Latest / Elon Musk Thinking / Elon Musks Speech in Tesla Shareholders Meeting.

Transcript

Elon Musks Speech in Tesla Shareholders Meeting. · Elon Musk Thinking

0:00
Welcome, Welcome to the Tesla shareholder Meeting. I just want to start up
0:04
by saying, hot, damn, I love you guys. We have the
0:07
most awesome shareholder base. I mean, it's just incredible any public company,
0:11
it's incredible. Wow, We've got a great sort of shareholder of meeting here
0:15
to recap the achievements of the company and tell you about where we're going.
0:19
I think it's incredible. I think we're not just opening a new chapter for
0:23
Tesla, we're starting a new book. So those of you who have been
0:27
following Tesla closely understand, like you get it, you know, and so
0:32
like the important thing is you get it. I mean, it's so the
0:36
Yeah, where things are going. I think it's just going to be absolutely
0:39
mind blowing. You know, we're obviously making great progress in solving the sustainable
0:44
energy problem, you know, with electric vehicles, with stationary storage, with
0:48
solar, and and then terms I think in terms of the value of the
0:52
company, autonomy is just such a mind blowing thing that I said, you
0:57
guys understand, but I think most of the world does not yet understand.
1:00
So and you know, talking to a lot of these sort of big institutional
1:04
investors I mean they're often in like New York and they don't drive cars.
1:08
So I'll be like, have you driven our car? Have you tried self
1:11
driving? You know, the version twelve point three, and they're like,
1:15
oh no, okay, well you should try it. That would be a
1:19
good thing to do. And if you just plot the points on the curve
1:23
of how well autonomy is progressing, and just believe the curve, it's headed
1:27
towards unsupervisedbule self driving very quickly at an exponential pace. If you just yeah,
1:34
it's I mean, yeah, So it's really as simple as that.
1:38
And in fact, i'd invite you to just do it personally, just say
1:42
okay, with each release, how many miles do you drive before you have
1:46
to intervene. It's like literally that simple. And with each release you'll see
1:51
there's a big improvement and it looks like an exponential so and it's very clear
1:56
that will actually go to the point where it is actually far safer than a
2:00
person driving the car. So, and I have some insight into, you
2:02
know, the next releases, because we've got basically for every release that's on
2:08
the road, we kind of see what the next generation releases and a little
2:12
bit about the release after that, so we have some insight into where it's
2:15
going, and it's really amazing, no question whatsoever, that will far exceed
2:19
human safety. Yeah, no question. So, and I think you know
2:23
I have mentioned this a bit, like sometimes it's like helpful to sort of
2:27
reiterate and connect some of the dots because sometimes people wonder, well, how
2:30
do you go from this big fleet to actually monetizing the fleet in an autonomous
2:37
situation? And it's actually a combination of like it's like Airbnb and Uber to
2:43
some degree. So like there'll be some cars that Teesla owns itself. It's
2:46
kind of like an Uber of fashion. But then for the fleet that is
2:50
owned by our customers, it will be like an Airbnb thing. You can
2:53
add or subtract your car to the fleet whenever you want. So you can
2:57
say, like I'm going away for a week, just one tap on your
3:00
Tesla app, your car gets added to the fleet and it just makes money
3:02
for you while you're gone. Yeah, you can say you can add it
3:07
to the fleet for a few hours, for a few days, for a
3:09
few weeks. Whenever you want it back, you can say come back,
3:12
and car will come right back. And I'm highly confident that it will far
3:15
exceed the value, Like the revenue made by the owner of the car will
3:21
far exceed the actual monthly payment, and then Test will obviously take a rev
3:24
share on that, but most of the money will go to the owner of
3:27
the car. And it's this is actually gonna work. This is what will
3:30
happen. So, I mean, i'd mark my words. This is simply
3:35
a matter of time. Now, admittedly I'm a little optimistic sometimes, you
3:38
know, I'm you know, so I don't have complete lack of self awareness.
