Latest / Elon Musk Podcast / Tesla Q3 2024 Financial Results and Q&A with Elon Musk

Transcript

Tesla Q3 2024 Financial Results and Q&A with Elon Musk · Elon Musk Podcast

0:01
Hey everybody. Welcome back to the Elon Musk
0:04
Podcast. This is a show where we discuss
0:07
the critical crossroads that shape SpaceX, Tesla X, The
0:11
Boring Company, and Neurolink. I'm your host, Will Walden.
0:15
Good afternoon, everyone, and welcome to Tesla's third quarter
0:18
2024 Q&A webcast. My name is Travis Axelrod, Head
0:22
of Investor Relations, and I'm joined today by Elon Musk, Devat
0:26
Teneja and a number of other executives.
0:29
Our Q3 results were announced at about 3:00 PM Central Time in
0:32
the update deck we published at the same link as its webcast.
0:36
During this call, we will discuss our business outlook and
0:39
make forward-looking statements. These comments are based on our
0:42
predictions and expectations as of today.
0:44
Actual events or results could differ materially due to a
0:47
number of risks and uncertainties, including those
0:50
mentioned in our most recent filings with the SEC.
0:53
During the question and answer portion of today's call.
0:55
Please limit yourself to one question and one follow up.
0:59
Please use the Raise Hand button to join the question queue.
1:02
Before we jump into Q&A, Elon has some opening.
1:04
Remarks, Elon, thank you. So to recap some listening,
1:11
something that would have been history of saying year over year
1:13
declines in order volumes in Q3. Tesla at the same time has
1:16
achieved record deliveries. In fact, I think if you look at
1:21
EV companies worldwide, to the best of my knowledge no EV
1:27
companies even profitable and I'm not to the best of my
1:31
knowledge there was no EV division of any company of any
1:35
existing order company that is profitable.
1:37
So it is notable that Tesla is profitable despite a very
1:43
challenging automotive environment and and this quarter
1:49
actually is a record Q3 for us. So we produced our 7 million
1:57
vehicle actually just yesterday. So congratulations to the teams
2:01
that made it happen in Tesla. That's the staggeringly advanced
2:04
amount of work to make 7 million cars.
2:09
So the C and we also have energy storage business is growing like
2:15
wildfire with strong demand for both Mega Pack and Powerwall.
2:22
And as people know on October 10th, we laid out a vision for
2:26
an autonomous analytical future that I think is very compelling
2:31
us. The Tesla team did a phenomenal
2:33
job there with actually giving people an opportunity to
2:37
experience the future where you have humanoid robots walking
2:43
among the crowd, not, you know, with the canned video
2:45
presentation or anything, but officially walking among the
2:48
crowd serving drinks and whatnot.
2:50
And and we had 50 autonomous vehicles.
2:53
There were NO20 cyber caps, but there were an additional 30
2:57
model wise operating fully autonomously the entire night,
3:00
carrying thousands, thousands of people's thousands of people
3:04
drink with with no incidents the entire night.
3:10
So, and for those who went there that it's worth emphasizing that
3:16
these, the cyber cap had no steering wheel or brake or
3:20
accelerated pedals, meaning there was no, there's no past,
3:23
there was no way for anyone to intervene manually even if they
3:26
wanted to. And the whole night worked very
3:29
smoothly. So regarding the vehicle
3:33
business, we are still on track to deliver more affordable
3:37
models starting in the first half of 2025.
3:42
You know, this is I think probably purely wanting to go,
3:46
what should they assume for vehicle, vehicle sales growth
3:49
next year? And at the risk of to take a bit
3:54
of risk here, I do want to give some, some rough estimate, which
3:58
is I think it's 20 to 30% vehicle growth next year.
4:05
You know, notwithstanding negative external events like if
4:08
there's some Porsche Majeure events, some big wall breaks out
4:11
or interest rates go sky high or something like that, then you
4:16
know, we can't overcome massive force mature events.
4:19
But I think with our lower cost vehicles, with the Advent
4:24
autonomy, something like a 20 to 30% growth next year is is my
4:29
best guess. And and then, and then cybercap
4:36
reaching volume production in 26.
4:39
I do feel confident of cybercap reaching volume production 26.
4:43
So just starting production, reaching volume production at 26
4:46
and that you know that that should be substantial.
4:51
Both we're aiming for at least 2 million units a year of Cyber
4:56
Cab. That'll be in more than one
4:59
factory, but I think it's at least 2 million units a year,
5:03
maybe 4 million ultimately. So yeah, this is just my best
5:09
guesses, but if you ask for my best guesses that those are my
5:13
best guesses. The the 46, the the cell 4680
5:18
lines, the team is actually doing great work there.
5:21
The 4680 is rapidly approaching the point where it is the most
5:28
competitive cell. So when you consider the the
5:31
fully get the fully landed, the cost of a of a battery pack
5:38
fully landed in the US net of incentives and duties, the 4680
5:45
is tracking to be the most competitive, maybe lower costs
5:49
between what are fully considered than any other
5:52
alternative, which is not quite there yet, But we're we're close
5:57
to being there, which I think is extremely exciting and we've got
6:03
several a lot of ideas to go well beyond that.
6:07
So if I think there's if we execute well the 4680 we'll have
6:15
the the Tesla internally produced cell will be the most
6:18
cost competitive cell in in North America, a testament to a
6:26
tremendous amount of hard work there from by the team.
6:30
So that's a rule we'll continue to buy a lot of cells from our
6:34
competitors. So our intent is not to make to
6:37
provided to make cells just internally.
6:41
So I don't want to sort of any log bells here.
6:44
We're also increasing substantially our vehicle output
6:49
and our stationary storage output.
6:52
So we need a lot of cells and most of them will still come
6:57
from suppliers. But but I think it is, it is
7:02
some good news that the Tesla internal cell is likely it's
7:08
track needs to be the most competitive in the US.
7:13
So with respect to autonomy as people are experiencing in the
7:18
cars really from week to week, there are significant
7:23
improvements in the miles between interventions.
7:28
So with the new version 12.5, the release of wholesale driving
7:34
and Cybertruck, the combining the code into a single stack so
7:40
that the city driving and the engine highway driving are one
7:44
stack, which is a big improvement for the highway
7:48
driving. So it's just all neural Nets and
7:52
the release of actually Smart Summon.
7:57
We try to have a sense of humor here at death and we're also so
8:04
that's 12.5 version 13 of FSD what is going out soon.
8:13
Shark will elaborate more on that later in the call.
8:19
We expect to see some roughly a 5 or 6 fold improvement in miles
8:23
between interventions compared to 12.5.
8:27
And actually looking at the the years a hole the improvement in
8:33
miles between interventions, we think we'll bring at least three
8:37
orders of magnitude. So that's that's a very dramatic
8:44
improvement in the course of the year and we expect that plan to
8:48
continue next year. So the current, the current
8:55
internal expectation, upside internal expectation or the
9:02
Tesla FSD having longer miles between and human is the second
9:11
quarter of next year, which means it may end up being the
9:16
third quarter, but it's next it seems extremely likely to be
9:19
next year. Sure.
