Latest / Elon Musk Podcast / Elon Musk Wows Tesla Crowd With Future Predictions

Transcript

Elon Musk Wows Tesla Crowd With Future Predictions · Elon Musk Podcast

0:00
So those of you who have been following Tesla closely
0:04
understand like you get it, you know, and so, so like important
0:12
thing is you get it. I mean, it's so the yeah, where,
0:18
where things are going. I think it's just going to be
0:21
absolutely mind blowing. You know, we're, we're obviously
0:25
making great progress in solving the sustainable energy problem,
0:30
you know, with electric vehicles, with stationary
0:32
storage, with solar. And and then I think in terms of
0:40
the the value of the company autonomy is just such a mind
0:46
blowing thing. That's like I said, you guys
0:49
understand, but I think most of the world does not yet
0:52
understand. So, and you know, and talking to
0:57
a lot of those sort of big institutional investors when
1:00
they're often in like New York and they don't drive cars.
1:05
So I'll be like, have you driven our car?
1:08
Have you, have you tried self driving, you know, the, the
1:11
version 12.3 and they're like, oh, no, OK, well, you should try
1:18
it. That would be a good thing to
1:19
do. And, and if you just plot the
1:23
points on the curve of how well autonomy is progressing and just
1:27
believe the curve, it's headed towards unsupervised full self
1:34
driving very quickly at an exponential pace.
1:37
If you just, if you just, yeah, it's, it's, I mean, yeah.
1:40
So it's, it's, it's, it's really as simple as that.
1:49
And in fact, I'd, I'd invite you to just do it personally, just
1:52
say, OK, with each release, how many miles do you drive before
1:56
you have to intervene? It's, it's, it's like literally
2:00
that, that simple. And with each release you'll see
2:05
it, there's a, there's a big improvement, big improvement.
2:08
And it looks like an exponential.
2:10
So, and it's very clear that that will actually go to the
2:13
point where it is actually far safer than a person driving the
2:16
car. So, and, and I have some insight
2:23
into, you know, the the next releases because we, we've,
2:27
we've got basically for every release that's on the road, we,
2:31
we kind of see what the next generation releases and a little
2:34
bit about the release after that.
2:35
So we have some insight into where it's going and it's really
2:37
amazing. No question whatsoever.
2:40
That will far exceed human safety.
2:44
Yeah, no question. So, and I think, you know, I
2:54
have, I have mentioned this bit, but like sometimes it's, it's
2:58
like helpful to sort of reiterate and connect some of
3:02
the dots because sometimes people wonder, well, how do you
3:08
go from this big fleet to actually monetizing the fleet in
3:13
an autonomous situation? And it's, it's actually a
3:16
combination of like it's like Airbnb and Uber to some degree.
3:21
So like there'll be some cars that Tesla owns itself.
3:24
It's kind of like an Uber fashion, But then for the fleet
3:27
that is owned by our customers, it will be like an Airbnb thing.
3:31
You can add or subtract your car to the fleet whenever you want.
3:35
So you can say like, I'm going away for a week at just one tap
3:39
on your Tesla app, your car gets added to the fleet and it just
3:42
makes money for you while you're gone.
3:45
Yeah. So you can say you, you can add
3:53
it to the fleet for a few hours, for a few days, for a few weeks,
3:57
whenever you want it back and say come back and Carl will come
4:00
right back. And I'm I'm highly confident
4:05
that it will far exceed the value of like the revenue made
4:08
by the the owner of the car will far exceed the actual monthly
4:12
payment. And then Tesla will obviously
4:14
take a rough share on that, but most of the money will go to the
4:17
the owner of the car and it's this is actually going to work.
4:21
This is what will happen. So you, I mean, I'd mark my
4:24
words, this is this is simply a matter of time.
4:27
Now, admittedly, I'm a little optimistic sometimes, you know,
4:32
I'm, I, you know, so I don't have complete lack of
4:34
self-awareness, but if I wasn't optimistic, this wouldn't exist.
4:39
This factory wouldn't exist, you know, So, so you know, that's
4:53
what my brother, who's awesome, he's here got just got
4:56
reelected. No, sorry.
4:57
Those things. So it's like a little anecdote,
5:06
like when we were kids in school, my, my brother would
5:09
tell me the the wrong time for the bus so that I wouldn't miss
5:12
the bus. And then I'd be upset with him.
5:16
Like, why'd you tell me the bus was going to come earlier than
5:18
it was? And, and, and he's like, well,
5:20
because otherwise you'd be late for the bus.
5:22
This is an actual thing that would happen.
