Exactly the progression Tesla described to the California DMV four years ago, it just took longer than anticipated.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable
- Thread starter AudubonB
- Start date
Meanwhile, in Italy, Elon's BFF Giorgia Meloni incentives more polluting cars over Electric ones.
cliff harris
Member
The short answer is that its absolutely physically impossible, regardless of how much cash you throw at it.
Tesla has 6 million cars distributed all over the planet being driven by random people, in random conditions, all collecting data and storing it in a vast data store, with custom built infrastructure training from that data.
This cannot be replicated in 3 years, for ANY amount of money. You need to build out significant IT infrastructure, hire a huge number of extremely specialized people (who will happily stay at Tesla with their stock options thankyou), and then you need 6 million cars and years of data collection.
No way, under no circumstances can it possibly be copied in a year. Maybe if you spend $100billion by literally giving away millions of cars and paying people to drive them to collect the data...
This is why FSD and the NN approach is such a staggering big deal. This is not 'use gigapresses' or '48v'. This is not NACS. This is not something you can steal from a hard drive, or something you can just copy by going to Tesla's parts suppliers. This is the absolute definition of a moat, even if Elon didn't intend to dig one, the way FSD is being solved turns out to be a natural moat.
IMHO this is a way stronger moat than Nvidia has (they don't even make the chips!) and we all know what's happened to their profits and stock.
I am not sure this is completely accurate. While it is true that tesla does have 6 million cars delivering data, the reason the jump from version 11 to version 12 was so substantial was because of AI / end to end Neural nets that tesla is using now. As we now know, a neural network is a method in AI that teaches computers to process data similar to a human brain. So now, tesla is “feeding” these human images for the AI/Neural nets to learn. And, Nvidia “bet” correctly on the GPU’s and chips to create and help this situation. So, it is almost like tesla went from version 1.0 before to 2.0 now. It is true that there are no car companies that have all the real world driving that tesla has, however, the end to end neural nets/AI learns so quickly that the playing field has been narrowed some. Tesla still has a lead, but this new AI compute GPU’s and chips that any car company can buy only gives tesla a smaller lead and other car companies can buy their way there, so maybe only a 1 to 2 year lead?. I did watch a video recently on this (seeing if I can find it) but there are videos on YouTube that discuss this, including videos from Nvidia discussing their AI tools for autonomous vehicle developers That people can view.
GhostSkater
Member
It's been a revelation to see how few EVs I see here
You see them obviously, but I was expecting way higher numbers, I drove for around 8 hours yesterday, ignoring Teslas at Superchargers, I don't think I counted even 10, with Teslas in Superchargers maybe 20
I would say that in a around town driving, the ratio between all cars to EVs might have been higher back home in Brazil than here, kinda hard to take any useful data since there is so much more traffic here
That I remember, Teslas obviously, including my first Plaid sighting
, ID3, eUP and a TaycanWhat about $3 Trillion dollars? That's around $1,000 a share...
If FSD is now hands-free with V12.3.3, this is really huge.
I'm leaning toward the affirmative that, yes, it is hands-free. Tesla is surely very careful with the language here. Since they now leave out language about "hands on the wheel", that must be intentional.
"Ford Blue Cruise and GM Super Cruise let you go hands-free on the highway." - hahahahahahahahahaha
cliff harris
Member
I suspect you cant just buy a billion dollars of AI accelerator chips and have them delivered on Monday. I suspect the lead times for this stuff are VERY long (and getting longer). Chip fabs are hilariously expensive to build, you cannot ramp production overnight. If you are TSMC, you are likely already fully booked out, and nvidia is not going to prioritize some new orders from General Motors (with shaky financials) over long term partners like Tesla.
But even assuming you can order the chips on Monday and get them in 6 months, you also need to build a datacenter (a huge one), and staff it with experts in large scale AI infrastructure. Who is giving up TSLA stock to go work at Ford's new data center, under management that have no idea what you do?
BTW a big AI datacenter is incredibly power hungry. Apply for the grid connection now, and maybe you get it (if you throw money at it) in 2 or 3 years. Thats optimistic.