9:21
Do you want to? Yeah, I'm mentioning miles
9:24
between critical interventions like you mentioned we already
9:30
made 100 X improvement with 12.5 from starting of this year and
9:34
then with V13 release we expect to be 1000 X from the beginning
9:39
from January of this year on my production release software.
9:44
And this came in because of technology improvements going to
9:47
end to end having higher frame rate partly also helped by
9:51
hardware force more capabilities so on.
9:56
And we hope that, you know, we continue to scale the neural
10:00
network, the data, the training compute etcetera.
10:03
By Q2 next year, we should cross over the average human minus per
10:09
critical intervention public collision.
10:11
In that case I mean that. Is just unbondished our internal
10:16
estimate, yes, yeah. So that's not sandbagging or
10:21
anything else. Our internal estimate is Q2 next
10:24
year to be safer than human and then to continue with rapid
10:30
improvements thereafter. So vast majority of humanity has
10:46
no idea that does those drive themselves.
10:49
So especially for something like like a Model 3 or Model Y, it
10:55
looks like a normal car. So you don't expect normal car
10:59
to be able to be intelligent enough to drive itself.
11:03
The cybercap looks different, Cyber truck looks different, but
11:07
Model Y and Model 3 are look, they're good looking cars, but
11:11
look, they look fairly normal. You don't expect a fairly normal
11:14
looking car to have the intelligence.
11:17
You have enough AI to be able to drive itself, but it does.
11:22
So we do want to expose that to more people.
11:25
And so we're doing every time we have a significant improvement
11:31
in the software, we'll roll, we'll roll out another sort of
11:34
30 day trial so to encourage people to try it again.
11:39
And we're all seeing a significant improvement in
11:42
adoption. So the the take rate for FSD is
11:48
has improved substantially, especially after the 1010 event.
11:57
Yeah. So there's no need to wait for
11:58
robot taxi or server cab for to experience full autonomy.
12:02
We expect to achieve that next year with the with our existing
12:05
vehicle line. I wanted to actually spot
12:12
someone gives a small taste of what it's going to look like.
12:15
The car able to drive itself to the user within private parking
12:19
lots. Currently it's speed limited,
12:21
but then it's going to quickly be increased and we already had
12:25
more than a billion usage items of smarts and then.
12:31
Yep. So and we actually, we have for
12:37
Tesla employees in the Bay Area, we already are offering a ride
12:41
hailing capability. So, so you can actually use with
12:46
the development app, you can request a ride and it'll take
12:50
you anywhere in the Bay Area. We do have a safety driver for
12:53
now, but that's not where we're required to do that.
12:58
We've, we've developed and I mean, David, do you want to
13:00
collaborate on that? Yeah, sure.
13:02
David. We showed some screenshots of
13:04
this in the Q1 shareholder deck. And yeah, this is real.
13:09
We've been testing it for the other part of the year and the
13:12
the building blocks that we needed in order to build this
13:14
functionality and deliberate to production we've been thinking
13:16
about working on for years. It just so happens that we've
13:20
used those building blocks to deliver great features for our
13:22
customers in the meantime, such as sharing your profile,
13:26
synchronizing it across cars so that every single car that you
13:29
jump into, whether it's, you know, another car that you own
13:31
or a car that somebody's loan to you or a rental car that you
13:33
jump into, it looks exactly like yours.
13:36
Everything's synchronized, seat near positions, you know, media
13:39
navigation, everything is the same.
13:41
Just what you would expect from one of our robo taxis.
13:46
But you know, we gave that functionality to our customers
13:49
right now because we built it intending for it to be used in
13:52
the future, releasing that functionality now all the and
13:58
then cybersecurity that we knew we were going to need to deliver
14:00
that functionality, sending a navigation to destination from
14:05
your phone to the vehicle. And so you know you're doing
14:08
that now with the with the ride hailing app, but it's something
14:11
that we've made available to customers for years, seeing the
14:15
progress on a route in the mobile app.
14:17
That's something you'll need for the ride hailing app.
14:19
But again, we released it in the meantime.
14:21
So it's not like we're just starting to think about this
14:23
stuff right now while we're building out, you know, the
14:25
early stages of our ride hailing network.
14:27
We've been thinking about this for quite a long time and we're
14:29
excited to get the. Functionality out there, yeah
14:31
and and we do expect to roll out ride hailing in California,
14:36
Texas next year to the public. But no, the California is
14:41
somewhat this quite a long regulatory approval process, but
14:45
it shouldn't, I think we should get a rule next year, but but
14:48
it's contingent upon regulatory approval.
14:50
Texas is a lot faster. So it's I'd say like we're,
14:57
we'll definitely have available in Texas and probably have it
15:01
available in California subject to regulatory approval and then
15:06
and maybe some other states actually next year as well, but
15:09
at least California and Texas. So I think that'd be very
15:15
exciting. That's really a profound change.
15:20
Tesla becomes more than a sort of vehicle and battery
15:26
manufacturing company at that point.
15:31
So we published that Q3 vehicle safety report which shows one
15:38
pack for every 700 miles of autopilot that compressed the US
15:41
average of water crash roughly every 700,000 miles.
15:45
So it's currently showing a 10X safety improvement relative to
15:50
the US average. And we continue to expand our AI
15:54
training capacity to accommodate the needs of both FSD and
15:57
Optimus. We're currently not not a
16:00
training compute constraint. That's probably the big slimming
16:06
factors that the the emphasis is actually getting so good that it
16:11
takes us a while to actually find mistakes.
16:15
And when you start getting to where it could take 10,000 miles
16:20
to find a mistake, it's it takes a while to actually figure out
16:25
which it is. Is this, is this software ball
16:29
better than a soccer ball A better than software ball B?
16:32
It actually takes a while to figure it out because neither
16:35
one of them making mistakes. What takes take a long time to
16:38
make mistakes? So that's actually the single
16:41
little bit big so many factors. How long does it take us to
16:44
figure out which which which version is better sort of high
16:49
class problem. Obviously having a giant fleet
16:55
is very helpful for breaking this out.
16:59
And then with Optimist we showed, we showed a massive
17:04
improvement in Optimist dexterity movement on October
17:06
10th and our next Gen. handed forum which has 22° of freedom
17:11
double which doubled the prior handed forum.
17:18
It's extremely human like and we'll say it's much better a
17:22
tactile sentencing. It's really I feel confident
17:28
saying that we have most advanced humanoid robot by
17:31
Longshot and we're moreover the only company that really has all
17:37
of the ingredients necessary to scale humanoid robots.
17:46
Because the things that what other companies are missing is
17:49
that they're missing the AI brain.
17:52
They're missing the ability to really scale to very high volume
17:56
production. So you sort of see some
18:00
impressive video demos, but what but they're they're like the
18:03
they like the localized AI and the voyage scale volume to very
18:08
high numbers. As I've said on a few occasions
18:13
before, I think optimists will ultimately be the most part I
18:20
think has a good chance of being the most valuable product ever
18:24
made for the energy business that's doing extremely well.