5:25
So I'm, I, I guess I've been sort of pathologically
5:27
optimistic from, you know, from birth, but, but that's the
5:32
reason we're this is all the stuff is happening anyway.
5:34
It's like I got to be somewhat pathologically optimistic, but I
5:38
do deliver in the end. That's the important thing.
5:49
So yeah, and it's, it's really just going to be something else.
5:57
I mentioned this before, but like the for Arkhanvest, I'd
6:00
recommend reading the Arkhanvest analysis That's the most
6:04
accurate there may be, you know, more accurate ones, but the most
6:06
accurate sort of analysis of the potential of autonomy, the most
6:13
accurate that I'm that I'm aware of is is Kathy Woods Arkhanvest.
6:18
So, and if I recall correctly, they they predicted somewhere
6:23
over a $5 trillion valuation for Tesla just based on vehicle
6:28
autonomy, not counting optimists.
6:31
Yeah, yeah, I agree with that. So, so I think, I think, I think
6:47
just based on vehicle autonomy, we can, we can 10X110X the value
6:51
of the company. I believe that's what will
6:55
happen. Yeah, I know.
6:56
It's just what what do you know? It's it's it's 4:20 PM.
7:02
Just noticed that. So you know, so so I think
7:16
that's actually, I think that's that's pretty accurate.
7:22
But there's there's it actually gets way crazier when you think
7:28
about the the optimist robot, which is really a humanoid robot
7:34
that is intended to, you know, be able to do anything you want
7:39
it to do to be, you know, it's, you know, your companion.
7:44
It can be at your house, it can sort of babysit your kids.
7:48
It could teach them be a teacher.
7:52
It it, you know, it can do factory stuff.
7:57
Like I think that the ultimate ratio of say, how many super
8:03
useful humanoid helper droids do you want?
8:06
Like who doesn't want to see 3PO, you know, you know, but
8:11
AC3PO plus R2D2 plus, you know, plus plus it would be pretty
8:17
awesome. I think.
8:19
I think everyone in the world is going to want one, like
8:22
literally everyone. And, and then there'll be
8:26
obviously robots in industry making stuff.
8:31
And so I mean, I, I think the ratio of humanoid robots to
8:36
humans will probably be at least 2 to one, something like that.
8:39
One to one for sure. So which means like somewhere on
8:42
the order of 10 billion humanoid robots, maybe maybe 20 or 30.
8:49
And so then it's like, OK, well, let's say, you know, you kind of
8:54
make, let's say the build rate is, I think the build rate will
8:59
be probably something ultimately like a billion a year humanoid
9:02
robots like actually, and if Tesla just has a 10% share of
9:08
that and it might be a lot more than 10%.
9:12
And there's, you know, we make like 100 million optimist units
9:16
a year. I just, I mean, for reference,
9:18
the auto industry is roughly 100 million vehicles per year.
9:23
So that, you know, sort of similar ballpark at least within
9:27
an order of magnitude. And I, I, I think we could make
9:33
one for a cost of maybe at, at, at really high scale of about
9:38
$10,000. It's, it's, it's smaller, it'd
9:40
be less expensive than a car. So, and I think if you sold it
9:46
for sell it for $20,000 or something, this is at large
9:48
scale volume, Tesla would basically make about a trillion
9:52
dollars of profit a year from that.
9:56
So yeah, if the price earnings multiple is say on a 20 or 25
10:06
something like that, that would mean a $20 trillion market cap
10:09
from Optimus alone and and probably 5 to 10 from autonomous
10:16
vehicles. So like, I think it's actually
10:22
conceivable it's when within the realm of possibility for Tesla
10:27
to achieve evaluation 10 times that of the most valuable
10:29
company today. So, so when I say this is like
10:42
we're starting a new book, I mean, it's going to be the best
10:45
book. Next level, next, next, next,
10:50
next level. Now we need to make sure these
10:54
robots are nice to us. And you know, that's very
10:57
important. So, so anyway, that's so with
11:05
that, let me get into the presentation.
11:09
Blah, blah, blah, blah, blah, blah, blah, blah.
11:17
Oh man, I, I think it might be worth us just putting a word
11:24
limit on future shareholder Tesla proposals.
11:30
It's like, it's like, like you can't have like a novel length,
11:34
you know, it's correct, but like you can't have a whole novel as
11:36
a, as a proposal. So, you know, as you can see,
11:42
our impact is accelerating, You know, we're, we're starting to
11:44
make a real noticeable bent in COVID emissions.
11:49
So we're, we're really, you know, making a lot of progress
11:51
towards achieving our goal of sustainability.
11:54
And one of the, the, the effects of autonomy is actually is going
11:59
to be an even greater effect on carbon emissions.
