Why Autopilot v9 could make Tesla too expensive to take private

[strangecosmos]

Why wouldn't the first mover to autonomy dominate, especially if it's Tesla?

It’s possible, yes. But I have never counted on Tesla dominating autonomy. If ARK Invest’s model turns out to be accurate, a 5% or 10% global market share over the long term would be plenty. That would equate to $500 billion to $1 trillion in annual revenue. At a revenue multiple of just 1x (the average for the S&P 500 is 1.5x), Tesla would increase its market cap 10x to 20x.

It would be ideal for me to invest in multiple companies working on autonomy, actually, if they had a similar market cap to Tesla and I had similar confidence in them. (I elaborate on my views on that topic in this thread.)

 

[Mo City]

It’s possible, yes. But I have never counted on Tesla dominating autonomy. If ARK Invest’s model turns out to be accurate, a 5% or 10% global market share over the long term would be plenty. That would equate to $500 billion to $1 trillion in annual revenue. At a revenue multiple of just 1x (the average for the S&P 500 is 1.5x), Tesla would increase its market cap 10x to 20x.

It would be ideal for me to invest in multiple companies working on autonomy, actually, if they had a similar market cap to Tesla and I had similar confidence in them. (I elaborate on my views on that topic in this thread.)

To clarify: I would NEVER count on Tesla dominating autonomy. I don't believe vision + sonics + one forward radar is enough no matter how fast the chip is. Full self-driving isn't necessary for Tesla to succeed.

However, IF Tesla wins the race to full autonomy with current external hardware + the new chip, they WILL dominate autonomy. The potential of a Tesla Network nukes the fleet vehicle systems competitors could eventually offer (later). Uber would be mortally wounded.

 

[strangecosmos]

If Tesla launches the Tesla Network while lidar still costs more than Model 3, then yeah, Tesla will dominate — at least for the first few years.

 

[Mo City]

Regarding software v.9 in particular,

The fact Elon had Andrej Karpathy on the Q2 conference call seemed to be a rather strong statement about Tesla's confidence in where Autopilot is headed (at least for now). I expect the improved capabilities of v.9 to be significant, both immediately and over the next year or so.

Even without reaching level 5, Autopilot's FSD features just need to become good enough that the vast majority of people must have it.

 

[heltok]

If Tesla tomorrow somehow got the software for self driving lvl5, with todays current hardware, that would be very very big. Compared to Google they already have >100k cars out on the street, they use electric which are cheaper for taxis, camera solution is much cheaper than Lidar today(that might change with solid state Lidars in a few years), and they have a much higher production forecasts. I would assume the value of the source code to be $100B-$1T. Also they should probably delay the Pickup and instead prioritize a pure taxi car as margins for the ride sharing network are much higher than for pickups.

 

[Krispykreme]

To clarify: I would NEVER count on Tesla dominating autonomy. I don't believe vision + sonics + one forward radar is enough no matter how fast the chip is. Full self-driving isn't necessary for Tesla to succeed.

However, IF Tesla wins the race to full autonomy with current external hardware + the new chip, they WILL dominate autonomy. The potential of a Tesla Network nukes the fleet vehicle systems competitors could eventually offer (later). Uber would be mortally wounded.

The problem is that Tesla won’t win and probably wouldn’t even be a leader in this front.

What would make anyone think that Tesla can catch up to likes of Waymo, Zoox, cruise???

 

[Johan]

The problem is that Tesla won’t win and probably wouldn’t even be a leader in this front.

What would make anyone think that Tesla can catch up to likes of Waymo, Zoox, cruise???

Soon 200k cars the roads driving around collecting real world data... (Yeah I know the first 50-100k have less capable sensors, but fast forward a year or two and do the math).

 

[Krispykreme]

Soon 200k cars the roads driving around collecting real world data... (Yeah I know the first 50-100k have less capable sensors, but fast forward a year or two and do the math).

That is the fallacy for outsider .

The quality of data that Waymo has accumulated since its inception is much better than what Tesla has done.