18:31
And there's the opportunity ahead is gigantic.
18:35
The Lathrop Megapack factory reached 200 Megapacks a week,
18:41
which is now a 40 GW hour a year run rate.
18:46
And we have a second factory in Shanghai that will begin with 20
18:51
GW hour a year run rate in Q1 next year, so just next quarter
18:57
and that that'll also scale up. It won't be long before we're
19:02
shipping 100 GW hours a year stationary storage in Tesla.
19:08
And that I mean that'll ultimately grow I think to
19:15
multiple terawatt hours per year.
19:18
It has to actually in order to have a sustainable energy
19:20
future. If you're not at the term what
19:22
scale, you're not really moving the needle.
19:26
So if you look at our mentally very complicated last master
19:33
plan, which I think actually has too much detail, I'll maybe last
19:38
grow up to analyze it, shorten it up, give us the TLDR on the
19:44
last master plan. But we we showed in that master
19:47
plan that it is possible to take all of us to a fully sustainable
19:52
energy situation using sustainable energy, power
19:55
generation and batteries and electric transport.
20:01
And there are no fundamental material limitations.
20:03
Like there's not some very rare material that we don't have
20:07
enough of on Earth. We actually have enough of the
20:11
raw materials to, you know, take all of human civilization, make
20:17
it fully sustainable. And even if civilization
20:21
radically increased, its electricity usage would still be
20:24
fully sustainable. You know, one way to think of
20:30
the progress of a civilization. Space.
20:33
Sounds a little esoteric, but is percentage completion of
20:36
Khadoshi scale. So Khadoshi scale one would be
20:41
you're using all the power of lava planet.
20:45
We were. We're currently less than 1% on
20:48
Khadoshi level level 1. Level 2 would be using all the
20:53
power of the sun and Level 3 all the power of the Galaxy.
20:58
So we got a long way to go Long way to go.
21:05
When you think in Carter ship terms, it becomes obvious that
21:08
by that the by far the biggest sources of energy is the sun.
21:13
Everything else is in the noise. So, so in conclusion, Tesla is
21:26
focused on building the future of energy, transport, robotics
21:29
and AI. And this is a time when others
21:32
are just focused on managing around near term trends.
21:36
We think what we're doing is the right approach and if we execute
21:40
on our objectives, I think we will Tesla.
21:44
My prediction is Tesla will become the most valuable company
21:46
in the world probably by a long by a long shot.
21:53
I want to thank the Tesla team once again for strong execution
21:57
in a tough operating environment, and we're looking
22:00
forward to building an incredibly exciting future.
22:02
Thank you. Great.
22:05
Thank you very much, Elon. And that, Bob, has some big
22:07
remarks as well. Thanks.
22:10
Our Q3 results were more positive and once again
22:13
demonstrate the scale to which businesses evolved what they
22:16
use. With the generation of record
22:18
operating cash flows of 6.3 billion, our automotive revenues
22:23
grew both quarter on quarter and year on year.
22:27
While we had unit volume growth, we did experience reduction in
22:30
ASPs primarily due to the impact of financing incentives.
22:35
As a reminder, we are providing these incentives primarily using
22:39
third party banks and financial institution and recognize the
22:42
cost of these incentives as a deferral reduction to them.
22:46
We released FSD for Cybertruck and other features like actually
22:50
smart someone like Peter talked about in North America which
22:53
contributed 326 million of revenues in the quarter.
22:58
We continue to see elevated levels of regulatory credit
23:01
sales with over 2 billion of revenues so far this year.
23:05
Expand on this at an industry level, China continues to
23:09
outperform US and Europe by a factor of 3 and if there is
23:15
something to be learnt from that, this gives a signal of
23:18
what is to come in other regions as customers acceptance of EV
23:23
growth and we feel that is the right strategy to build
23:28
affordable and more compelling leads.
23:31
Our focus remains on growing unit volume while avoiding a
23:34
build up of inventory. To support this strategy, we are
23:38
continuing to offer extremely compelling vehicle financing
23:41
options in every lock. When you compare any vehicle in
23:46
a lineup with other OEMs, believe our vehicles provide
23:50
much better value, particularly when you consider the safety
23:54
features, performance and unbalanced software.
23:58
Functionalities like David also talked about include also what
24:03
Ashok had talked about about autonomy, music options,
24:07
parental controls and much more. While every vehicle in our
24:11
lineup comes up with these capabilities, there is an
24:14
awareness gap not just with buyers but at times even with
24:18
existing owners. We plan on making these more
24:21
visible in our interactions with both existing and future
24:24
customers. Automotive margins improved
24:28
quarter over quarter as a result of A50 features release
24:32
discussed before. Increase in our overall
24:35
production and delivery volume, continued benefit from the
24:38
marketing pricing and more localized deliveries in region,
24:42
which resulted in lower freight and duties.
24:45
Sustaining these margins in Q4, however, will be challenging
24:49
given the current economic environment.
24:52
Note that we are focused on the cost per vehicle and there are
24:56
numerous work streams within the company to squeeze that cost
24:59
without compromising on customer experience.
25:03
Yeah. Something that doesn't helpful,
25:05
hopefully helpful macro trend is if there's a decline in interest
25:09
rates, this has a massive effect on the on automotive demand.
25:17
The vast majority of people is the demand is driven by the
25:21
monthly payment, can they put monthly payment.
25:24
So like most likely we'll see continued decline interest rates
25:28
which helps with affordability vehicles.
25:31
Yeah. I mean, that is one trend which
25:33
we observed in the industry that you know, because of
25:36
affordability being impacted because of interest rates,
25:41
people are holding onto the cars longer, especially in the US and
25:45
that is actually having an impact on all of the industry
25:48
team. As we discussed earlier, as we
25:56
discussed impact orders, energy deployments fluctuate quarter
26:00
over quarter due to customer readiness, location of orders
26:04
being fulfilled and not necessarily an indicator of
26:09
demand or production within the quarter.
26:10
While we did see a decline in Q3, we expect to grow deployment
26:15
sequentially in Q4 to end the year, but more than doubled of
26:19
last year. Energy margins in Q3 for a
26:24
record at more than 30%. This is a function of mix of
26:27
projects being deployed in the quarter.
26:30
Note that there will be fluctuation in margins as we
26:33
manage through deployments and our inventory.
26:36
Our pipeline and backlog continue to grow quarter over
26:40
quarter as we fill our 2025 production slots and we are
26:44
doing our level best to keep up with the demand.
26:48
Just coming back on automotive margins, I talked about, sorry,
26:54
I talked about what is happening.
26:56
One other thing which I want to also share is that we're that
27:02
you know, we will continue to keep whatever we can to squeeze,
27:07
like I said before about squeezing out the cost, but this
27:10
is something which we also are very capable of.
27:13
I mean just in Q3 we reached our lowest cost per vehicle and that
27:19
is a trend which we want to keep focus off.
