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

[Todd Burch]

An actor famous for the "Fast and furious" franchise (which involves a lot of cars and a lot of South America ;-)).

And sadly he passed away because the Porsche GT3 he was in did not have a stalk.

 

[unk45]

You’re too old car people. And it didn’t start as you described. It was a low budget opening film with lesser known actors and actresses. It was a hit and as the series moved along the big name actors came onboard etc…

Yes! The reality does confirm that it is a merchandising ploy, obvious when Mattel is in that act. Not a problem if consumers buy the merchandise. I admit to helping too promote a number of products i would never buy myself. I am less 'too old' than 'too interested in other things'. OTOH I am definitely old enough to be a candidate for US President!

 

[cab]

TSLA up 3.6%, NVDA down 1.4%, I'll take it

Sir. Please. Can we not.

This is why we can't have nice things...

 

[2daMoon]

The real(TM) reason the SP is falling is that some people want to sell shares to pay taxes in April, and the MMs know this is a prime opportunity to get them to capitulate while the price is driven down. They are just shaking the tree to see what falls out, because they can.

/s (mostly)

Follow up. Full disclosure, I was one of those fishing for some folding money to proffer the tax man, and now feel somewhat redeemed after selling a few shares at 180 today.

At least this will relieve any stress incurred while watching for the best price to sell at for the rest of the month. That is a relief.

If my buddy Murphy is paying attention, this could result in a catalyst for a big upbeat in the SP,...
just to put me in my place for exhibiting such arrogance.

 

[Rarity]

I've been thinking about how Tesla is approaching its AI computing needs. We and most others have focused on training needs (Dojo and H100s), but NVIDIA reports that 40% of its data center business is for inference rather than training. It seems likely that as AI use scales, a larger percentage will be devoted to inference.

Because of this, from a computing portfolio perspective, it seems that NVIDIA will not be a large part of the Tesla AI story. It's true that Tesla currently is paying a king's ransom to NVIDIA for training computing hardware. But as Dojo matures as a supplement on training and the masses start utilizing FSD (entirely inference), the percentage of Tesla's overall computing portfolio handled by NVIDIA hardware probably will be relatively low.

Likewise, I wouldn't be surprised for Tesla to offer its inference resources to the broader market in some fashion. How competitive is the current FSD computer to NVIDIA's offering? Honestly, I have not been following the evolution of the FSD computer as I just considered it merely good enough for Tesla's purposes. That's all it needed to be. But maybe a closer look at where it stands from a technology perspective would be warranted.

 

[gillfoto]

An actor famous for the "Fast and furious" franchise (which involves a lot of cars and a lot of South America ;-)).

Thank you, now moving on . . . anyone heard of 'car people' Steve McQueen?

 

[EnzoXYZ]

And sadly he passed away because the Porsche GT3 he was in did not have a stalk.

Paul died in a ~2005 Porsche Carrera GT driven by a profession racecar drive on public streets. Not to be confused by many of the 911 variants as this car was production limited. The Carrera GT was also known as a widow maker as it did not have any fancy electronic traction stabilizer. Car crashed and burned while the passengers was in the car. RIP.

There are few times I am glad Elon does not allow us to switch off traction control on our Tesla especially the performance variants. Things can go horribly wrong real fast.

 

[SmokyPeat]

Thank you, now moving on . . . anyone heard of 'car people' Steve McQueen?

Nope, but I hear they're going to do another remake of Vanishing Point where Kowalski drives a Cybertruck. Totally different ending.

 

[SOULPEDL]

...Likewise, I wouldn't be surprised for Tesla to offer its inference resources to the broader market in some fashion. How competitive is the current FSD computer to NVIDIA's offering? Honestly, I have not been following the evolution of the FSD computer as I just considered it merely good enough for Tesla's purposes. That's all it needed to be. But maybe a closer look at where it stands from a technology perspective would be warranted.

Tesla's inference (oops corrected... I was thinking Training) need is infinite... and for eternity. IMO. The value of a new robot skill or more safety in FSD are two good reasons why Tesla Dojo would (likely) always be worth more internally than renting out compute space. I don't think this will ever be different, and they'd likely be renting to some of their own competitors.