12:03
Because I, I, I suspect that we'll go from passenger cars
12:08
having about 10 hours of usage per week, about an hour and a
12:11
half per day to probably a third of the hours in the week, maybe
12:15
5055 hours, which means the same car will, will be used five
12:20
times more often. So, so you need a certain amount
12:23
of resources to make the car, but it'll have five times the
12:26
usage. In my opinion, I think that's
12:29
quite likely, which will be an even more dramatic impact on
12:33
carbon emissions. So, you know, we, we're, we care
12:40
a lot about sustainable manufacturing.
12:43
Our, our factories are beautiful.
12:44
You walk around the factory, it's, it's, it's actually, you
12:48
know, contrary to these, like, what do you mean the press?
12:50
It's like a good vibe in the factory.
12:52
Like people are smiling and happy.
12:54
It's like, it's nice. So, so yeah.
13:09
So I mean, we do care. We actually do care a lot.
13:13
It tells us about doing the right thing.
13:14
And we, you know, we're not going to be perfect, but we do
13:17
care a lot about doing the right thing.
13:20
So our vehicles are water efficient, energy efficient.
13:24
You know, we try whenever possible to use renewables for
13:28
powering the factory and you know, do do our best to do the
13:32
right thing. Our batteries are lasting a lot
13:37
longer. I got too many things, man.
13:42
Anyway, the batteries that last lasting a lot longer, which is,
13:45
which is good safety is is is good in the factory.
13:54
We're improving the affordability of BBS, you know,
13:57
model wise comparable to BMW X3, but it's actual cost of
14:00
ownership on a monthly basis is much lower.
14:06
We were number one in EVs globally last year.
14:09
So, yeah, so yeah, so let's see and grass the Tesla team on
14:27
making 6,000,000 vehicles all time.
14:29
So, so that I mean the Tesla fleet is really becoming very
14:38
substantial. I mean it's going to be 7
14:40
million vehicles by the end of this year, over 7 million
14:43
vehicles. So we had, and our, our factory
14:52
in Fremont is currently the highest volume order order
14:56
factory in North America. So, and, and we broke the prior
14:59
production record from when it was new me.
15:02
So it's, it's actually pretty wild that there's a, we have a
15:05
giant car factory in the San Francisco Bay Area.
15:08
It's, it's like it's not exactly the cheapest place to have a car
15:11
factory. It's like that and the Swiss
15:13
Alps or something, you know, but, but we, we still managed to
15:17
make, you know, great cars at high volume.
15:20
And that's a testament to the, the great team we've got in
15:22
Fremont. So Congrats to the Fremont team.
15:35
And yeah, a lot of people were saying the cyber truck is like
15:38
fake. It's never going to come out,
15:40
you know, and now we're shipping a lot of cyber trucks.
15:44
We we had a, a weekly record of 1300.
15:47
So and you know, I think that with a cyber truck, it really is
16:00
something special. Like if you'll sometimes have
16:04
like different opinions on the cyber truck.
16:07
But if you really want to know if there's a, if something is,
16:11
is cool, if it's, if it's a great product, like show it to a
16:14
kid, OK, like no filter. OK, the kids got like no filter,
16:21
like a 5 year old, 6 year old, something like that and or even
16:24
3 year old and say, which car do you like Cybertruck?
16:31
So it's like that's how you know, and it's it's finally
16:38
something that it it just looks like the future.
16:41
It's and it drives so well. It's it, you know, it's
16:47
comfortable. It's it's just a fantastic
16:49
product. I think it's our I think it's
16:51
our best product and it's a lot of innovation.
16:55
It's got a 48 Volt low voltage system finally, instead of 12
17:02
volts, 800 Volt drivetrain, the world's biggest castings.
17:07
You know, it's bulletproof, all, all the cool things can, you
17:09
know, out pull AF350 diesel, which is a very impressive truck
17:14
and it can, you know, it, it can do a quarter mile faster than a
17:20
Porsche 9/11 while towing a Porsche 9/11.
17:24
So that's insane. So yeah.
17:36
And we think if you think about it like how how often do
17:39
companies make products that move your heart that are really
17:45
special, It's it's so rare. And I think this is one of them.
17:49
So yeah. And we launched the upgraded 3,
18:01
which is actually a fantastic car.
18:03
I recommend trying it out. It's it's really a great car and
18:09
it's only $216 after a gas savings like total cost of
18:13
ownership. It's basically 200 and something
18:15
dollars when compared to a gasoline car.
18:18
Some like gasoline car. So it's it's it's really a great
18:23
deal and a fun car. And the performance multi
18:27
performance is well, it's it's like, you know, fast and a
18:31
Porsche 9/11 it's just a great, great car.