And even if you just offer $50million signing bonuses to a hundred AI experts to come work there, and even if you have the data centers, and even if you have the chips, and the over-the-air connectivity arrangements and infrastructure to collect the data, and the grid connection to power the chips...
You still need 5 million connected EVs to collect data over several years, to collect all weather/geography combinations.
MAYBE BYD could do it. Maybe. But 3 years is pushing it. And thats to get where Tesla are today, you would still be 3 years behind. And those 5 million EVs? you had to sell every one at a big loss, because otherwise whose buying an FSD-less car when Tesla exists. At least $2k loss on each. Thats $10billion bare minimum set on fire. Total cost to only be 3 years behind Tesla? I'd say minimum $40 billion.
I don’t think it’s quite so clear a path to AI “commoditization” when it comes to vehicle integration. I would sooner short NVDA than bet other vehicle makers will catch up to Tesla in less than 3-5 years.
Edit: Ditto what Cliff said!
Realistically I think in markets that are less deeply established with wide stream local BEV adoption, as in most of the world. EV’s are ubiquitous in some areas (e.g. San Francisco area and nearly invisible in others (e.g. North Dakota). Other than those two examples some places in Italy have many EV’s (in Bergamo several years ago I parked my Model X with six other Teslas) while in most areas they’re far less common. In Brazil the recent entry of BYD is almost instantly increasing BEV visibility outside of wealthy communities.It's been a revelation to see how few EVs I see here
You see them obviously, but I was expecting way higher numbers, I drove for around 8 hours yesterday, ignoring Teslas at Superchargers, I don't think I counted even 10, with Teslas in Superchargers maybe 20
I would say that in a around town driving, the ratio between all cars to EVs might have been higher back home in Brazil than here, kinda hard to take any useful data since there is so much more traffic here
That I remember, Teslas obviously, including my first Plaid sighting
Just considering Norway, the Western Hemisphere, almost everywhere, has major steps to go before really beginning to seem to have penetrated markets. Just looking at areas that do have high BEV adoption, all have major government initiatives, as most major European cities now have restrictions on ICE , major Chinese cities also.
Southern Europe, the Americas and other places make policy changes we probably will not have the major adoption increases that we all imagine, with only TSLA at 20 million per year. Even the US IRA that seemed so momentous is rapidly being watered down. In Brazil, with all the impetus to improve energy resilience there is still public resistance to utility stationary storage, still investing in Diesel peaker plants.
Solar panel projects abound, but not storage, except for private sector, like Vale.
There is no doubt that Tesla is at a crucial juncture, especially with new vehicle markets, like Italy and even much of US. Only Tesla and BYD seem to be pursuing both the energy infrastructure and vehicle markets. Both are thriving, although in different ways.
Tesla securities analysts see only a limited scope car maker, suffering market saturation. Their infection of FUD has become virulent.
So, Tesla has said for some time that 2024 would be transitional. Hearing that, analysts and many investors heard something far less
optimistic.
Others see this is a time to HODL, watch carefully, be diligent, and remember the Hitchhiker’s Guide to the Galaxy and it’s immortal words
“DON’T PANIC”.
So far the competition is at around 2020 fsdb messing around and moving to Birds Eye View net. This is all due to Tesla showing their cards AI day 1 and 2.
Lots of Chinese copying Teslas approach and essentially waiting for their AI day 3 to get the answers. I have checked bili bili China youtube on consumer delivered fsd beta and there are none still from the competition, mobile eye shipped EAP and that's about it.
Souce for this information is 99T, the Chinese tesla AP engineer who worked for them over 8 years. He has watched all the autonomous presentations from competition in China and noticed they just cut and pasted Teslas AI day presentation. This is why AI day 3 is either not happening or will not have many technicals.
I got the FSD trial invite last night, did a bunch of driving with it this morning (or rather I sat in the car while it drove itself!), I'm on the latest version of FSD and I'm still getting the nags. I don't think it's completely hands off yet, but the nags are certainly less frequent than I'm used to.