27:23
Then going on to service and other, we continue to show
27:26
improvements in Q3. This was a result of better
27:29
performance both in our service business which includes
27:32
collision parts sales and merchandise and continued growth
27:36
in supercharging. These field based revenues will
27:38
continue to grow as the overall field size increases.
27:43
Our operating expenses declined quarter over quarter in a year
27:46
on year basis. This is partially due to the
27:48
restructuring we undertook in Q2.
27:51
Cost saving from these initiatives were partially
27:54
offset by increase in costs related to our AI efforts.
27:58
We've started using the GPU cluster based out of a factory
28:02
in Austin ahead of schedule and on track to get 50K GPUs
28:06
deployed in Texas by the end of this month.
28:09
One thing which I'd like to elaborate is that we're being
28:12
very judicious on our AI compute spend too and saying how best we
28:17
can utilize the existing infrastructure before making
28:20
investments. On the CapEx front, we had over
28:24
3 1/2 billion in the quarter. This was a sequential increase
28:28
largely because of investments in AI Confused, we not expect
28:32
the CapEx for the year to be in excess of 11 billion.
28:37
We shared our vision for the future at the WE Robot event at
28:40
the beginning of the month. The Tesla team is hyper focused
28:44
on delivering on that version. The all efforts are underway to
28:47
make it a reality. While we've achieved significant
28:50
progress this year, it will take time to get this as we find in
28:55
new and incredibly complex technologies and navigate a
28:59
fragmented regulatory landscape. Future is incredibly bright and
29:04
I want to thank the Tesla team once again for all the help.
29:09
Great. Thank you very much for all.
29:11
Now we'll go to investor questions.
29:13
The first one is, is Tesla still on track to deliver the more
29:17
affordable model next year as mentioned by Elon earlier?
29:20
And how does it align with your AI products right now?
29:24
Sure. I mean as Elon and Bible both
29:25
said, you are in plan to meet that in the first half of next
29:29
year. Our mission has always been to
29:31
lower the cost of of the of our vehicles to increase the
29:34
adoption of sustainable energy and transport.
29:37
Part of that is is lowering the cost for current vehicles, which
29:40
is where all of the personally owned vehicles that we sell
29:43
today come in. But the next stage in that
29:45
really is it fits into AI road map is when we bring in robo
29:48
taxis, which lowers the initial cost of of getting into an EV.
29:52
And those that's really where we see the marriage of of EV road
29:56
map in the AI road map. Yeah.
30:00
It'll be like with with with incentive sub sub 30K which is
30:05
kind of a key threshold. Right.
30:09
Thank you very much. Similar question next, when can
30:14
we expect Tesla to give us the $25,000 non robot taxi regular
30:17
car model? We're not making it on.
30:21
Yeah. All our vehicles today are.
30:23
Robotaxi, I think we've made very clear that we're the future
30:26
is autonomous. I mean, it's going to be, you
30:32
know, actually said this many years ago, but that my strong
30:38
belief, and I believe that is panning out to be true.
30:40
So it'll be very obvious in retrospect, is that the future
30:44
is autonomous electric vehicles and non autonomous gasoline
30:50
vehicles in the future will be like riding a horse and using a
30:53
flip phone. It's not that there are no
30:57
horses. Yeah, there are some horses, but
30:59
they're unusual, they're niche and you know, so, so really just
31:06
everything's going to be electric, autonomous.
31:10
I think this this is blind, like should be frankly blindingly
31:14
obvious at this point. That is the future.
31:18
So a lot of automotive companies or most automotive companies
31:22
have not not internalized this, which is surprising because
31:27
we've been shouting this from the rooftops for such a long
31:30
time. And it will accrue to their
31:33
detriments in the future. But all of our vehicles in the
31:38
future we yes, sorry, older vehicles that we've really made
31:43
obviously have millions of vehicles.
31:45
The vast majority are capable of autonomy.
31:49
And you know, we're currently making on the order of 30 by
31:54
1000 autonomous vehicles a week. If you compare that to say Wayne
31:58
More's entire fleet, it's less than they have less than 1000
32:01
corners. We make 35 KA week.
32:04
Yeah, and our cars look normal. Yeah, they mostly look normal.
32:08
Cybertruck looks yes, thankfully, you know, looks
32:13
abnormal. And then the, you know, cyber
32:17
cab slash robotaxi. You know, we wanted to have
32:21
something futuristic looking. It does look futuristic.
32:26
It's worth starting with respect to the Cyber Cab, it's not, it's
32:30
actually not just a revolution in vehicle design, but
32:33
revolution in vehicle manufacturing that is also
32:36
coming with this with the Cyber Cab.
32:39
The the cycle time like the the units per per hour of of the
32:48
Cyber Cab line is like this is just really something special.
32:53
I mean, this is probably, yeah, half ordered magnitude better
32:59
than other car manufacturing lines.
33:02
Like like, like like not in the same league, is what I'm saying.
33:08
Not in the same league. So so it's it's it's you know,
33:13
and I said like several years ago that the maybe the most
33:19
hottest Tesla point coffee will be the foundry.
33:22
Yeah, just like buy a. Factory.
33:23
Yeah. You can't reverse engineer a
33:28
factory. This up to Mile, Yeah.
33:29
It's like, you know, yeah, so the and as we're so we're
33:37
rapidly evolving and manufacturing technology.
33:40
So so anyway, there's like basically I think having a
33:46
regular 25K models pointless. Yeah, it would be silly.
33:50
Like it would be completely at odds with what we believe.
33:52
In, in an autonomous world, what matters is the lowest cost per
33:56
mile of, of efficiency of that vehicle.
33:58
And that's what we've done with, yeah, with the rope taxi.
34:00
Exactly. Autonomous, it's fully
34:02
considered cost per mile is what matters.
34:05
And if you try to make a car that is, you know, essentially a
34:12
hybrid manual automatic car, it's it's not going to be as
34:17
good as a dedicated autonomous car.
34:22
So yeah, Sarah cab is just not going to have steering wheels
34:26
and paddles. Well, you design optimized for
34:30
autonomy, but now it'll it'll it'll, you know, cost on the
34:35
order of cost roughly 25K. So it is a 25K car and you can,
34:40
you won't be able to buy one on it exclusively if you want.
34:45
So just won't have steering wheel paddles.
34:49
You don't need it. Great.
34:53
Thank you very much. The next question is what is
34:56
Tesla doing to alleviate long wait times at service centers?
35:00
So we aim on solving problems at the source, so at the factory,
35:04
before they can even affect our customers.
35:07
We believe the best service is no service and.
35:10
Really is. Don't even have them.
35:12
If the car doesn't break. Yeah, exactly.
35:13
That's the best thing. Don't see anyone with the test
35:15
shirt you need to do it fix the issue upstream or you would
35:18
remotely do it through software. They can be at work or at home
35:22
and you know car can be parked and we address the fixed issue
35:26
and we've partnered the field with service to make sure we're
35:29
looking at the same issues. And additionally just in Q3 and
35:32
Q4 of this year long we we have opened and will open in total at
35:36
nearly 70 locations. And in North America, we
35:40
significantly expand the size of each location and have doubled
35:42
the size last year compared to this year.