18:38
So yeah, So yeah, yeah. And of course model Model Y
18:52
became the best selling vehicle globally.
18:54
And this this is something that we predicted.
18:56
So, you know, like I said in like 2022, it would Model Y
19:02
would be the biggest the car in the world by dollar volume
19:06
sales. And at 2023 it will be the
19:08
biggest in unit volume. And it was and again this year
19:14
it will be the best selling car on earth.
19:17
So yeah. And we obviously we've got the
19:30
Tesla Semi. You know, we're we're in low
19:33
volume production of the Tesla Semi.
19:34
And just last week, I've I approved the plans for volume
19:40
production of the Tesla Semi. So we're we're going to make a
19:55
lot of semi trucks. So we're we're going to start
19:58
off with, yeah, it's, it's, it's something I think it'll actually
20:03
move the needle financially. It's not like it's not a small,
20:05
small thing. And the thing about like
20:09
commercial vehicles is that. Companies that use them are
20:12
super objective. It's not like it's actually, I
20:15
mean, obviously we're going to make it have cool style and be
20:19
an awesome car. But for commercial vehicles, the
20:22
companies that make the buying decisions, they they just look
20:25
at say like, what are the numbers?
20:26
Like is this cost less to transport or it costs more to
20:30
transport than than say a diesel truck?
20:33
And The thing is that economics are much better than than diesel
20:36
truck. It's kind of basically a no
20:38
brainer. If you're a transport company,
20:40
you don't use an electric, the Tesla electric semi, you're just
20:43
losing money. Why would you do that?
20:46
Do you not like money? OK, but if you do like money,
20:52
then I recommend using the Tesla semi.
20:57
And so, so I think this is, this is really going to actually sell
21:08
at a scale that people will be surprised at and, and we'll
21:11
actually move the needle financially and, and do a lot of
21:12
good and also for, for CO2 and, and, and sustainability because
21:19
semis are driven all the time, actually even at, at much
21:23
smaller numbers, they have a, a very big effect on total carbon
21:27
emissions solved. So obviously we've got some new,
21:34
new products that we're working on under the covers and I think
21:41
these these are, I think this is going to be pretty special.
21:46
So, you know, some of them I think people at maybe at first
21:51
may think, oh, it's not going to be that amazing, but just wait,
21:57
it will be so, so our supercharge network is
22:13
continuing to grow. You know, rumors of the death of
22:17
Supercharger are greatly exaggerated.
22:20
We, we are in fact continuing to grow the supercharged network
22:24
significantly. In fact, this year I think we'll
22:26
put more, we'll deploy more Superchargers this year that are
22:31
actually working than the rest of industry combined, just FYI.
22:39
So now, now we are going to be more careful about like you
22:47
know, the capital efficiency of where we deploy supercharges.
22:52
But for sure any place that has congestion, any sort of missing
22:56
parts of the map that we're missing, we're going to put the
22:58
superchargers there. So even for the remainder of the
23:02
year, we, we expect to spend about half a billion dollars on,
23:06
on supercharge of deployment. So it's, it's very significant
23:09
and to be a well spent half a billion.
23:11
So this is a, you know, it's, it's a lot.
23:16
So, and we're this is actually not showing a bunch of new
23:20
markets that will be opening up later this year and next.
23:24
So we're obviously what we want Tesla to be worldwide.
23:27
And this is a lot of countries that we, you know, have zero to
23:32
very few Tesla's. And we obviously we, we want
23:34
Tesla everywhere. So there are a lot of new
23:36
markets that we're going to open up later this year or next.
23:43
We're also opening the Supercharger of Superchargers to
23:47
other companies with the that's, that's the adapter to the old
23:54
somewhat somewhat clunky connector.
23:57
But our goal is to be helpful to other car companies.
24:01
So any other car companies that are going electric, we do want
24:04
to, we do want to be helpful to them with our Supercharger
24:06
network. So we're providing them with
24:09
adapters and enabling them to have access to the Tesla
24:13
Supercharging network because we think it's better to do that
24:16
than to create a walled garden that inhibits EV growth.
24:28
Oops. There's a bit of latency here.
24:32
And we're also innovating a lot in battery production with the
24:35
4680 that's built right here. And it is it's a hard problem,
24:41
you know, like there, there are entire companies that just make
24:45
battery cells. That's all they do.
