Latest / Elon Musk Thinking / Elon Musk Said At Teslas Q1 2024 Financial Results.

Transcript

Elon Musk Said At Teslas Q1 2024 Financial Results. · Elon Musk Thinking

0:00
To recap. In Q one, we navigated several unforeseen challenges as well as
0:04
the ramp of the updated Model three in Fremont. There was As people have
0:07
seen, the eb adoption rate globally is under pressure, and a lot of
0:12
other water manufacturers are pulling back on ebs and pursuing plug and hybrids instead.
0:17
We believe this is not the right strategy and electric pickles will ultimately dominate the
0:21
market. Despite these challenges, the Tales of team did a great job executing
0:24
in a tough environment, and energy storage deployments a megapack in particular reaching all
0:29
time high in Q one, leading to record profitability for the energy business,
0:33
and that looks likely to continue to increase in the quarters and years ahead.
0:36
It will increase. We actually know that it will be significantly faster than the
0:39
car business, as we expected. We also continue to expand our AI training
0:44
capacity in Q one, more than doubling our training compute sequentially. In terms
0:49
of the new product roadmap, there's been a lot of talk about our upcoming
0:52
vehicle line in the next In the past several weeks, we've updated our future
0:57
vehicle lineup to accelerate the launch of new models. A head previously mentioned start
1:00
production in the second half. We expect it to be more like the early
1:03
twenty twenty five, if not late this year. These new vehicles, including
1:07
more affordable models, will use aspects of the next generation platform as well as
1:11
aspects of our current platforms, and we'll be able to be produced on the
1:15
same manufacturing lines as our current vehicle lineup, but it's not contingent on any
1:18
new factory or massive new production line. It'll be made on our current production
1:22
lines much more efficiently, and we think this should allow us to get to
1:26
over three million vehicles of capaccy when relates to the full extent regarding FSC version
1:30
twelve, which is the pure AI based self driving people that if you haven't
1:34
experienced this, I strongly vote you to try it out. It's profound and
1:38
the rate of improvement is rapid, and we've now turned that on for all
1:44
cars with the cameras and inferenced computer and everything from hardware three in North America.
1:49
So it's been pushed out to think around one point eight million vehicles and
1:53
we're seeing about half of people use it so far, and that percentage is
1:57
increasing with each passing week. So we now have over three hundred billion miles
2:00
that have been driven with FSDB twelve since the launch of full self driving,
2:06
so supervise full self driving. It's picked up very clear that the vision based
2:08
approach with end to end neural networks is the right solution or scalable autonomy.
2:13
It's really how humans drive. Our entire road network is designed for biological neural
2:17
nets and eyes, so naturally cameras and digital neural nets are the solution to
2:23
our current road system. To make it more accessible, we've produced the subscription
2:27
price to your ninety nine dollars a month, so it's easy to try out,
2:30
and as we've announced, we will be showcasing our purpose built robot taxi
2:34
or cybercap in August. Regarding AI compute, over the past few months,
2:38
we've been actively working on expanding Tela's core AI infrastructure. For a while there
2:43
we were training constrained in our progress. We are at this point no longer
2:46
training constrained, and so we're making rapid progress. We've installed and commissioned,
2:52
meaning they're actually working, thirty five thousand, h one hundred computers for GPUs.
2:55
GPU is wrong with they need a new word, feel like a WinCE
3:00
when I say GPU, because not defuse that Gcentral Graphics and Cindero Graphics.
3:04
But you know, roughly thirty five to age one hundreds AP proactive and we
3:07
expect that to be probably eighty five thousand or their abouts by the end of
3:10
this year in training. Just for training, we're making sure that we're being
3:14
as efficient as possible in our training. It's not just about the number of
3:17
AGE one hundreds, but how efficiently they're used. So in conclusion, we're
3:21
super excited about our autonomy roadmap. I think that it should be obvious to
3:23
anyone who's driving version twelve and a test that that is only a matter of
3:29
time before we acceed the reliability of humans, and not much time at that.
3:32
And we're really headed for an electric vehicle and autonomous future. And to
3:38
go back to something I said several years ago that in the future, gasoline
3:42
cars that are not autonomous will be like riding a horse and using a flipboot,
3:45
and that will become very obvious in hindsight. We continue to make the
3:47
necessary investments that will drive growth in profits and TAILS in the future. And
3:51
I wanted to thank the TAILS the team for incredible execution during this period and
3:54
look forward to everything that we have planned to add. So what is the
3:57
current status of Optimists. We are able to do simple factory tasks, or
4:00
at least I should say factory tasks in the lab. In terms of actually,
4:04
we do think we will have Optimists in limited production in the factory and
4:09
natural factory itself doing useful tasks before the end of this year, and I
4:12
think we may be able to sell it externally by the end of next year.
4:15
These just guesses. As I've said before, I think Optimists will be
4:19
more valuable than everything else combined. If you've got a antient humanoid robot that
4:27
is able to navigate reality and do tasks typer request, there is no meaningful
4:32
limit to the size of the economy. So that's what's going to happen.
4:35
And I think Tesla is the best position of any humanoid robot maker to be
4:41
able to reach volume production with efficient inference on the robot itself. This perhaps
4:47
is the point that is worth emphasizing. Tesla's inference efficiency is vastly better than