3:42
But if I wasn't optimistic, this wouldn't exist, This factory wouldn't exist,
3:46
you know. So you know, that's when my brother, who's supports
3:50
he just got reelected. No, sorry, this is so like a little
3:53
anecdote when we're kids, is cool. My brother would tell me the wrong
3:58
time for the bus so that I missed the bus, and then it'd be
4:00
upset with him, like why did you tell me the bus was going to
4:02
come earlier than it was, And he's like, well, because otherwise you'd
4:05
be late for the bus. This is an actual thing that would happen.
4:09
So I'm I guess I've been sort of pathologically optimistic from from you know,
4:14
from birth. But that's the reason where this is all the stuff is happening
4:16
anyway, It's like I gotta be somewhat pathologically optimistic, but I did live
4:21
in the end. That's important thing. So yeah, and it's it's really
4:25
just going to be something else. I mentioned this before, but like the
4:29
for ARC invest, I'd recommend reading the ARC invest analysis. That's the most
4:32
accurate. There may be, you know, more accurate ones, but the
4:34
most accurate sort of analysis of the potential of autonomy, the most accurate that
4:42
I'm aware of is Kathy Wood's ARC invest. So and if I recall correctly,
4:46
they predicted somewhere over a five trillion dollar valuation for Tellla just based on
4:51
vehicle autonomy, not counting optimists. Yeah, yeah, I agree with that.
4:55
So so I think I think just based on vehicle autonomy, we can
4:59
we can tend act one ten x the value of the company. I believe
5:01
that is what will happen. Hey, it's just what do you know,
5:04
it's it's it's four twenty pm. Just notice that, so, you know,
5:11
So, so I think that's actually I think that's pretty accurate. But
5:15
there's this It actually gets way crazier when you think about the Optimus robot,
5:20
which is really a humanoid robot that is intended to you know, be able
5:25
to do anything you wanted to do, to be you know, it's your
5:28
you know, your companion. It can be at your house. It can
5:30
sort of babysit your kids. It could teach them, be a teacher,
5:34
you know, it can do factory stuff. Like I think that the ultimate
5:40
ratio of you, say, how many super useful humanoid helper droids do you
5:46
want? Like who doesn't want to see three po you know, you know,
5:48
but a C three P O plus R two D two plus you know
5:51
plus it would be pretty awesome. I think everyone in the world is going
5:55
to want one, like literally everyone, and then there will be obviously robots
5:59
in industry making stuff, and so I mean, I think the ratio of
6:03
humanoid robots to humans will probably at least two to one something like that,
6:09
one to one for sure, which means like somewhere on the order of ten
6:12
billion humanoid robots, maybe twenty or thirty. And so then it's like,
6:16
okay, well, let's say, you know, you kind of make they
6:19
said the build rate is I think the build rate will be probably something ultimately
6:24
like a billion a year humanoid robots like actually, and if Tesla just has
6:29
a temper cent sure of that, and it might be a lot more than
6:31
and this you know, who make like one hundred million Optimists units a year.
6:35
I just I mean, for reference, the auto industry is roughly one
6:38
hundred million vehicles per year, so that you know, sort of similar Bullpark
6:43
at least within an order of magnitude. And I think we could make one
6:46
for a cost of maybe at really high scale of about ten thousand dollars.
6:50
It's smaller, it'd be less expensive than a car. So and I think
6:55
if you sold it for sell for twenty thousand dollars something this is at large
6:58
scale volume, Tesla would basically make about a trillion dollars of profit a year
7:02
from that. So yeah, if the price oarning is multiple is say on
7:10
a twenty or twenty five something like, that would mean a twenty trillion dollar
7:14
market cap from Optimis alone, and probably five to ten from autonomous vehicles.
7:18
So like, I think it's actually it's within the realm of possibility for Tesla
7:26
to achieve evaluation ten times that of the most valuable company today. So when
7:30
I say this is like we're starting a new book, I mean it's going
7:32
to be the best book next level, next, next level, and we
7:36
need to make sure these robots are nice to us, and you know that's
7:40
very important. So so anyway that so with that, let me get into
7:43
the presentation. Blah blah blah blah blah blah. I think it might be
7:47
worth us just putting a word limit on future shareholder TELA of proposals. It's
7:53
like you can't have like a novel length, you know, it's kind of
7:56
like you can't have a whole novel as a TESLA present. So, you
8:01
know, as you can see, our impact is accelerating. You know,
8:03
we're starting to make a real noticeable event in corbon emissions. So we're really,
8:09
you know, making a lot of progress towards achieving our goal of sustainability.