24:48
And but we're making good progress, like all the cyber
24:51
trucks that you see use the Tesla 4680 cell and that we have
24:57
a clear path to the 4680 being, we think probably the most
25:02
competitive cell from it's from a manufacturing efficiency
25:06
standpoint. And so it's, it's, but it is, it
25:11
is a hard problem. I have to say there's, there's
25:14
quite a lot of brain damage required to be good at cell
25:17
manufacturing. It's like a brain damage high.
25:22
But I think it does, it does give Tesla resilience in the
25:27
face of if there are changes in the, I don't know, geopolitical
25:32
situation somehow, but it like it's good to have so many
25:37
independence in cell manufacturing.
25:42
So, yeah, and we're making steady progress.
25:45
And we also have the the cathode refinery, which you can see
25:48
behind the the, the main factory.
25:50
So you can sort of see it in the picture there.
25:53
That's a cathode refinery. And then we've got the lithium
25:56
refinery in South Texas. And so we're just making sure
25:59
we've kind of got the piece of the puzzle and delete.
26:05
I mean, if you were to look at sort of a video of how Tesla
26:09
does say cathode and, and lithium refining and how the
26:12
rest of industry does it, it's night and day.
26:14
I mean, you can sort of eat off the floor in the, in the Tesla
26:17
refinery. And I would not recommend doing
26:19
so in the others. So, you know, it's, it's really,
26:23
it's, it's, it would definitely go very, very vertical here.
26:29
But I think this, this is a, it's a wise investment that will
26:32
that will pay off more than people realize.
26:35
So I don't think this thing is
26:51
working. So this year we're also on track
26:57
to complete a massive number of energy deployments.
27:01
So that is a gigantic increase in in deployment capability.
27:08
So we're really, we seem to be tracking to sort of a 2 to 300%
27:15
year over year growth and and energy storage deployment and
27:19
stationary pack. So it's giant.
27:23
And the, the limiting factor really is being able to build
27:26
more mega packs and build more power walls.
27:30
So we're ramping up production of the power Wall 3, which is
27:34
really a game changer for at, at the personal level, the
27:37
Powerwall 3, it usually takes about 3 iterations of, of any
27:42
given technology to, to really for, for it to be something
27:47
where it's like, OK, now we're really hitting the sweet spot.
27:49
So Powerwall 3 is like an epic product.
27:53
The Mega Pack is also, it's even it's, I'd call it sort of
27:58
iteration 2 of mega Pack is, is also an epic product with, with
28:03
the mega pack three, which is on a probably a couple years away.
28:09
We'll we'll start actually absorbing more and more of the
28:13
substation of the, of the, of the sort of power electronics.
28:17
So I sort of think you want to get to the point with the mega
28:20
pack where you can literally just take the high voltage power
28:22
lines and plug them in. Just there's no substation.
28:25
We just just plug them in, just drop them down and drop it down
28:29
and plug it in. That's, and now it works, which
28:35
is like mind blowing for the utility industry, by the way.
28:38
Like they're like what? Yeah, usually just plug the
28:43
wires in and it'll work for, for for very high voltage.
28:48
So that that I think is actually going to be a major.
28:53
So this is anyway. So, yeah, this is this is great,
28:56
excellent work by the energy storage team.
29:08
Services and other are also looking good.
29:11
So we're actually now making money on the services and other
29:14
stuff. So this is a kind of a big deal,
29:22
right? Yeah.
29:24
So see. Yeah.
29:27
And Tesla's obviously way more than a car company.
29:30
We've, you know, we've, we do a lot of software at Tesla, you
29:34
know, about, I don't know, roughly half the, or a very huge
29:39
percentage of the, of the engineering we do is actually
29:41
software engineering. So Tesla's, I'd say as much a
29:45
software company as it is a hardware company.
29:47
This is a very big deal because car companies are not software
29:52
companies normally. So this, this matters a lot.
29:55
So things like our auto bit of software for energy storage, all
29:59
the software that controls the cars, the mega packs, the the
30:01
power walls of the solar at the obviously for AI and and full
30:08
self driving are big deal insurance service and collision.
30:14
That Tesla also writes a lot of software internally that that
30:19
helps that like we call like the Tesla operating system
30:21
internally that is head and shoulders above what any other
30:26
company has. I think probably better than any
30:28
Fortune 500 company. The Tesla internal software is
30:33
is just way better. So it's yeah, it's just far more
30:39
than what people normally think of as a car company.
30:44
And Tesla's also the world the leader in real world AI.
30:48
So there's really this is a big deal like Tesla is a head of of
30:55
Google Matter, you know, opening eye anyone on real world
30:59
software actually look, taking in video and making decisions
31:03
based on video. No one can no one is even close.
31:08
And it's getting better, I'd say with each passing month, if not
31:14
each passing week. So yeah,
31:30
and and it's it's it's also worth noting that Tesla is
31:33
actually pretty good at chip design.