Edit: Put another way, do you see Tesla renting out factory space?

 

[2daMoon]

Nope, but I hear they're going to do another remake of Vanishing Point where Kowalski drives a Cybertruck. Totally different ending.

The Bulldozer loses?

 

[Usain]

Nice report on February sales in Norway and some other European countries.

About 1 in 4 cars sold in Norway was a Model Y. The Viking makes the point that Norway is the only mature EV market today.

If this is what EV sales are going to look like in other markets when they mature, Tesla will absolutely crush the world.

 

[Rarity]

Tesla's inference need is infinite... and for eternity. IMO. The value of a new robot skill or more safety in FSD are two good reasons why Tesla Dojo would (likely) always be worth more internally than renting out compute space. I don't think this will ever be different, and they'd likely be renting to some of their own competitors.

Edit: Put another way, do you see Tesla renting out factory space?

Just to confirm in case it isn't clear: Tesla Dojo has nothing to do with inference. Rather, it is for training. I agree with you that Tesla's training computing appetite will be huge, so it would be a while if ever before an external offering is made other than to perhaps xAI.

I will edit my post to be clearer.

 

[GrandEnigma]

Got assigned a VIN today for my CT order, its in the 3,000s. Also came with a disclosure that none of the computer vision stuff is shipping yet and I had to acknowledge it before I can progress with delivery. Didnt note anything about the missing elocker or hub caps, which I assume they later disclose or one would find out on their own at delivery.

In any event, decent ramp so far as it looks like they should hit about 3k production this quarter.

 

[FEERSUMENDJIN]

and... how does someone program both sets of key cards together? Thx

The car has to be taught the key. Standard in menu option.

 

[Stretch2727]

Got assigned a VIN today for my CT order, its in the 3,000s. Also came with a disclosure that none of the computer vision stuff is shipping yet and I had to acknowledge it before I can progress with delivery. Didnt note anything about the missing elocker or hub caps, which I assume they later disclose or one would find out on their own at delivery.

In any event, decent ramp so far as it looks like they should hit about 3k production this quarter.

Based on the Q4 delivery report Tesla probably will not release the CT number separate from S and X.

I have this odd goal that Tesla will deliver more CT's in the first full quarter of production than GM has delivered Hummers in 2 1/2 years. Similar priced trucks in the same market. It would probably take about 4500 to do this. Probably not but they won't be too far from it!

 

[SOULPEDL]

Just to confirm in case it isn't clear: Tesla Dojo has nothing to do with inference. Rather, it is for training. I agree with you that Tesla's training computing appetite will be huge, so it would be a while if ever before an external offering is made other than to perhaps xAI.

Thanks, I didn't know the proper use of this word. Then wouldn't an FSD license to other companies be such an effort? If so, then yes.

And... I just figured out what you meant. I think I'll have another cup of coffee...

Moments later...

Better answer, and it's a good question: NVidia has no internal mission-oriented effort to use their own inference capacity, unless to design better hardware and chips. So for them, renting it out might be more lucrative as compared to Tesla who is using 150% of their compute for FSD and Optimus.

So, for my latest answer - I doubt Tesla will sell, rent, or lease inference compute like NVidia, at least until they have their own house in order which is like >5 yrs out.

 

[Xd85]

slightly OT: will NACS non tesla vehicles "from 2025" charge with pre-NACS wall connectors?

 

[MP3Mike]

slightly OT: will NACS non tesla vehicles "from 2025" charge with pre-NACS wall connectors?

Depends on the wall connector and how it is setup. Most of them support J1772 so should be fine. But they can be set to Tesla only mode. (There can be some delay as newer Wall Connectors start with Tesla single-wire CAN communications, and fall back to J1772 after a timeout.)

 

[Xd85]

Depends on the wall connector and how it is setup. Most of them support J1772 so should be fine. But they can be set to Tesla only mode. (There can be some delay as newer Wall Connectors start with Tesla single-wire CAN communications, and fall back to J1772 after a timeout.)

you mean J1772 as the communication protocol, not the pins right? That makes sense. I have a couple of gen 3s and didn't know there was a tesla-only protocol. Only mildly considering adding a NACS vehicle next year.

 

[SAPharmD]

When are P&D numbers released?