Number of car matters if the cars a equipped to capture good data. None of current Tesla sensor suite, camera systems can match the data quality from Waymo sensor suites.

Tesla sensors suites are about 2 generations behind. The only thing that would come close is their upcoming hi-resolution image radar that would finally provide true point cloud data.

 

[strangecosmos]

Soon 200k cars the roads driving around collecting real world data... (Yeah I know the first 50-100k have less capable sensors, but fast forward a year or two and do the math).

200,000+ Hardware 2 cars have been produced, according to Lex Fridman’s estimate.

 

[Mo City]

The problem is that Tesla won’t win and probably wouldn’t even be a leader in this front.

What would make anyone think that Tesla can catch up to likes of Waymo, Zoox, cruise???

Let's talk when the others begin installing their hardware into mass-produced vehicles being sold to the general public.

Everybody's path to full autonomy is murky at this point, not just Tesla's.

 

[BioSehnsucht]

(moved this reply from Market Action to here as it's closer to being a relevant thread and there's too many AP HW threads to choose from)

Ok thanks for this quote. It clarifies a few thing. It is indeed a custom ASIC so it will definitely fit into the current thermal envelop and also not need any licensing. The GPU will probably stay to do some general pixel by pixel pre processing of the image (which they are really good at), but can also be run at lower power. In this case the mod is right to ban KK.

Bashing tesla's architecture without actually understanding it, thinking that tsla use the same architecture as KK's own Autonomous driving team.

Decent chance that the GPU portion of the existing HW 2.0/2.5 Tegra System-on-Chip (it's an ARM CPU plus small Nvidia GPU plus some other stuff) has enough juice if they keep it to do the basic image pre-processing, and just replace the big GPU with a "big" (chances are it's physically smaller even though more powerful for the application) ASIC.

Alternatively fixed function image manipulation hardware has been around since the 80's, they can put that in the ASIC too, if they know for sure what they need, it's trivial as these things go - and fixed function doesn't have to mean without runtime variation (i.e. the same hardware can be used to scale different parts of different images differently), just purpose built to do one thing. Either way, it's a non-issue.

 

[Causalien]

(moved this reply from Market Action to here as it's closer to being a relevant thread and there's too many AP HW threads to choose from)


Decent chance that the GPU portion of the existing HW 2.0/2.5 Tegra System-on-Chip (it's an ARM CPU plus small Nvidia GPU plus some other stuff) has enough juice if they keep it to do the basic image pre-processing, and just replace the big GPU with a "big" (chances are it's physically smaller even though more powerful for the application) ASIC.

Alternatively fixed function image manipulation hardware has been around since the 80's, they can put that in the ASIC too, if they know for sure what they need, it's trivial as these things go - and fixed function doesn't have to mean without runtime variation (i.e. the same hardware can be used to scale different parts of different images differently), just purpose built to do one thing. Either way, it's a non-issue.

My guess for the asic is that they actually plan on stuffing some of the neural net decision making on it according to the engineer's comment. Not sure where it is done right now, possible on the gpu (jerry rigging a rasterizer to act as "one neuron") and slow on the arm chip.

 

[ValueAnalyst]

The coast to coast self driving demo will be game changing. It's just a question of when that will happen.

End of 2017

 

[strangecosmos]

Well, I have to admit I was wrong about Autopilot v9 and Navigate on Autopilot. Part of this is that Navigate on Autopilot does not yet complete Tesla’s roadmap for full on-ramp to off-ramp driving with no driver input (just monitoring and intervention).

Perhaps this will change when that roadmap is complete. Or perhaps it will change when other functionalities are added, like automatic stopping for stop signs and ted lights. Or perhaps the autonomous coast to coast demo will change market participants’ minds.

In the meantime, I’m happy to keep buying more stock at a price I suspect underestimates the expected value of Tesla’s discounted future cash flows.

 

[_jmk]

I guess investors should remember that people have the tendency to be overly positive about the things they care about. There are no fast magic bullets here, only slow and steady progress. Also mind you, the pace is likely to get even slower after the core features are in place and they start to iron out the never ending corner cases.