31:36
So the, the AI inference chip that's, that was designed by
31:40
Tesla, that's in cars. We had sort of our hardware 3 AI
31:45
in furnace cars made in the, you know, over the past year have
31:48
the hardware 4. We've just completed design on
31:52
hardware 5, which we're now calling AI 5 because it, it's,
31:58
it's really is, it's, it's, it's, there's still actually is
32:01
not a, a chip from NVIDIA or from.
32:05
And I have a lot of respect for NVIDIA from any company that we
32:09
would prefer to put in our car that is better than what we have
32:12
in the car. So, so we, we, we started from
32:24
scratch in chip design, just as we started from scratch in AI
32:28
software and have the best real world AI software and the and
32:32
the best AI and furnace chip in the world from nothing.
32:36
So this is, you know, a big deal.
32:44
And the, the capability of the of the, of the chips in the car
32:48
is, is dramatic. Like, I mean, right now all the
32:51
cars are actually training like we, we have hardware four run
32:56
hardware 3IN emulation mode. So we'll continue to make
33:02
significant progress in hardware 3, but later this year we'll
33:06
actually bifurcate, continue working on hardware 3, sort of
33:10
training on hardware 3, but but then do separate training on
33:15
hardware 4. That'll be the the sort of
33:17
training cluster that we're building at the South side of
33:19
the the Gigafactory and that that'll be dedicated hardware 4
33:25
video and inference. And you know, so that the
33:31
hardware 4 has cameras that have four or five times better
33:35
resolution. And depending on how you count
33:38
the, the sort of hardware 4, it's, it's about anywhere from 3
33:43
to 8 times better than hardware 3.
33:46
So it's, but, but everything you're seeing thus far is just
33:48
hardware 3. And we still have a long way to
33:50
go before we get, we reach the limits of hardware 3.
33:53
So hardware three, I think will, will still will do amazing
33:55
things, but but hardware four, I think we'll probably do about
34:00
five times better then hardware 5, which comes out in about 18
34:06
months or so is 10 times more capable than hardware 4.
34:11
It's it's like a, so if you know NVIDIA hardware, it's it's sort
34:15
of AB200 class computer. So you know, that's like this.
34:26
So what we'll just progressively do is improve how many nines of
34:29
reliability the car gets. And I and then of course that
34:35
will go into Optimus as well. So the same chip will go into,
34:39
you know, hardware fog will go into is, is an Optimus hardware
34:46
AI5, which we're calling switching from hardware 5 to AI5
34:49
will be an Optimus and in, in all cars in in about 18 months.
34:54
And it's really just a staggering amount of compute and
34:58
it's very, it's very power efficient to compute.
35:00
So it's, it's got to be because if you're in a mobile
35:03
application like a humanoid robot or a car, you, you can't
35:06
just be sucking down 10 kilowatts, you know, like you
35:09
can in a data center. So you you've got to be very
35:12
power efficient. So the, you know, sort of
35:18
hardware three and four are only a few 100 watts.
35:23
Yeah, Yeah, this is hard. Now hardware 5 will will be able
35:35
to go probably up to about 7 or 800 watts It but it'll, it'll
35:39
power fluctuate contingent upon the complexity of the scene that
35:43
it is that it is in. So if it is in a, you know,
35:46
parking lot stationary, it's like, you know, you know, you
35:49
don't have to think very much just like a person, like if
35:51
you're in a complicated traffic scenario, you've got to think a
35:54
lot more than if you're just cruising along, you know, on an
35:58
empty Rd. But something that I think is
36:03
potentially interesting down the road is like at some point the
36:07
Tesla fleet I think will probably be, you know, over 100
36:10
million vehicles. And if each vehicle has a
36:14
kilowatt of efficient inference compute, I think there's a,
36:21
there's, I think there's an sort of an Amazon Web Services, AWS
36:25
type opportunity. Because if you got 100 million
36:28
vehicles with a kilowatt of efficient inference compute,
36:33
you've got 100 gigawatts of compute, like 100 gigawatts of
36:38
computers a lot. And it's distributed all over
36:40
the world. So Even so, when the car is not
36:43
in autonomous mode, which I think probably is, you know,
36:48
doing robo taxi work maybe 50-60 hours a week, but about 100
36:53
hours a week, it's it's not, it's probably stationary.
36:57
So, so there's 100 hours of 100 gigawatts of Infernus compute,
37:03
which I think we should use. Why not?