8:13
And one of the effects of autonomy is actually is going to be an
8:18
even greater effect on carbon emissions because I suspect that we'll go from passenger cars
8:24
having about ten hours of usage per week, about an hour and a half
8:28
per day, to probably a thirty the hours in the week, maybe fifty
8:31
fifty five hours, which means the same car we'll be used five times.
8:37
So you need a certain amount of resources to make the car, but it'll
8:39
have five times the usage. In my opinion, I think that's quite likely,
8:45
which will be an even more dramatic impact on carbon emissions. So you
8:52
know, we care a lot about sustainable manufacturing. Our factories are beautiful.
8:56
Like you walk around the factory, it's actually, you know, a contrary
8:58
to these like what you're in the press. It's like a good vibe in
9:03
the factory, like people are smiling and happy. It's like it's nice.
9:07
So yeah, so I mean we do care. We actually do care a
9:09
lot. It tells about doing the right thing. And you know we're not
9:13
going to be perfect, but we do care a lot about doing the right
9:15
thing. So our vehicles are water efficient, energy efficient, We try whenever
9:20
possible to use renewables for powering the factory, and you know, do our
9:24
best to do the right thing. Our batteries are lasting. Look longer,
9:26
I get too many things, man, Anyway, the batteries the last lasting
9:30
a lot longer. It's just good. Safety is good in the factory.
9:33
We're improving the affordability of EBS you know model why it is comparable to a
9:37
BMW X three, but it's actual cost of ownership on a monthly basis is
9:41
much lower. We were number one in EBS globally last year. So yeah,
9:46
so yeah, so let's see, I Congress, that tails the team
9:48
on making six million vehicles all the time. So, I mean the tail
9:54
of the fleet is really becoming very substantial. I mean it's gonna be seven
9:58
million vehicles by the end of this year, over seven million vehicles. So
10:01
we have and our factory in Fremont is currently the highest volume auto factory in
10:07
North America. So and we broke the prior production record from when it was
10:11
new me So it's it's actually pretty wild that there's a we have a giant
10:13
car factory in the San Francisco Bay area. It's like it's not exactly the
10:18
cheapest place to have a car factory. It's like that in the Swiss Alps
10:20
or something, you know, but we still managed to make, you know,
10:22
great cars at high volume. And that's a testament to the great team
10:26
we've got in Fremont. So, Congress, the Fremont team, and you
10:28
know, a lot of people were saying the cyber truck is like fake,
10:31
it's never going to come out, you know. And now we're shipping a
10:35
lot of cyber trucks. We had a weekly record of thirteen hundred. And
10:39
you know, I think with a cyber truck, it really is something special.
10:43
Like you feel sometimes have like different opinions on the cyber truck, but
10:46
if you really want to know if there's a if something is cool, if
10:48
it's a great product, like show it to a kid, okay, like
10:54
no filter. Okay, the kid's got like no filter, like a five
10:56
year old, six year old something like that, and or three even three
11:00
year old and say which car do you like? Cyber truck? So it's
11:03
like that's how you know, And it's finally something that it just looks like
11:09
the future. It's and it drives so well, you know, it's comfortable,
11:13
it's it's just a fantastic product. I think it's out. I think
11:16
it's our best product, and it's a lot of innovation. It's got a
11:20
forty eight volt a low voltage system finally instead of twelve vaults, eight hundre
11:24
voult drive train, the world's biggest castings. You know, it's bulletproof.
