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

[StealthP3D]

The cost of replacing HW2 to HW2.5 to HW3 to HW4 for FSD owners is easily justified as the cost of developing FSD...Tesla has been putting the best HW they have in production into all vehicles to build the fleet and collect the data they need to perfect FSD. If that means upgrading HW for FSD owners a few times....drop in the bucket IMO.

Exactly. And I think some people were not understanding my earlier examples. Because Tesla has taken money from car buyers, not down-payments but paid in total, Tesla cannot develop FSD to level 5 on any hardware without providing that hardware to all purchasers of FSD or making FSD work on all the cars sold with FSD, not in maybe "two weeks" or two years but when they have the ability to provide it. And considering that FSD is currently 10K and most pre-orders have paid at least $7K, and a substantial amount of that revenue is still unrecognized, it will not be "expensive" for Tesla to upgrade all the old hardware to enable Level 5 on all cars that have paid for it. Claiming it's not yet safe enough for previous hardware versions is not an option.

The good thing about this is that announcing there will be HW4 coming shouldn't Osborne sales of HW3 vehicles.

 

[Christine69420]

Deep dive/first impression of Dojo. So mind blowing that they are skeptical but still can't believe their eyes.

TLDR: "Chips are not the unit of scale for Tesla, the 25 chip tiles are. This tile far surpasses anything from Nvidia, Graphcore, Cerebras, Groq, Tenstorrent, SambaNova, or any other AI training geared start up in per unit performance and scale up capabilities.

All of this seems like far out technology, but Tesla claims they already have tiles running at 2 GHz on real AI networks in their labs.

Rolling it all up, cost equivalent versus Nvidia GPU, Tesla claims they can achieve 4x the performance, 1.3x higher performance per watt, and 5x smaller footprint. Tesla has a TCO advantage that is nearly an order magnitude better than an Nvidia AI solution. If their claims are true, Tesla has 1 upped everyone in the AI hardware and software field."

Tesla Dojo – Unique Packaging and Chip Design Allow An Order Magnitude Advantage Over Competing AI Hardware

Tesla has flipped the hardware world on its head by creating a chip with GPU levels of compute, more than the highest end networking chips IO, and high levels of flexibility. They utilize a integra…

semianalysis.com

This quote in particular was informative:

My mouth is salivating.

Not often I see white collar analysts express themselves like that.

 

[Electroman]

Very impressed with Ashok. He seemed very knowledgeable and loved his gentle presentation style with no hyperbole or dramatics.

 

[Singuy]

This quote in particular was informative:



Not often I see white collar analysts express themselves like that.

Yeah the bandwidth numbers they showed were ridiculous and can scale beyond the tiles without losing bandwidth is mind blowing. It's almost like one gigantic chip made up of 3000 clusters of D1 chips. Order of magnitudes more than the competition. However one must not lose sight of their performance metric which is currently 30% higher per watt than Nvidia's best(which is a very good number..but not super crazy).

 

[LiveLong&Profit]

can someone please relate this to NVDA offerings - this is above my head

Quickly and superficially:

• Nividia builds very good GPUs - which are use for Machine Learning as well (Tesla does too)
• Nividia started making their own custom software suite +10 years ago. A lot of game devs love these tools. They are kind of industry standard.
• Based on gaming chops, Nvidia also gained popularity re. ML, being cheap yet powerful.
• Nividia GPU chips are popular for ML, but not native ML chips
• DOJO D1 is solely built for tesla-specific ML (in a sense) and so *very* optimized for that.
• In the short to medium term Nividia will have lots of customers. In the long term Tesla may become the dominant AI platform. Especially if the Tesla Robot/Tesla Bot/'Optimum sub-prime' takes hold...

 

[Skryll]

Did I hear right in the presentation that CT will debut with HW4 in a year? I thought we are closer to CT production than that...

 

[LiveLong&Profit]

OMG. Tesla is going to be a humanoid robot company...

YES!
Called it

 

[XHabjab]

Tesla doesn’t need to buy Boston Dynamics, if their best employees decide to come work for Tesla instead.

Think 4d.

I see 2 reasons Tesla might want to buy Boston Dynamics (not to exclude a possibly bigger slice of the Hyundai pie):

1. In "large" companies, mergers and acquisitions often occur to absorb the $$$ and other costs of patent fees, royalties, court costs and delay of product. A company buying/merging another company with a single patent might be able to see ROI because of that one royalty. I think we've seen that BD has technology that Tesla is interested in and Tesla is certainly operating in that area of study. Not unlikely at all.

2. Boston Dynamics is currently owned by Hyundai, a Korean company. A lot of BD's product has been put to use by the US military. This might make it (officially or otherwise) strategic in nature. The US military might be interested in them having US ownership. [Remember that Tesla is a US company...]

 

[Singuy]

Did I hear right in the presentation that CT will debut with HW4 in a year? I thought we are closer to CT production than that...

One probably need to distinguish HW4 being debuted on CT in a year vs CT being in production in a year. There's a chance that CT will be released with HW3 and HW4 will show up first on CT.

 

[thx1139]

As predicted, the street considers any kind of android announcement to be a distraction for the company.

Amazing that wallstreet doesnt think that a massive company cant handle multiple projects at once. How about ICE manufacturers being distracted with ICE business with company like Tesla taking away market share of cars.