37:09
You know, when, when, when people looked at Amazon, we
37:12
started out obviously as an online bookseller and it's on,
37:16
you know, it's going to be like this incredible place where you
37:22
can buy anything. And, and then they, and then
37:26
Amazon Web Services like, well, they got all these computers
37:29
that only really see peak usage sometimes, but what are they
37:35
going to do when it's not peak usage?
37:38
And sometimes the Amazon servers are down like 10%.
37:40
So that's when they said, well, let's do Amazon Web Services.
37:44
And then Amazon Web Services became more valuable than the
37:47
entire rest of Amazon. And, you know, I think there's,
37:52
there's some kind of opportunity there that's pretty significant
37:55
for, for Tesla down the road. You know, again, that's really
38:01
nobody's really factoring that in, but I think that that
38:05
actually will be quite significant.
38:08
We also know are no longer compute constrained for
38:11
training. So I, I check in with the team,
38:16
it's like, is there anything we could do to improve the pace of
38:20
progress with respect to training and inference?
38:23
And currently that is not the limiting factor.
38:25
In fact, the limiting factor right now is that the, the
38:29
amount of miles between interventions is so long that it
38:32
takes quite a while to figure out which version is better than
38:36
the other version because none of them are requiring any
38:39
interventions. So it's like, you know, if
38:42
you're not getting to like thousands of miles between
38:44
interventions or you're like 10,000 miles to get an
38:47
intervention, then like, well, the average person only drives
38:50
about 10,000 miles in a year. And if it's in an urban
38:54
environment and the average speed is 20 miles an hour, I
38:58
mean, so our professional test drivers get pretty bored,
39:01
frankly. You know, they, they're like,
39:04
OK, I drove all week and there was no intervention.
39:07
Like, like the highlight of the week would be like, yes, an
39:09
intervention. Finally, it's like getting to
39:12
that point. So this is where actually having
39:17
a giant fleet is extremely important because we can deploy
39:22
a new FSD model and run it in shadow mode and see what see how
39:25
well it performs. You know, compare how the human
39:30
drives the car versus the the new self driving build and and
39:36
then analyze that delta in shadow mode, like the shadow
39:39
knows or doesn't know, as the case may be, and and then be
39:44
able to assess by getting billions of miles very quickly
39:50
with the giant fleet like that. Basically that that data engine
39:53
is incredibly helpful. Like I actually, it's not
39:57
possible to solve the self driving problem without having
40:00
millions of vehicles on the road.
40:02
So, so that's actually the, it's like figuring out clever ways to
40:08
test which how good the next build FSD build is, is, is
40:15
actually the limiting factor right now.
40:17
And then of course, like I said, we are building another training
40:20
data center right here, which will be dedicated to Hardware 4
40:25
training. So we'll, we'll bifurcate
40:28
Hardware 3 and Hardware 4 training later this year.
40:31
We'll, we'll keep improving Hardware 3, but we're going to
40:34
uncork the full capability of Hardware 4 as well.
40:39
So yeah, I think, I think most people here have tried out
40:50
version 12 as we did say like sort of unsupervised full self
40:58
driving, full self driving would be version 12.
41:01
So we we are actually just keeping the version arbitrarily
41:05
at 12 and then like pulling a 12.412 point 312 point 412.5.
41:10
But it's actually really like version 13, version 14.
41:14
But anyways, this is an arbitrary designation, so 12.4
41:19
is actually like a whole different version than than 12.3
41:23
and 12.5 is a whole different version 12.4.
41:27
So you'll see really giant improvements.
41:32
I think sometimes factor of 10 improvements between a
41:36
successive versions. And then as I mentioned, the way
41:46
it'll work for the existing fleet is are we able to add or
41:49
subtract your car to the fleet as as you'd like?
41:53
And, and there will also be Tesla owned cars.
41:58
So it'll be a combination of like Uber plus Airbnb, but it'll
42:04
be pretty wild that the, that there'll be a software update
42:06
and then the whole, the whole fleet suddenly becomes
42:09
accessible. It's like suddenly you've got
42:12
7,000,010 million cars that can ultimately 10s of millions of
42:17
cars that can do autonomous driving and instead of being
42:22
used 10 hours a week, can be used 50 or 60 hours a week.
42:38
So it's it's pretty wild to just, you know, be in Palo Alto
42:41
with a bunch of cubes and then a humanoid robot just walks past.
42:49
We've made it a massive amount of progress with Optimists in a
42:52
short period of time, from someone pretending to be a robot
42:55
dancing in a suit to a pretty hodgepodgey robot, to a robot
43:01
that is actually doing useful tasks in the factory today.
43:06
So we have two optimist robots in our Fremont factory that are
43:09
doing basically this task, which is taking cells off the end of
43:13
the line and placing them in a shipping container.