35:46
Yeah, I think, I think it's like actually a lot of merits that
35:48
have any large service centers because you can, you can have
35:52
specialization of Labor, you can start, you can start to
35:55
approach. Yeah, it should be more factory
35:57
like, you know, where you can have dedicated lanes for
36:00
particular types of servers, OK. And it's way easier for somebody
36:03
to become expert in a few different types of repairs than
36:07
in every repair. Exactly, this has helped us with
36:09
the base of these heavy repairs like clogging up the lane.
36:13
They've dedicated lanes for different type of repairs and so
36:16
it's throughput matters and really treating it like a
36:18
factory. Yeah, this is, this is where
36:22
Tesla structure, I think Tesla has a structural advantage
36:25
relative to the rest of the auto industry because we, we make the
36:29
cars and we service the cars. Whereas I think there's a bit of
36:33
a conflict of interest with the dealer model and the sort of
36:38
traditional OEM, a dealer model where the dealerships make most
36:41
of their money on service. And so they don't, they
36:46
obviously just incented to reduce the servicing cost.
36:50
Whereas in our case, we are incented to reduce the servicing
36:53
cost because we we carry that servicing cost and we've got a
36:57
good feedback loop with our with our cars.
37:00
Yeah. With the factory, with the
37:02
service leaders together and send people from the factories
37:05
to the field and field to the factory to see it first hand.
37:09
Provide suggestions for, you know, manufacturing as well As
37:11
for engineering on design. Yeah.
37:13
So I view this as a structural, fundamental structural advantage
37:17
of Tesla versus the rest of the auto register.
37:22
Also, doing a bunch of work on the software side, not only to
37:24
automate, you know, diagnostics, so identifying, you know, what
37:28
needs to be done to a car before it comes into service, but also
37:31
automating all of the preparation work and aligning
37:34
all the resources that are necessary in order for the car
37:37
would be very efficiently worked on once it arrives.
37:39
So the parts are there like the lift is scheduled, the
37:42
technician's schedule like everything that we like.
37:44
The farm said this is what's wrong with me and tell us to
37:46
fill the service center the. Car.
37:48
Everything ready in event? Yeah.
37:50
Please fix me. And this is what's wrong, Yeah.
37:52
This is what you're going to do now, Yeah.
37:55
Instead of customer trying to translate, the car is telling us
37:57
directly and we're pulling that. Yeah, yeah.
37:59
You don't need, but most of the time you don't need to diagnose
38:01
the car exactly when it arrives. The car, Yeah, this is like,
38:05
again, a fundamental technology advantage and structural
38:09
advantage compared to the rest of the auto industry.
38:11
Yeah, it's. I think it's under appreciated
38:15
as to what all we are able to do and that's why because like you
38:19
said before, most of our cars except for cyber drug look the
38:23
same. Yeah, right.
38:24
So people don't realize that it has so much capability.
38:28
Yeah, yeah, yeah. But look, that's better than
38:30
other cars. But they're not like, obviously,
38:33
like super futuristic, yeah. Great.
38:38
Thank you very much. The next question is please
38:42
provide an update on the semi. What will the next stage of
38:45
growth look like and when will FSD be ready?
38:48
Sure. So as you, we posted in the
38:50
earnings back, we're progressing something on the build of the
38:52
semi factory in our data factory in Reno.
38:56
We've released all our major capital expenditures for that
38:58
program and we're on track to start pilot builds in the
39:01
second-half of next year with production starting in the first
39:04
half of 2026 and rampant really throughout the year to to full
39:07
production semi, you know, growth will largely depend on
39:11
our customers adoption of the product.
39:13
Well, I, I, I don't think we're going to be divided limited.
39:14
Honestly, yeah, what should I say?
39:15
Which is like, you know, trainer for foot semi because it's
39:18
really a commodity of total. Cost of ownership, Yes, exactly.
39:21
It's it's good we we have kind of ridiculous demand for the
39:26
semi. In that world where it's about
39:29
how much do I spend? Yeah, it excessinates per mile.
39:32
It's a no brainer. Yeah, fundamentally, if you've
39:35
got a semi work, the fully considered cost per mile per per
39:39
ton of transport is better than say diesel truck.
39:43
Any company that doesn't adopt an electric semi will will lose.
39:47
It's not a, it's not a subjective thing.
39:50
It's like whether do you like? It's a competitive, I mean, we.
39:52
We like, we, we want the style we want, we want to have a
39:54
beautiful semi truck. But frankly if you made an Oxy
39:56
semi truck it wouldn't matter. And, and this is proving so in
40:01
our fleets and in Pepsi's partner.
40:04
In fact, the the Pepsi actually said last week they're having
40:07
nobody want their drivers don't want to go back once.
40:10
Yeah, yeah. As soon as we give anyone the
40:12
electric semi, it's it's, it's like the that's like the choice.
40:17
It's the what they want to. Drive yeah yeah, yeah that's
40:19
like like so the like the most senior like their top drivers
40:22
will they get to drive to tell the same one and it's it's the
40:26
it's the thing they want to drive it's it's it's super fun
40:29
to drive it's also very. Easy to drive.
40:31
It's easy. It's easy to drive and it holds
40:34
ass and it's like fast. Maybe too fast.
40:38
Well, but I mean like, you know, like, like you've seen like the
40:41
videos of where like the electric semi, like, you know,
40:44
can go uphill speeding past like the diesel truck.
40:50
Yeah, in cars. So like it's responsive.
40:53
It's you know, you know, you floor it and that the truck
40:57
actually, but. And that's a benefit not only
40:59
for for the driver and for the goods, but also for safety in
41:02
terms of other drivers on the road.
41:03
You don't get stuck behind the semi.
41:05
You're not like, you know, in a slow down situation in the
41:08
homeland. I mean, how that plays into, you
41:10
know, FSD, which is the second part of the question.
41:12
All of the semis have been since the couple 100 we've deployed
41:15
already and and the ones that we'll be building next year and
41:17
throughout should you have all of the hardware in the cameras
41:21
necessary to to deploy FSD and we're currently training with
41:25
that small that we have and as soon as the fleet is trained and
41:28
the neural Nets are up, you know we'll we'll get FSD onto that
41:31
platform. Yeah, I mean, it'll be a massive
41:33
improvement in driver fatigue, you know, because and driver
41:38
safety. We've got sort of the anti Jack
41:41
knifing software. We're, you know, you don't have
41:44
to worry about your brakes overheating if you go down a
41:46
down a steep hill because because the we, we use
41:48
regenerating like that, that energy goes back in back into a
41:52
pack. Actually, when we leave ring on.
41:54
Something it's just like it's it's like radically better than
41:58
than a diesel assembly. It's what what the drivers love
42:00
it, OK. Great, guys.
42:03
Thank you very much. Our next question is when will
42:05
Tesla incorporate X and CROC in all the Tesla vehicles?