11:28
All the cool things can you know outpull a F three P fifty diesel,
11:33
which is a very impressive truck, and it can you know, it can
11:37
do a quarter mile faster than a portion nine to eleven while towing a portion
11:41
nine eleven, So that's insane. So yeah, I think if you think
11:54
about it, like, how often do companies make products that move your heart
12:01
that are really special. It's so rare, and I think this is one
12:07
of them. So yeah, And we launched the upgraded three, which is
12:18
actually a fantastic car. I'd recommend trying it out. It's really a great
12:24
car, and it's only two hundred and sixteen dollars after a gas savings like
12:30
total cost ownership, it's basically two hundred and something dollars when compared to a
12:35
gasoline car, some of the gasoline cars, so it's really a great deal
12:41
and a fun car. And the performance of wealthy performance is well it's like
12:46
you know, fasting portion nine eleven. It's just a great car. So
12:54
yeah. So yeah, And of course one of Way became the best selling
13:00
globally, and this is something that we predicted, so you know, I
13:03
think I said in like twenty twenty two it would well away would be the
13:07
biggest car in the world by dollar volume sales, and that twenty twenty three
13:11
it would be the biggest in unit volume, and it was, and again
13:16
this year it will be the best selling car on Earth. So and we
13:22
were We've got the Tesla Semi. You know, we're in low volume production
13:26
of the Tesla Semi. And just last week I proved the plans for volume
13:30
production of the Tesla Semi. So we're gonna make a lot of semi trucks.
13:46
We're going to start off with. Yeah, it's something I think it'll
13:50
actually move the needle financially. It's not like it's not a small thing.
13:54
And the thing about like commercial vehicles is that companies that use them are super
13:58
objective. It's not like it's actually I mean, obviously we're going to make
14:03
it have cool style and be an awesome car. But for commercial vehicles,
14:07
the companies that make the buying decisions, they just look at it and say,
14:11
like, what are the numbers, Like, is this costs less to
14:15
transport or it costs more to transport than say a diesel truck. And the
14:18
thing is that the economics are much better than a diesel truck. It's kind
14:22
of basically no brainer. If you're a transport company, you don't use an
14:26
electric the test electric semi, you're just losing money. Why would you do
14:31
that? Do you not like money? Okay, but if you do like
14:37
money, then I recommend using the Tesla semi. And it's so I think
14:50
this is really going to actually sell it a scale that to you will be
14:54
surprised at and we'll actually move the needle financially and do a lot of good
14:58
And also for CO two and sustainability because semis are driven all the time and
15:03
actually even at much smaller numbers, they have a very big effect on total
15:09
carbon emission solved. So obviously we've got some new products that we're working on
15:18
under the covers, and I think these are I think is going to be
15:22
pretty special. So you know, some of them, I think people maybe
15:26
at first may think, oh, it's not going to be that amazing,
15:31
but just wait, it will be. Our supercharge network is continuing to grow.
15:35
You know, rumors of the death or supercharger are greatly exaggerated. We
15:39
are in fact continuing to grow the supercharge network significantly. In fact, this
15:45
year I think we will put more, deploy more superchargers this year that are
15:50
actually working than the rest of industry combined, just FI. So now we
16:00
are going to be more careful about, like you know, the capital efficiency
16:04
of where we deploy supercharges, but for sure any place that has congestion,
16:08
any sort of missing parts of the map that we're missing, we're going to
16:12
put the supercharges there. So even for the remainder of the year, we
16:15
expect to spend about half a billion dollars on as charge of deployment, so
16:21
it's very significant and well spent half a million, so this is it's a
16:25
lot so and we're this is actually not showing a bunch of new markets that
16:30
we'll be opening up later this year and next. So we're obviously we want
16:34
Tesla to be worldwide, and this is a lot of countries that we you
16:40
know, have zero to very few Tesla's and we obviously we want Tesla everywhere.