43:17
And yeah, we actually have quite a few of these cruising around
43:21
our offices in in Palo Alto. So there's and, and I think
43:29
we've got kind of like one major hardware revision, which should
43:34
be done by end of this year or early next before.
43:39
And and then we'll move into a limited production next year of
43:42
Optimists limited production for use in our factories where we'll
43:45
test out the product kind of, you know, you know, as I said,
43:50
sort of eat our own dog food or whatever the electronic
43:53
equivalent of that is. So, but I think like next year,
44:00
my prediction is next year we'll have over over 1000, maybe a few
44:04
thousand Optimus robots working at Tesla and, and, and things
44:18
are going to scale up very rapidly from there.
44:23
So we'll iron out the bugs. It'll like the degree of
44:28
autonomy will be radically better.
44:31
You'll just literally be able to talk to it and say, please do
44:33
this task or I'm going to show you something now do that the
44:37
thing that I'm showing you and, you know, get to the point where
44:43
it can watch a video of something like a person and and
44:47
then learn just by looking at that video and and do that task.
44:52
So yeah. So really it's going to be quite
44:57
something. And and I'm confident of the
45:02
prediction that there will be more like the ratio of humanoid
45:10
robots to humans will be greater than one to 1.
45:13
So there'll be, you know, more than 10 billion humanoid robots
45:16
in the world, probably 20 or more.
45:20
And Tesla's going to be by far the leader in that.
45:24
You're seeing a lot of robot startups and, but I think it's
45:30
actually very challenging to, to do Optimus as a robot startup
45:35
because what we found to make Optimus work, we've had to
45:40
design, design from first principles from scratch every
45:45
part of the robot. So the motor, the gearbox, the
45:48
sensors, the power electronics, the communication system,
45:54
everything had to be done from scratch.
45:56
We, we found that there's basically nothing.
45:58
There's no supply chain. So even though there are many
46:02
electric motors made in the world, there's no supply chain
46:05
for the types of motors and sensors and gearboxes that are
46:08
needed for a humanoid robot. And you, I mean, what you're
46:14
saying here is our current generation hand and arm, but our
46:18
next, which has 11° of freedom, our next generation has 22° of
46:22
freedom. It will be able to play the
46:25
piano. So it's, it's really like, wow.
46:31
Now, of course, like I said, we need to make sure you don't have
46:33
the Terminator scenario. That's very important.
46:36
So safety of the humanoid robot will be very important, but
46:43
because it requires so much ground up design, designing
46:47
every motor, gearbox, sensor, power electronics from scratch,
46:50
it's, it's very hard for a startup to, if not impossible
46:53
for a startup to replicate that. But at Tesla, we have the
46:56
world's best electrical engineering.
46:57
I think we've got the world's best mechanical engineering for,
47:01
for gearboxes and, and, and for the electric motors, power
47:06
electronics. So, you know, we have the
47:10
resources to do that it it applies quite well.
47:13
And then you also have to have the, the brain you need the you
47:17
need the, the you need a power efficient inference computer,
47:21
which we've got for the car and we'll be using an optimist.
47:25
You need AI real, you need to be the best in real world AI and
47:29
tells us the best in real world AI.
47:31
So you need all of these. You need a very strong hand of
47:34
cards in order to make a compelling robot.
47:39
And then you also need to be very good at scale
47:41
manufacturing. So in order to have the robot
47:44
not cost like hundreds of thousands of dollars, in order
47:47
to make it cost like, you know, 10 or $20,000, actually need to
47:51
design for manufacturing and be very good at manufacturing.
47:56
And what in my experience prototypes are, are easy
48:00
compared to volume manufacturing.
48:02
Prototypes are easy. Production is hard, relatively
48:05
speaking. So Tesla has the production
48:08
capability, it has the engineering capability and has
48:11
the AI hardware and software capability and even the most
48:18
optimistic estimates that I've seen for for optimists, the
48:23
optimist optimist, I think under under count the magnitude of of
48:32
what this robot will be able to do.
48:38
You know, as I said at the beginning of the of the
48:41
presentation, I, you know, I agree with the Arkan best
48:44
analysis that autonomous transport is called sort of a 5
48:50
to $7 trillion market cap situation.
48:54
Optimus, I think is is a 25 literally $25 trillion market
49:00
cap situation. Now, I don't want to, I don't
49:10
want to trivialize, trivialize what's necessary to get there.
49:12
I mean, it's an immense amount of work that is required to get
49:15
there, like super difficult, but we are moving very fast down
49:21
that road. We're going to make it happen.
49:23
So thank you.