42:12
I mean, these are relatively small 5 things, you know, but
42:16
yeah, I think we'll keep expanding, you know what, what
42:23
is available in the car on the spot screen and also improving
42:28
like the browser. So like just generally you can
42:30
access anything you want in the car.
42:31
In fact, for the Tesla, you know, once you get to full
42:34
autonomy, you actually want fully a, a system that is you
42:41
can do anything like if you want to browse the Internet, if you
42:43
want to, you know, ask, ask AI questions, if you want to watch
42:48
a movie, if you want to play a video game, if you want to do
42:51
some productivity thing, you can do anything you want and an
42:55
autonomous vehicle because you don't need to drive.
42:58
And so that's why the Cybercast got a nice big screen and it's a
43:03
great sound system. So you can watch it, watch a
43:07
great movie with it's like bringing like a.
43:09
Personal movie theatre. Yeah, personal movie theatre.
43:14
Awesome. Yeah, This is why we've been
43:16
building this functionality, adding gaming to the car, adding
43:19
these and other all sorts of different media applications to
43:21
the car, because you know, the cars, that's what you're going
43:25
to. That's, yeah, the cars that we
43:26
built today, There's some. Really fun games by the way.
43:28
People haven't tried it. There's like Castle Doobad and
43:32
Paulotopia and a bunch of really fun games in the car.
43:35
Yeah, we're, we're constantly looking at, you know, what
43:37
features that annex and we're paying attention to what's most
43:41
commonly refunded by our customers.
43:42
Yeah, play Castle Bimbat, you wonder.
43:47
Great. Thank you guys very much.
43:49
The next question is, Elon mentioned unsupervised FSD in
43:52
California and Texas next year. Does that mean regulators have
43:56
agreed to it in the entire state for existing Hardware 3 and 4
43:59
vehicles? No, as as I said earlier,
44:04
California loves regulation, but they have a pathway.
44:07
Yeah, I mean, there's a pathway. Obviously Waymo operates in
44:09
California. So there's just a lot of forms
44:10
about and a lot of approvals that are required.
44:14
I mean, I'd be shocked if we don't get approved next year,
44:16
but it's, it's just not something we totally control.
44:19
But I think we will get approval next year in California and
44:22
Texas and, and and towards the end of the year it will branch
44:26
out beyond California and Texas. I mean, I think it's important
44:29
to reiterate this like you're certifying a vehicle at the
44:33
federal level in the US is done by meeting FMBSS regulations.
44:37
All our vehicles today that are produced that are autonomous
44:39
capable meet all those regulations.
44:41
The cyber cattle meet those regulations.
44:42
And so the deployment of the vehicle to the road is not a
44:45
limitation, but is a limitation is what you said.
44:47
At the state level where they control autonomous vehicle
44:50
deployment. Some states are relatively easy
44:51
as you mentioned for Texas and other ones have half place like
44:54
California that they take a little longer, other ones
44:56
haven't. Set up anything yet and so we
44:58
will work for those state by state I, I do think we should
45:01
have a federal. I agree that like autonomous
45:04
vehicles should be approved. It should be.
45:06
It should be possible to. Congress, if you're listening,
45:08
let's get a federal AB. There should, there should be a
45:10
federal approval process for autonomous vehicles.
45:13
I mean, that's that's how the FMBSS.
45:15
Is starting to work. Federal Motor Vehicle.
45:18
The FMBS is federal. Yeah.
45:20
So I mean in 2017 and 18 that we, you know, some regulators
45:24
started looking at it and it's really kind of stalled since
45:26
then. But we would appreciate and
45:28
would support helping out it. Really needs to be at like a
45:31
national approval is is important.
45:34
You know, if there's a department of government
45:35
efficiency. I'll try to help make them open.
45:38
And you took for everyone, not just Tesla, but you know, just
45:43
we, we like like some, some things in the US are state by
45:46
state regulated, like for example, insurance.
45:48
And it's like a it's incredibly painful to do it state by state
45:52
50 states. And and I think we should have,
45:55
there should be a national approval process for autonomy.
46:00
So. Great.
46:02
Thanks guys. The next question is what is the
46:05
plan for 2025I? Don't think we're just talking.
46:10
Yeah, we, I mean, basically we talked through this and there's
46:12
a lot going on. You know, ALDI mentioned that
46:15
we're working on cheaper models to come out.
46:18
I mean, they're work which the team is doing to get the
46:21
factories ready today to try and make that happen.
46:24
Yeah, by the way, the amount of work required to make a lower
46:28
cost car is insanely high. But like, it is harder to get
46:34
like 20% of the cost out of a car than it is to design the car
46:37
and build the entire factory in the 1st place.
46:40
It's like excruciating and it's and there's not a lot of movies
46:44
made about the the the heroes who got 20% of the cost out of a
46:48
car. But let me tell you, there
46:49
should be. That is incredibly heroic.
46:56
It's a little change, isn't it? It's not like.
46:58
This yeah, but it's like there should be the the heroes who got
47:00
20% the cost out of a car is like, damn, I'm going to respect
47:03
to them. It's like, you know, I think you
47:08
probably could make a compelling movie, but it's just no, no,
47:12
like if you actually saw how hard, if people actually saw how
47:15
hard it was to do that, you'd be like, whoa, that's damn hard.
47:20
Just yesterday we were talking about potty.
47:22
Yeah. I mean, honestly, like
47:24
literally, yeah. I mean, there's a lot of what I
47:27
do call it sort of like getting cost out of things.
47:29
It's kind of like, it's like Game of pennies.
47:31
So it's like Game of Thrones, but pennies.
47:34
You know, first approximation you've got if you've got 10,000
47:36
items in a car. Very rough approximation.
47:40
And each of them costs $4.00. Then you have a $40,000 car.
47:44
So if you want to make a, you know, a $35,000 car, you're
47:48
going to get $0.50 on average out of the 10,000 Landers.
47:51
Every. Every part, yeah.
47:54
And it's like, you know, and then obviously the best is you
47:57
delete some parts. In fact, we've not been able to
47:59
delete a lot of parts. I'm very, I'm not very excited
48:03
about the, the Sava cap design and, and, and the but you know,
48:09
how how, how rethinking the design of a car for the side of
48:12
a cab, designing it will offer high volume production and then
48:16
design machine that builds the machine that is that I think is
48:21
also revolutionary. And it's just that there's no
48:24
other car company that's even trying to do what we're doing.
48:28
Like I've even heard of it actually.
48:30
In fact, I'm certain there isn't one Like like like I'm, I think
48:35
this, this the new machine that builds the machine like it's so
48:40
it's, it's inherent like that. It's it's designed to be like
48:48
five times better than provisional factory like cycle
48:53
time. Cycle time and like heart
48:56
deletion, I don't think any other car company has the same
49:00
level of like integration of thought that we have when it
49:03
comes to like when you design a part from a white sheet of paper
49:06
who's going to make it where's it going to be made how's it
49:08
going to be shipped how's it going to be assembled into the
49:11
vehicle And like at any one point if something is done in a
49:15
silo, it becomes a bottleneck of either cost or time or
49:19
efficiency. But with the with the row taxi,
49:21
you know the, the development like we've done a good job on
49:26
like combining all that and then like blowing up how it's made
49:29
and saying it should be made this way and rethinking it all.
