose GPU with tensor extensions. The dedicated control flow is much smaller than in a GPU.

The chip is an extremely impressive feat of engineering. A feature shrink to say 7nm will allow some combination of higher performance and lower power.

Samsung is working on advanced process nodes as well. 7 nn with 5nm “risk production” in 2019:

Samsung Foundry Updates: 8LPU Added, EUVL on Track for HVM in 2019
https://www.anandtech.com/show/13329/samsung-foundry-updates-8lpu-for-2019

 

[SOULPEDL]

Something just hit me. Since he is targeting the TN in the US, then sales need to continue focusing growth in the US. So relying on new EU countries or China sales although good for cash flow (higher $ versions), he needs more "units" in US. Two competing agendas - Profit vs more networking units. He said SR+ will be their focus. So should we expect a slow year on the financials, or does he make it up with great margins and FSD sales to keep profits going +? If that is the plan, are we there now and could show profit already in Q1?

 

[Singuy]

This guy is bat-sugar. Process node alone (“nm”) does not totally define performance. Architecture is a huge differentiator. Tesla put a convolutional neural network (CNN) directly into hardware instead of a general purpose GPU with tensor extensions. The dedicated control flow is much smaller than in a GPU.

The chip is an extremely impressive feat of engineering. A feature shrink to say 7nm will allow some combination of higher performance and lower power.

Samsung is working on advanced process nodes as well. 7 nn with 5nm “risk production” in 2019:

Samsung Foundry Updates: 8LPU Added, EUVL on Track for HVM in 2019
Samsung Foundry Updates: 8LPU Added, EUVL on Track for HVM in 2019

Still scratching my head how they freaken made this chip as a side project and suddenly become as competitive as Nvidia while blowing AMD's attempt out of the water. It's really crazy and I don't think these investors really understand what they were looking at.

 

[heltok]

It sounds like we can expect FUD that the FSD hardware isn't all that great, because "it's only 14nm and modern fabs can do better". Like most FUD that contains a kernel of truth, and if Tesla were a fab company it'd be damning. For example if Samsung couldn't do better than 14nm and didn't have a roadmap to get there, that'd be a problem for Samsung. But Tesla isn't a fab company.

All Tesla needs is an FSD computer that's fast enough, cheap enough, small enough, and doesn't use too much power. It doesn't matter if it's 14nm or 45nm. Today's FSD computer fits the requirements perfectly. Paying for a smaller process size would probably be a waste of money, and wouldn't necessarily improve FSD performance.

An interesting point here is that Intel has been struggling with they <10nm chips, as explained here:

The video also explains chiplets which the twitter chip expert bear(who someone here linked) was unaware of.

The main takeaway is that as the transistors get smaller yields have been falling and many chips needs to be discarded. This is improving, but for Tesla going 14nm instead of 7nm in 2018 likely improved their costs and time to scale significantly. 14nm was good enough for now, 7nm or better might be ready for cost/scale in 2021.

 

[S3XY]

It was fake in that it was "gamed," and it didn't reflect the state of the their technology at the time. They had to start over from scratch soon after that because MobileEye ditched them, and the AP2 didn't even have path planning at the time. Per Karpathy, they only recently added path planning for on and off ramps within the last 4-5 months.

Umm, most of us already knew that last part. We've been using NoA ourselves and have known what capabilities it's had. So no, we weren't 'gamed', unless you count the Ataris that were also added recently.

 

[printf42]

This theory probably needs to be tamed. FSD will not be availabel everywhere at once for regulatory reasons. So the amount of income from such is limited at first and gradual. And tesla cars will only increase in price at those places.

Also from the demo, it would seem that places like Italy will require longer training because the overall environment is just so different. Asia as well. I don't see fsd getting release in Vietnam for a long time.

So extreme positive income at 2020 when fsd is turned on is not possible. More like a gradual increase. The next 2 years is still basically just cash flow and manufacturing. And the biggest hype being the truck.

Agree that this is not going to happen overnight everywhere at the same time, but for the location it goes live, I can imagine them taking over Uber+Lyft+Taxi overnight, and slowly takes over private car ownership as well, just because of the economy of not having a driver in the cost structure.

So when one metro area goes online, it could soak up big chunk of supply immediately, so that cars will flow in from other part of the country and affect supply and demand on a much larger scale.