16:45
So there are a lot of new markets that we're going to open up
16:48
later this year and next. We're also opening the supercharge of supercharges to other
16:53
companies with the that's the adapter to the old somewhat clunky connector. But our
17:02
goal is to be helpful to other car companies, So any other car companies
17:06
that are going electric, we we do want to be helpful to them with
17:10
our supercharge network. So we're providing them with adapters and enabling them to have
17:15
access to the taels of supercharging network because we think it's better to do that
17:19
than to create a world garden that inhibits ev growth hopes. There's a bit
17:33
of latency, and we're also innovating a lot in battery production with the forty
17:37
six eighty that's built right here, and it is it's a hard problem,
17:42
you know, like there are entire companies that just make battery cells. That's
17:47
all they do. But we're making good progress, Like all the cyber trucks
17:51
that you see use the tells of forty six eighty cell and we have a
17:55
clear path to the forty six eighty being we think probably the most competitive cell
18:00
from it's from a manufacturing efficiency standpoint, and so it's but it is a
18:07
hard problem. I have to say, there's quite a lot of brain damage
18:10
required to be good at cell manufacturing. It's like a brain damage high.
18:14
But I think it does give Tesla resilience in the face of if there are
18:18
changes in the I don't know geopolitical situation somehow, but it's like it's good
18:23
to have something in the independence in cell manufacturing. So yeah, we're making
18:30
study progress. And we also have the cathode refinery, which you can see
18:33
behind the main factory, so you can sort of see it in the picture
18:37
there that's a cathode refinery. And then we've got the lithium refinery in South
18:41
Texas, and so we're just making sure we've kind of got the piece of
18:45
the puzzle and the I mean, if you would to look at sort of
18:51
a video of how Tesla does say cathode and lithium refining and how the rest
18:56
of industry does it, it's night and day. I mean, you can
18:59
sort of eat off the or in the Tesla refinery, and I would not
19:02
recommend doing so in the others. So you know, it's really we're definitely
19:06
going very vertical here. But I think that this is a it's a wise
19:10
investment that will pay off more than people realize. So I don't think this
19:30
thing is working. So this year we're also on track to complete a massive
19:36
number of energy deployments, so that is a gigantic increase in deployment capability.
19:41
So okay, we're really we seem to be tracking to sort of a two
19:47
to three hundred percent year of a year growth and energy storage deployment and stationary
19:52
pack so it's giant and the limiting factor really is being able to build more
19:59
megapacks and build more power walls. So we're ramping up production of the power
20:03
wall three, which is really a game changer for at the personal level.
20:07
The power wall three, it usually takes about three iterations of any given technology
20:12
to to really for it to be something where it's like, Okay, now
20:17
we're really hitting the sweet spot. So power Wall three is like an epic
20:19
product. The megapack is also it's even it's i'd call it sort of iteration
20:26
two of the Megapac is also an epic product. With with the Megapac three,
20:32
which is probably a couple of years away, will we'll start actually absorbing
20:37
more and more of the substation of the power of the sort of power electronics.
20:41
So because I sort of think you want to get to the point with
20:45
the megapac where you can literally just take the high voltage power lines and plug
20:48
them in just there's no substation, we just plug them in just drop them
20:52
down and drop it down and plug it in. That's and networks, which
20:57
is like mind blowing for the utility industry by the way, like they're like,
21:03
what, yeah, just plug the wires in and it'll work for very
21:08
high voltage. So that I think is actually going to be a major So
21:12
this is this great, excellent work by the energy storage team. Services and
21:18
other are also looking good. So we're actually now making money on the services
21:22
and other stuff. So this is kind of a big deal, right,
21:26
yeah, so let's see. Yeah, and Tesla's obviously way more than a
21:30
car company. You know, we do a lot of software Tesla you know
21:36
about I don't know, roughly half the or a very huge percentage of the
21:41
engineering we do is actually software engineering. So Tesla's i'd say as much as
21:45
software company as it is a hardware company. This is a very big deal
21:48
because car companies are not software companies normally, so this matters a lot.
21:53
So things like our order bit of software for energy storage, all of the
21:56
software that controls the cars, the megapacks, power walls of the solar obviously
22:03
for AI and and full self driving, a big deal, insurance service and
22:07
collision. Tesla also writes a lot of software internally that that helps that like
22:14
we call like the Tesla operating system internally that is head and shoulders above what
22:18
any other company has. I think, probably better than any Fortune five hundred
22:22
company. The Tesla internal software is is just way better. So it's yeah,
22:29
it's just far more than what people normally think of as a car company.