49:31
So it's the most efficient factory possible that shows in
49:33
our we will ship in our CapEx efficiency when when we deploy
49:36
it, it shows in the number of parts, it shows in the
49:38
simplicity of the vehicle, but also how it performs in, in
49:41
terms of like end user state. Yeah.
49:44
Just to close up, just on the energy front, also in 25, we
49:48
will have started manufacturing of the manufacturer.
49:52
We'll continue to increase our storage deployments with
49:55
Powerwall 3. We plan to continue expanding
49:58
our supercharging network, getting more OEMs on our network
50:02
4680 in that cell ramp as you all talked about that would keep
50:07
going. And then there's a, we're also
50:09
we'll have a little refinery starting to produce.
50:12
So there's a lot which is going on.
50:13
Great. Yeah, so many things, Yeah.
50:17
I mean, the great like crazy thing is like Tesla's winning
50:20
basically on, on almost every single thing we're doing.
50:23
If we're not winning now, we're trying to run in arenas where
50:27
there are entire large companies that that's the only thing they
50:30
do, yeah. I mean that's it's a company.
50:34
There are multiple companies within the company.
50:36
Yeah, Tilda's, like many companies they want yeah.
50:40
That's it. Thank you guys.
50:41
Just a few more. What is going on with the Tesla
50:44
Roadster? Fun things.
50:47
Well, I just thought you'd like to thank our long-suffering
50:50
deposit holders of the Tesla Roadster.
50:53
You know, the reason it hasn't come out yet is because it is
50:55
the roaster is not not just the icing on the cake, it's the
50:58
cherry on the icing on the cake. And so, you know, our larger
51:05
mission is to accelerate the progress towards a sustainable
51:09
energy future, you know, trying to do things that maximize the
51:12
probably the future is good for humanity and for Earth.
51:17
And and so that necessarily means that like the things like
51:22
that are kind of like dessert. But we'd like we'd all love to
51:25
work on the Tesla, the next Gen. Tesla rose.
51:27
We're super it is super fun and we are working on it, but it has
51:31
to come behind them. The more things, the things that
51:35
have a more serious impact on the good of the world.
51:38
So just thank you to all Lord suffering Tesla roster deposit
51:42
holders. And we are actually finally
51:45
making progress on that. And we're we're close to
51:50
finalizing the design on that. It's really going to be, I
51:55
think, spectacular. You know, a friend of mine,
52:00
Peter Thiel. Yeah.
52:04
And sometimes people like Peter Thiel and our rivals, we're
52:07
really good friends. You know, Peter, you know, was
52:11
lamenting how, you know, the future doesn't have flying cars.
52:15
Well, we'll see more to come. Yeah.
52:20
Great. Thank you very much.
52:23
The next one is quite similar to other questions we've had, so I
52:26
might combine it with the final question.
52:28
So briefly, could you just detail how Robotaxi will roll
52:32
out? Will it start with a Tesla
52:34
deployed fleet and then allow customers to add theirs on like
52:39
a subscription model and then we'll hardware 3D capable of
52:41
level 5? Because you're going hardware 3.
52:49
What we saw with 12.5 was it was easier to make progress with
52:52
starting with Hardware 4 and figuring out the solution and
52:55
then back porting into Hardware 3 instead of directly working on
52:58
hardware 3. Given that hardware 4 has more
53:00
like fundamental hardware capabilities.
53:03
I think that trend will continue into the next few quarters as
53:07
well. But we first figured the
53:08
solution rapidly with AI4 and then back ported, right, the
53:13
kernels. It just takes longer to develop
53:15
those things because it's not fundamentally supported in the
53:17
hardware and it's emulated. But yeah, it's initially working
53:23
on hardware 4, backboarding it to Hardware 3.
53:26
Yeah. So it's the answer is we're not,
53:29
we're not 100% sure, but, but as a short mention, because by some
53:39
measures Hyper 4 has really several times the capability of
53:43
hardware 3. It's easier to get things to
53:45
work with hardware 4 and then it takes a lot of effort to sort of
53:51
squeeze that functionality into hardware 3.
53:55
And there is some chance that hardware 3 is does not achieve
54:02
the safety level that allows for unsupervised FSD.
54:10
You know, there was some chance of that.
54:13
And if that turns out to be the case, we will upgrade those who
54:18
have bought hardware 3 FSD for free and we have designed the
54:22
system to be upgradable. So, and it's really, it's really
54:28
what you know, just to switch, sort of switch out the computer
54:32
type thing like the camera, the cameras are, you know, they're
54:39
capable. But anyway, we don't, we don't
54:42
actually know the answer to that.
54:43
But if it does turn out, we'll take, we'll make sure we take
54:47
care of those who have bought MSD on Hot Route 3.
54:51
Great. And in the last few minutes that
54:53
we have left, we will try to get in some analyst questions.
54:57
The first question will be coming from Pierre Ferrigan at
55:00
New St. Pierre.
55:02
Please feel free to unmute yourself.
55:07
Thanks a lot guys for taking my question.
55:11
I was wondering about like the compute you're you're ramping
55:14
up. So you gave like interesting
55:16
statistics on how much you have and you said you you don't feel
55:19
your compute constraint. And I was wondering, you know,
55:23
how you are putting to work this additional compute.
55:27
Is that a game for you of creating like larger and larger
55:31
models, like next generation of models that are larger the way
55:35
open AI go from GPT 3 to GPT 4? Or is that more like you're set
55:40
on your model and you need to throw more and more compute to
55:44
accelerate the pace of learning to to improve reliability?
55:52
And then I had a quick photo pretty quick on you roll out in
55:55
Texas and in in California next year.
56:00
The plan as you see today is it to roll out like a fleet of two
56:06
with cars that we start with like a supervisory like like
56:11
some on board supervision, someone sitting at the wheel
56:15
just in case and removing the supervisors progressively or are
56:19
you aiming for going fully fledged without even a human
56:23
super supervisor when you get? Started, OK, well, I guess we're
56:28
going to, I'll answer. Yeah.
56:31
The first part of the question, Dean, the nature of real world
56:34
AI is different from say an LLM in that you have a massive
56:40
amount of context. So that like the, the you've got
56:45
a case of Tesla 7 or 8 cameras that you know, 9 up to 9 if you
56:50
include the internal camera that, that, that so you've got
56:57
gigabytes of context and, and that that is then distilled down
57:02
into a small number of control outputs.
57:07
You know, whereas it's like you don't really, it's very rare to
57:09
have, in fact, I'm not sure any ALM out there can do gigabytes
57:13
of context. And then you've got to, you've
57:17
got to then process that in the car with a very small amount of
57:22
compute power. So, you know, it's, it's a, it's
57:31
all doable and it's happening, but it is a different problem
57:35
than than what's say a Gemini or an open AI is doing.