 

[MP3Mike]

Did I hear right in the presentation that CT will debut with HW4 in a year? I thought we are closer to CT production than that...

No, that isn't what Elon said.

 

[Electroman]

For those asking why a Tesla Robot.

What does a Robot do?: A Robot takes in Visual data and clues, and audible data and clues, understand the environment and react to it. So there are two parts to it:

a) Ability to process images and sounds and understand at a fast rate. Needs to be context aware
b) Perform the physical action.

Tesla has already solved (or will be solving) the former through the FSD project. A car driving at 70 mph is in a more challenging environment than a servant robot, but essentially they are performing a very similar function.

The latter part of actually doing the physical job, is not easy but solvable problem. Tesla can get that done relative easily and it depends on the specific physical job.

If I tell my robot, 'put the red screwdriver in its place'.

a) it should know what a screwdriver is, where it is, its orientation, know where the top drawer is, how to open it, and lastly the path to get to the various end points. All of that FSD has solved.
b) The physical part of picking up, walking, opening the door and putting it in, still needs to be solved.

So Tesla is more than half way there in getting a Robot ready.

 

[Discoducky]

Very impressed with Ashok. He seemed very knowledgeable and loved his gentle presentation style with no hyperbole or dramatics.

He's as solid of an engineer as I've ever worked with and the fact that he's done so well at Tesla over the past 7 years is phenomenal.

 

[TheWalkingDad]

Another hitpiece on Tesla. Amazing how they can spin this. I mean some people will say glass half empty and you should be working to finalize the projects you have already proposed but why can't a company have a vision?

Tesla’s humanoid robot gets muted welcome from Wall Street

 

[LiveLong&Profit]

But why talk robot now? Talk "other usecases to navigate real world" - work on the robot in secret and do an 2007 iPhone moment when pulling it our first time fully running?

Or is it to recruit away the BostonDyn engineers to Tesla to built this beast?

Think Elon just sees the future clearly.
He more or less said that they are going to be built regardless, so Tesla should build them and make them safe.
Almost dutybound.

Also:
- Recruiting raw talent is perhaps the most limiting factor re. how fast Tesla can grow. For a long time, talented and capable people scorned manufacturing. But - what is cooler than mass producing robots? Not much...
This will get talent to choose Tesla. And sure, they will get to work on robotics. But also cars and other stuff.
- Nice to be able to scale solar-roofing by x5, x10 or more by having a work-force of robots who can fall down without it being a major problem.
- Robots are another angle to solve transport and mitigate climate change: How many routine task no longer need you or actual humans to go do some mundane task when you can order a robot to do it. fx by logging into the robots eyes and choose just the apples and grapes you need, or pick up medicine, or whatever. Combine with Starlink then you have telepresence. Want to visit Rome or Tokyo or Shanghai in 5 minutes? The robot can walk the streets for you. And much more...
- Tesla can use the extra income and profit to further accelerate re-inventing the entire supply-chain, especially battery production to support a carbon-free economy faster.

 

[carterm2]

"Latency" The big message from AI day was about the framework for solving FSD. We were shown the progress and roadmap for how the 9's will fall. More importantly and what seems to have been willfully ignored is Tesla's ability to create massively parallel computing platform for themselves. This brings a new level of vertical integration. There is also a possible business in selling compute and possibly compete with Nvidiain the future, as well as licensing FSD to legacy auto.

Q. I would like to know how the NN can fit into a compute platform like the HW3 and actually manage driving in such a complex world. How could there be enough weights, memory, or cpu performance?

 

[RationalOptimist]

The thought that Tesla might acquire Boston Dynamics misses the fact that Tesla's AI architecture is completely different from BD. The latter is essentially algorithmic. it's not based on the deep-learning neural net approach Tesla has. BD have taught their dogs to do amazing specific tricks in specific contexts. But they don't remotely have the kind of general use that Tesla has in mind. It wd be a complete distraction and confusion for Tesla to make this acquisition.

 

[Drumheller]

 

[StealthP3D]

I am away visiting my elderly mum helping out. She could so use a TeslaBot with a nursing app. And a butler app.

That comment highlights just how much money people will need in the future to be considered to be "living well".

Now this is looking more than a decade into the future but if you're happy with your bot just being able to fetch you another beer, you might get by with the base model. But you're gonna want some upgrades. Do you want your bot to be capable of expertly cleaning your robo-taxi? Yup, that's an upgrade. Who wouldn't want their bot to have the same skills as a masterful chef? That's a hardware and a software upgrade! Personal tailor? That only comes with the high dexterity hands, an expensive upgrade if you haven't already bought the high-dexterity hands and Clear Vision eyes! Capable of general carpentry tasks to build that shop you've always wanted? More money! Fine cabinetry making skills? $$$$.

Yes, young people today need to invest well if they want to live well in the future. Being a doctor isn't going to cut it unless you have very modest needs (medicine is about to undergo a seismic shift). Buy TSLA (and other companies of the future) now and over the next several years and you might actually be able to live well in the future.

 

[Doit]

Humans are not energy intensive compared to robots. In general, biological systems are very energy efficient. A human consumes around 100 Watt. I doubt Tesla will build an entire robot within that power envelope.

Well, not that efficient actually as biological systems, particularly humans consume a lot to obtain that 100 watts and also require a rather restrictive environmental conditions to achieve that: temperature (cooling/heating), water consumption, waste process and disposal, etc.