22:33
And Tesla's also the leader in a real world AI, so there's really
22:40
this is a big deal. Like Tesla is ahead of Google matter, you
22:45
know, opening anyone on real world software actually looking taking in video and making
22:49
decisions based on video. No one can. No one is even close.
22:55
And it's getting better, I say, with each passing month, if not
22:59
each passing week. Soya. And it's also worth noting that Tela is actually
23:04
pretty good at chip design. So the AI inference chip that's that was designed
23:10
by Tela, that's in cars. We had sort of our hardware three AI
23:14
inference cars made you over the past year have the hardware four. We've just
23:18
completed design on hardware five, which we're now calling AI five because it's really
23:25
is it's there's still actually is not a chip or from and I have a
23:30
lot of respect from a video any company that we would prefer to put in
23:33
our car that is better than what we have in the car. So so
23:45
we started from scratch in chip design just says we started from scratch in AI
23:52
software and have the best real world AI software and the and the best AI
23:56
infernence chip in the world from nothing. So this is you know, a
24:03
big deal, and the capability of the chips in the car is dramatic.
24:11
Like I mean, right now all the cars are actually training, like we
24:15
have Hardware four run Hardware three in emulation mode. So we'll continue to make
24:21
significant progress in hardware three, but later this year we'll actually bifurcate continue working
24:25
on hardware three, training on hardware three, but then do separate training on
24:33
hardware four. That'll be the sort of training cluster that we're building at the
24:36
south side of the gigafactory, and that'll be dedicated hardware for video and inference.
24:42
And you know, so that the hardware four has cameras that have about
24:48
four or five times better resolution, and depending on how you count the sort
24:53
of harderfor it's about anywhere from three to eight times better than hardware three.
25:00
But everything you're seeing thus far is just hardware three, and we still have
25:03
a long way to go before we get we reach the limits of hardware three.
25:07
So hardware three I think we'll still we'll do amazing things. But hardware
25:10
four, I think we'll probably do about five times better. Then Hardware five,
25:14
which comes out in about eighteen months or so, is ten times more
25:18
capable than hardware full. It's like a if you know, in video hardware,
25:23
it's sort of a B two hundred class computer, so you know,
25:27
that's like this. So what we'll just progressively do is improve how many nines
25:33
of reliability the car gets, and then of course that will go into optimists
25:38
as well. So the same chip we'll go into you know, hardware full,
25:42
we'll go into opto is an optimist. AI five, which you're switching
25:47
from hardware five to a F five will be an optimist and in all cars
25:51
in about eighteen months. And it's really just a staggering amount of compute,
25:55
and it's very power efficient compute. So it's got to be because if you're
25:59
in an oval application like a humanoid robot or a car, you can't just
26:03
be socking down ten kilowatts, you know, like you can in a data
26:07
center, So you've got to be very power efficient. So the you know,
26:11
sort of hardware three and four are only a few hundred watts. Yeah,
26:15
yeah, it's hard hardware five will be able to go probably up to
26:26
about seven or eight hundred watts, but it'll power fluctuate contingent upon the complexity
26:32
of the scene that it is in. So if it is in a parking
26:36
lot, station areas, like, you know, you know, you don't
26:38
have to think very much, just like a person like if you're in a
26:41
complicated traffic scenario, you've got to think a lot more than if you're just
26:45
cruising long, you know, on an empty road. But something that I
26:48
think is potentially interesting down the road is like, at some point, the
26:52
Tesla fleet I think will probably be you know, over one hundred million vehicles,
26:57
and if each vehicle has a killer watt of efficient, I think there's
27:04
a there's I think there's a sort of an Amazon Web Services AWS type opportunity
27:11
because if you've got one hundred million vehicles with a killowatt of efficients in furnace
27:15
compute, you've got one hundred gigawatts of compute, like one hundred gigawatts of
27:22
computers a lot, and it's distributed all over the world. So even so
27:26
when the car is not in autonomous mode, which I think probably is you
27:30
know, doing ROBOTAXI work maybe fifty sixty hours a week, but about one
27:37
hundred hours a week it's not it's probably stationary. So so there's one hundred
27:42
hours of one hundred gigawatts of infernce compute, which I think we should use.