I think there’s a misunderstand of how much work goes into building a NN like this. If you were to just give another company Tesla NN code and al their raw data they would still take years of ever achieve what Tesla has today. The amount of development that went into all the tools and techniques for processing everything and getting the results that Tesla has is what gives Tesla the lead. Even if you were to give another company the NN AND THE WEIGHTS I think they would be unable to make any improvements on what they would have stolen while Tesla would be continuously making compounding improvements.
Imagine a baby. Or even a young adult. You put it through school, maybe even college. And then dump it in a foreign country with no money and no tools to communicate in the local language. It could walk. It would know how to dress itself. It knows how to eat but would probably actually struggle to acquire anything to eat. It’s still heavily dependent on the ecosystem that built it despite the fact it is itself a meat sac guided by a NN that can function autonomously. Without the system that it knows how to learn from it is going to struggle both to function and to learn to function.
A human in this case will learn to adapt. But the level of AI we are talking about today is not advanced enough to do so.
It’s a lot more complocated than just NN + data = amazing.
Would love to hear what @Discoducky would say about this. I am no expert, just watching videos on YouTube, and as we all know, everyone on the internet is always correct
. @Discoducky, would you be so kind to share your view on this topic as you had worked for tesla and have actual experience. Thanks!There are only so many people willing to beta test self driving software and they already own Teslas.
Been thinking about "the moat" lately as FSD gets incredibly good we are marching the 9s.
What is the FSD moat? A fleet of millions able to collect video/map/weather/environment/vehicle dynamic/cabin driver data capable of being turned into ground truth for NN training that is distributed out to the entire fleet via OTA anywhere with a cycle time of less than 2 weeks.
With infinite money (100B+), how long would it take to 'catch up' to this? Well, never, why? Let's break this down and I'm leaving out the pesky parallelized bits like standing up and building out your multi-exaflop NN infra.
You'd need a fleet of vehicles with Tesla's capabilities. Does this exist today? No. Okay, so how long to build it?
You'd need to re-design your vehicles that you are producing today. Um, I'm not making cars today. Great, then just start from scratch on the important techy bits. Okay, how long does that take? Well, how much is different? The entire communications infrastructure, all the ECUs and sensors that process and send data, full bring up and stack development on vision NNs which nVidia can sell you somewhat. Okay, do I currently have engineers that can do this? No. These are folks you'll need to hire. Okay, let's hire them! Well, these folks are super expensive and you'll need to figure out how to get them to agree to work at your company. Okay, I have money, great, then let's get started! And that my friends, is the story of Project Titan at Apple.
There's another story here that is currently playing out however and that is licensing. This story is yet to be written, but I feel that is DOA and here is why.
I've got a hankering that Tesla is about to solve FSD, so I think to myself, self, how can I get that tech working on my vehicles. And let's say I can convince Tesla to provide me a giant spec book of everything I need to do. Wonderful, now I just need to start making those changes. How long will that take? In the past I've been able to make brand new vehicles in 3 to 5 years. But those cars were largely just the same techy bits with new ICE engines. Ugh, so I'm back to hiring techy bit folks who are super expensive. Gotcha, need a boat load of capital, a super focused VP and a CEO who is all in. This is what I think is going to happen with Ford as Doug Field (my former manager at Tesla) and Jim seem to walking this path in the coming year and the earliest model year to support Tesla FSD would be 2030 at the earliest. And if they don't start this year, the earliest would be 2031.
But I think this is moot, assuming Tesla builds/completes at least 4 more gigafactories by 2028 and saturates the market with robotaxi's. But yeah, quite a moat.
Also, I should mention, no other company is pursuing vision only at scale. Competitors in China are the only thing that could eclipse Tesla for manufacturing and AI, but that is just in China, the race is now getting Europe robotaxi ready and India completely built out for manufacturing and robotaxi's.
In summary, the moat makes it moot.
What is the FSD moat? A fleet of millions able to collect video/map/weather/environment/vehicle dynamic/cabin driver data capable of being turned into ground truth for NN training that is distributed out to the entire fleet via OTA anywhere with a cycle time of less than 2 weeks.