57:41
And now part of the way you can make up for the fact that the
57:45
first computer is, is quite small is by spending a lot of
57:52
effort on training. And just like, just like a human
57:57
like you, the more you train on something, the less, the less
58:04
mental workload it takes when you try to when you when you do
58:08
it like when the first time, like a human starts driving, it
58:11
absorbs your whole mind. But then as you train more and
58:16
more on driving get very good then you, the driving camera
58:20
becomes a background task. It doesn't it only absorbs a
58:23
small amount of your mental capacity because you have a lot
58:25
of training. So we can make up for the fact
58:29
that the inference computers, it's tiny compared to, you know,
58:34
A10 kilowatt Bank of GPUs because you've got a few 100
58:38
watts of difference compute. We can make up that with heavy
58:44
training. So yeah, that's and then there's
58:52
also vast amounts. The actual petabytes of data
58:55
coming in are tremendous. And then sorting out what
58:59
training is important with, you know, all of the vast amounts of
59:03
video training, video data coming in fleet, what is
59:07
actually most important for training that's also quite
59:12
difficult. But as I said, we're we're not
59:18
currently training compute constraints.
59:20
I'm sure you want to leverage. Yeah, like you mentioned the
59:25
training. Has more trained larger models
59:27
also to train quicker. But in the end, we've still got
59:31
to pick which models are performing better.
59:33
So the validation effort to picking the models because the
59:36
miles per intervention is pretty large.
59:39
We're driving a lot of miles and going close to, we do have
59:42
simulation and other ways to get those metrics.
59:45
Those two help. But in the end, that's the big
59:47
bottleneck. Yeah, that's why we're not
59:52
training computer constraint alone.
59:54
And there's other axis of scaling as well, which is the
59:56
data. Figuring out which data is more
59:58
useful. That that is an important task.
1:00:01
We're focusing on that. Yeah, so here.
1:00:05
So as it relates to the second part of your question, Pierre
1:00:08
about safety drivers and rolling it out, each state has different
1:00:11
requirements that you know in terms of how many miles and how
1:00:14
much time you need to have a safety driver and not have a
1:00:16
safety driver. We're going to follow all those.
1:00:18
We're not going to violate whatever regulations are out
1:00:20
there, but safety is a priority. But the goal is obviously that
1:00:23
when we're ready and safety is there, we'll we'll drive from
1:00:27
from the right chair. Yeah.
1:00:29
I mean, I guess like we, we think that we'll be able to have
1:00:34
driverless Teslas during paid rides next year, you know,
1:00:39
sometime next year. Arie, thank you.
1:00:44
And our next question comes from Adam Jonas at Morgan Stanley.
1:00:48
Adam, please feel free to unmute yourself.
1:00:52
OK. Thanks everybody.
1:00:54
Just had a question about the relationship between Tesla and
1:00:58
XAI. Many investors are still not
1:01:01
clear how the work at XAI is is truly beneficial to Tesla.
1:01:05
Some even take the view that the two companies may even be in
1:01:08
competition with each other in terms of talent and tech and
1:01:12
even your time. Elon, so what's your message to
1:01:14
investors on that relationship between Tesla and XAI and where
1:01:18
do you see it going over time? Thanks.
1:01:24
Well, I should say that, you know, XAI has been helpful to
1:01:28
Tesla AI, you know, quite a few times in terms of, you know,
1:01:34
things like scaling up a lot like training, You know, just
1:01:39
even like recently in the last week or so, improvements in all
1:01:44
toner training where if you you're doing a big training 1
1:01:47
and I know it fails, being able to continue training and and
1:01:56
easy to recover from on a train when the exam has been pretty
1:01:58
helpful. So, but there are different
1:02:04
problems. You know, XAI is working on
1:02:07
artificial general intelligence or artificial super intelligence
1:02:12
tells us trying to make autonomous cars and autonomous
1:02:15
robots. They're different problems, so
1:02:22
yeah, I mean. I think we've said this before
1:02:25
also like all, not all AI is equal, right?
1:02:28
I mean, there's AI is a broad spectrum, yeah.
1:02:30
And we have our old swim lengths.
1:02:32
Yeah. There are certain things which
1:02:34
we can collaborate on if needed, but for the most part we're
1:02:38
solving different issues. Yeah, it doesn't tell us focus
1:02:42
on real world and and that was saying it is quite a bit
1:02:46
different from because you have massive context in the form of
1:02:53
video and some amount of audio that's going to be still like
1:02:58
with extremely efficient inference compute.
1:03:03
I do think tells us the most efficient in the world in terms
1:03:09
of inference compute like because out of necessity, we
1:03:13
have to we have to be very good at inferred efficient inference.
1:03:22
We can't pretend 10 kilowatts of GPUs in a car.
1:03:25
We've got a couple 100 watts. And so, you know, it's pretty
1:03:30
well designed Tesla AI Cho, but it's still a couple 100 watts.
1:03:39
But there are different problems.
1:03:40
I mean, it's just, you know, like the stuff that XI is like
1:03:45
when it's running in ferns. I mean, it's it is running in
1:03:49
ferns like answering questions, answering testing questions on a
1:03:53
on A10 kilowatt rack. It's like, yeah, put that in the
1:03:57
car. It's a different problem.
1:03:59
Please. No, no, exactly.
1:04:03
So you know, XAI is because because I felt there wasn't
1:04:09
there wasn't a truth seeking digital super intelligence
1:04:13
company out there. Like that's what it came down to
1:04:15
that they needed to be a truth secret like like an AI company
1:04:20
that is very rigorous about being truthful.
1:04:24
And so I'm not saying XAI is perfect, but that's but that
1:04:27
that is at least the explosive aspiration.
1:04:31
Even if something is, you know, politically incorrect, it should
1:04:34
still be truthful. I think this is very important
1:04:38
for AI safety. So anyway, I think AI XAI will
1:04:44
it has been helpful to Tesla and will continue to be helpful to
1:04:47
Tesla, but there are very different problems.
1:04:52
Great. And and and I mean like like
1:04:55
what is like what, what are the car company has?
1:04:57
It has a world class trip design team.
1:05:02
Like 01 of the car company has a world class AI team like Tesla
1:05:06
does 0. Those are all startups.
1:05:10
They're creative from scratch. Great.
1:05:15
Thank you, Ilan. And I think that's unfortunately
1:05:17
all the time that we have for today.
1:05:18
And we appreciate all of your questions and we look forward to
1:05:21
you next quarter. Thank you very much and goodbye.
1:05:28
Hey, thank you so much for listening today.
1:05:29
I really do appreciate your support.
1:05:31
If you could take a second and hit the subscribe or the follow
1:05:34
button on whatever podcast platform that you're listening
1:05:36
on right now, I greatly appreciate it.
1:05:39
It helps out the show tremendously and you'll never
1:05:42
miss an episode. And each episode is about 10
1:05:45
minutes or less to get you caught up quickly.
1:05:48
And please, if you want to support the show even more, go
1:05:50
to patreon.com/stagezero and please take care of yourselves
1:05:56
and each other and I'll see you tomorrow.