27:48
Why not. You know, when people looked at Amazon, which started
27:52
out obviously as an online bookseller, and it's you know, has grown to
27:56
be and then they and then Amazon Web Services, like, well, they
28:00
got all these computers that only really see peak usage sometimes, but what are
28:07
they going to do when it's not peak usage? And sometimes Amazon servers are
28:11
down at like ten percent. So that's when they said, well, let's
28:12
do Amazon Web Services. And then Amazon Web Services became more valuable than the
28:18
entire rest of Amazon. And anyway, I think there's there's some kind of
28:22
opportunity there that's pretty significant for Tesla down the road. You know, again
28:26
that's really nobody's really factoring that in, but I think that that actually will
28:30
be quite significant. But we also are no longer compute constrained for training,
28:34
So I check in with the team, is like, is there anything we
28:38
could do to improve the pace of progress with respected training and inference? Currently
28:42
that is not the limiting factor. In fact, that the limiting factor right
28:45
now is that the amount of miles between interventions is so long that it takes
28:52
quite a while to figure out which version is better than the other version because
28:56
that none of them are requiring any interventions. So it's like, you know,
29:02
if he's not getting to like thousands of miles between interventions or like ten
29:04
thousand miles to get an intervention, then like, well, the average person
29:08
only drives about ten thousand miles in a year, and if it's in an
29:14
urban environment and the average speed is twenty miles an hour, I mean,
29:18
so our professional test drivers get pretty bored. Frankly, you know, they're
29:22
like, Okay, I drove all week and there was no intervention, Like
29:26
the highlight of the week would be like, yes, an intervention. Finally,
29:30
it's like getting to that point. So this is where actually having a
29:37
giant fleet is extremely important because we can deploy a new FST model and run
29:41
it in shadow mode and see how well it performs relative you know compare how
29:48
the human drives the car versus the new self driving build and then analyze that
29:53
delta in shadow mode like the shadow knows or doesn't know, does the case
29:57
maybe, and then be able to assess by getting billions of miles very quickly
30:03
with the giant fleet like that. Basically, that data engine is incredibly helpful.
30:08
Like I actually, it's not possible to solve the self driving problem without
30:14
having millions of vehicles on the road. So that's actually the is like figuring
30:19
out clever ways to test which how good the next build FSD build is actually
30:26
the limiting factor right now. And then, of course, like I said,
30:29
we are building another training data center right here, which will be dedicated
30:33
to hardware for training, So we'll bifurcate hardware three and hardware four training later
30:38
this year. We'll keep improving hardware three, but we're gonna uncork the full
30:42
capability of hardware four as well. So yeah, I think most people here
30:56
have tried out version twelve. I'd say, like sort of unsupervised full self
31:03
driving, full self driving version twelve, So we are actually just keeping the
31:07
version arbitrarily at twelve and then like calling a twelve point four, twelve point
31:11
three, twelve point four to twelve point five, but it's actually really like
31:15
version thirteen, version fourteen. But anyway, this is an arbitrary designation.
31:19
So twelve point four is actually like a whole different version than twelve point three,
31:26
and twelve point five is a whole different version twelve point four. So
31:30
you'll see really giant improvements I think sometimes factor of ten improvements between successive versions.
31:38
And then, as I mentioned, the way it'll work for the existing
31:41
fleet is be able to add us subtract your car to the fleets as you'd
31:45
like, and there will also be Tesla owned cars, so it'll be a
31:48
combination of like Uber plus Airbnb. But it'll be pretty wild that there'll be
31:55
a software update and then the whole fleet suddenly becomes accessible. It's like suddenly
31:59
you've got seven million, ten million cars that can ultimately tens of millions of
32:02
cars that can do autonomous driving, and instead of being used ten hours a
32:07
week, can be used fifty or sixty hours a week. So it's pretty
32:23
well to just be in Palo Alta with a bunch of cubes and then a
32:29
humanoid robot just walks past. We've made a massive amount of progress with Optimists
32:32
in a short period of time, from someone pretending to be a robot dancing
32:37
in a suit to a pretty hodgepodgy robot to a robot that is actually doing
32:42
useful tasks in the factory today. So we have two Optimus robots in our
32:45
Fremont factory that are doing basically this task, which is taking cells off the
32:52
end of the line and placing them in a shipping container. And yeah,
32:57
we actually have quite a few of these cruising around our office in pal Alto.