With infinite money (100B+), how long would it take to 'catch up' to this? Well, never, why? Let's break this down and I'm leaving out the pesky parallelized bits like standing up and building out your multi-exaflop NN infra.
You'd need a fleet of vehicles with Tesla's capabilities. Does this exist today? No. Okay, so how long to build it?
You'd need to re-design your vehicles that you are producing today. Um, I'm not making cars today. Great, then just start from scratch on the important techy bits. Okay, how long does that take? Well, how much is different? The entire communications infrastructure, all the ECUs and sensors that process and send data, full bring up and stack development on vision NNs which nVidia can sell you somewhat. Okay, do I currently have engineers that can do this? No. These are folks you'll need to hire. Okay, let's hire them! Well, these folks are super expensive and you'll need to figure out how to get them to agree to work at your company. Okay, I have money, great, then let's get started! And that my friends, is the story of Project Titan at Apple.
There's another story here that is currently playing out however and that is licensing. This story is yet to be written, but I feel that is DOA and here is why.
I've got a hankering that Tesla is about to solve FSD, so I think to myself, self, how can I get that tech working on my vehicles. And let's say I can convince Tesla to provide me a giant spec book of everything I need to do. Wonderful, now I just need to start making those changes. How long will that take? In the past I've been able to make brand new vehicles in 3 to 5 years. But those cars were largely just the same techy bits with new ICE engines. Ugh, so I'm back to hiring techy bit folks who are super expensive. Gotcha, need a boat load of capital, a super focused VP and a CEO who is all in. This is what I think is going to happen with Ford as Doug Field (my former manager at Tesla) and Jim seem to walking this path in the coming year and the earliest model year to support Tesla FSD would be 2030 at the earliest. And if they don't start this year, the earliest would be 2031.
But I think this is moot, assuming Tesla builds/completes at least 4 more gigafactories by 2028 and saturates the market with robotaxi's. But yeah, quite a moat.
Also, I should mention, no other company is pursuing vision only at scale. Competitors in China are the only thing that could eclipse Tesla for manufacturing and AI, but that is just in China, the race is now getting Europe robotaxi ready and India completely built out for manufacturing and robotaxi's.
In summary, the moat makes it moot.
Krugerrand
Meow
It’s as if you don’t bother reading any other posts in this investment thread, or the other investment threads, or the other trader threads -
By the time you type your post and then wait for someone to answer you that you rarely acknowledge back, you could have scrolled a few pages and found your answer at least six times.
No auto company is catching up to Tesla in self driving; technologically or commercially. Nvidia and Google will probably eventually establish some kind of foothold and set up an iOS/Android style duopoly with Tesla being iOS. Tesla can forestall that some by licensing its software.
Tesla has a huge marketplace advantage for the next few years. They created the market for subscription based self driving just like they created the market for long range electric cars. And just like Tesla was synonymous with long range electric cars for a long time, they will be synonymous with self driving in customer’s minds for a long time.
Legacy auto still mostly sells ICE; no one is thinking of putting FSD in ICE. They don’t sell enough EVs to even start collecting data. They lack SW and chip design competency. They are cash strapped. Their customer base is not interested in beta testing kludgy self driving software especially when a much better system already exists. They have no data.
It’s just so much easier to license.
The only exception to this will be in China where the CCP will put a huge thumb on the scale to make sure a viable local competitor to Tesla emerges. They will not totally block Tesla; they recognize the value of competition and know that former Tesla employees can be used to seed local competitors.
cliff harris
Member
Good point on vision only. Really now that FSD12 looks so good, any 'competitor' who mentions Lidar should just be laughed out of the room. Until competitors understand that vision-only is the way, they cannot even START to compete.
Honestly the biggest threat to Tesla FSD may well be regulatory. Lobbyists may insist that this tech is in all cars, and insist Tesla license it at affordable prices.
Honestly the biggest threat to Tesla FSD may well be regulatory. Lobbyists may insist that this tech is in all cars, and insist Tesla license it at affordable prices.
Similar threads
- Locked
- Locked