33:00
So there's and I think we've got one major hardware revision which should be
33:08
done by end of this year or early next before and then we'll move into
33:13
a limited production next year of Optimists, limited production for use in our factories
33:17
where we'll test out the product kind of you know, you know, as
33:22
I say, sort of either and dog food or whatever the electronic equivalent that
33:25
is. So, but I think like next year, my prediction is next
33:31
year we'll have over over one thousand, maybe a few thousand Optimus robots working
33:37
at Tesla, and and things are going to scale up very rapidly from there,
33:43
so well iron out the bugs. It'll like the degree of autonomy will
33:49
be radically better. You'll just literally be able to talk to it and say,
33:52
please do this task, or I'm going to show you something. Now
33:55
do that the thing that I am showing you, and you know, get
33:59
to the point where it can watch a video of something like a person and
34:02
then learn just by looking at that video and do that task. So yeah,
34:07
so really it's going to be quite something, and I'm confident of the
34:12
prediction that there will be more, like the ratio of humanoid robots to humans
34:17
will be greater than one to one, so that there'll be, you know,
34:21
more than ten billion humanoid robots in the world, probably twenty or more,
34:25
and Tella is going to be by far the leader in that. You're
34:30
seeing a lot of robot startups, but I think it's actually very challenging to
34:34
do Optimists as a robots startup because what we found to make Optimists work,
34:40
We've had to design every design from first principles from scratch. Every part of
34:46
the robot, so the motor, the gearbox, the power, electronics,
34:51
the communications system, everything had to be done from scratch. We found that
34:55
there's basically nothing, there's no supply chain. So even though there are many
35:00
electric motors made in the world, there's no supply chain for the types of
35:05
motors and sensors and gearboxes that are needed for humanoid robot. I mean,
35:10
what you're saying here is our current generation hand and arm, but actually which
35:15
has eleven degrees of freedom, our next generation has twenty two degrees of freedom.
35:20
It will be able to play the piano. So it's really like wow.
35:24
Now, of course, like I said, we need to make sure
35:27
we don't have the terminator scenario. That's very important. So safety of the
35:31
humanoid drobot it will be very important. But because it requires so much ground
35:38
up design, designing every motor, gearbox, sensor, par electronics from scratch,
35:43
it's very hard for a startup to if not impossible for a startup to
35:45
replicate that. But it tes that we have the world's best electrical engineering.
35:51
I think we've got the world's best mechanical engineering for gearboxes and for the electric
35:55
motors. Par electronics. You know, we have the resources to do that.
36:00
It applies quite well. And then you also have to have the brain.
36:05
You need the you need a power efficient inference computer, which we've got
36:09
for the car, and we'll be using an optimist, you need AI real
36:14
you need to be the best in real world AI, and Tela's the best
36:16
in real world AI, So you need all of these. You need a
36:20
very strong hand of cards in order to make a compelling robot. And then
36:24
you also need to be very good at scale manufacturing. So in order to
36:29
have the robot not cost like hundreds of thousands of dollars, in order to
36:32
make it cost like ten or twenty thousand dollars, you actually need to design
36:37
for manufacturing and be very good at manufacturing. And what in my experience,
36:43
prototypes are easy compared to volume manufacturing, prototypes are easy. Production is hard,
36:47
relatively speaking. So Tela has the production capability, it has the engineering
36:52
capability, and it has the AI hardware and software capability. And even the
36:59
most optimistic estimates that I've seen for optimists, the optimist I think under account
37:07
the magnitude of what this robot will be able to do. You know,
37:09
as I said at the beginning of the presentation, you know, I agree
37:14
with the ark and Best analysis that autonomous transport is called sort of a five
37:20
to seven trillion dollar market cap situation. Optimists, I think is a twenty
37:25
five literally twenty five trillion dollar market cap situation, the