4:53
anyone any other company. There's no company even close to the infer efficiency of
4:57
Tesla. We've had to do that because we were constrained by the inference hardware
5:00
in the car would never choice, but that will pay dividends in many ways.
5:03
What is the assessment of the pathway towards regulatory approval for unsupervised suff as
5:09
the in the US, it's actually been pretty helpful that other autonomous car companies
5:13
been cutting a path through the regulatory jungle. But the which is that's actually
5:16
quite helpful, and they have obviously operating in Temples Go for a while.
5:20
I think they got approval for City of LA. So these approvals are happening
5:24
rapidly. I think if you've got at scale, it's a statistically significant amount
5:29
of data that shows conclusively that the autonomous car has let's say half the accident
5:34
rate of a human driven car. I think that's difficult to ignore because at
5:39
that point, stopping autonomy means killing people. So I actually do not think
5:44
that there will be significant regulatory barriers provided there is conclusive data that the autonomous
5:47
car is safer than a human driven car, and it might be this will
5:51
be much like elevators. Elevators used to be operated by a guy with the
5:55
relays, which but sometimes I would get tired or burnk or just make a
5:59
mistake and sure somebody in half between floors. So now we just have we
6:02
just get in an elevator from press button. We don't think about it.
6:05
In fact, weird if somebody is standing there with relay switch. That'll be
6:08
how cars work. You just summon a car using your phone. You get
6:12
in, it takes you to a destination, you get out. You don't
6:14
even think about it. Just like an elevator, it takes you to your
6:16
floor. That's it. Don't think about how the elevator is working or anything
6:19
like that. And something I should clarify is that Tesla will be operating the
6:24
fleet, so you can think of like how Tesla I think some combination of
6:28
Airbnb and Uber, meaning that there'll be some number of cars that Tesla owns
6:31
itself and operates in the fleet. There'll be some number of cars, and
6:35
there'll be a bunch of cars where they're owned by the end user. But
6:39
that end user can out or subtract their car to the fleet whenever they want,
6:43
and they can decide if they want to only let the car be used
6:46
by friends and family, or only by five star users, or by anyone
6:49
at any time. They could have the car come back to them and be
6:53
exclusively theirs. Like an Airbnb, you could run out your guest room or
6:57
not anytime you want. As our fleet grows, we have seven million cars
7:00
going to nine million car is going to eventually tens of millions of cars worldwide,
7:06
with a constant feedback loop. Every time something goes wrong, that gets
7:11
added to the training data and you get this training flywheel happening in the same
7:15
way that Google Search has the sort of flywheel. It's very difficult to compete
7:18
with Google because people are constantly doing searches and clicking, and Google's getting that
7:23
feedback loop. It's the same with Tesla, but at a scale that is
7:27
maybe difficult to comprehend, but ultimately be tens of millions. I think there's
7:30
also some potential here for an AWS element down the road where if we've got
7:35
a very powerful inference because we've got a hardware three in the cars, but
7:41
now all cars are being made with hardwor four, harbwor five is pretty much
7:45
designed and should be in cars hopefully towards the end the next year, and
7:48
there's the potential to have for to run when the car is not moving,
7:53
to actually run distributed inference like AWS, but distributed in ference like It takes
7:59
a lot of is to train an AI model, but many orders of magnitude
8:01
less compute to run it. So if you imagine the future paths where there's
8:05
a fleet of one hundred million Teslas and on average they've got like maybe a
8:11
kilowatt of inference compute, that's one hundred giggle whattts of infurance compute distributed all
8:15
around the world. It's pretty hard to put together one hundred gigawts of compute.
8:18
And even in an autonomous future where the car is paths used, instead
8:22
of being used ten hours a week, is used fifty hours a week,
8:24
it still leaves over one hundred hours a week where the car infurence computer could
8:28
be doing something else, and it seems like it would be a wayte much
8:31
to use it. We do have some insight into how good the think things
8:33
will be, and like, let's say three or four months, because we
8:37
have advanced models that are for more capable than what is in the car,
8:41
but have some issues with them that we need to fix. So they're like,
8:45
there'll be a step change improvement in the capabilities of the car, but
8:48
it'll have some quirks that that need to be addressed in order to release it
8:52
as a Showcus saying, we have to be very careful in what we release
8:56
the fleet or to customers in general. So if we look at say twelve
8:58
point four and twelve point five, which I really could arguably even be version
9:01
thirteen version fourteen, because it's pretty close to total retrain of the neural nets,
9:07
and in each case or substantially different. So we have been insight into
9:11
where the models, how to call, how well to call perform. In
9:15
say three or four months, can we get an official announcement of the timeline
9:18
for the twenty five thousand dollars vehicle. Really the way to think of Tesla
9:22
is almost entirely in terms of solving autonomy and being able to turn on net
9:26
autonomy for a gigantic fleet. And I think it might be the biggest asset
9:30
value appreciation in history when that happens, when you can do unsupervised full self
9:33
driving. Have any of the legacy automakers contact the test lava possibly licensing f
9:39
ASD in the future. We're in conversations with one major ortamaker regarding licensing epesty.
9:45
Can we make f as the transfer permanent until etizes fully delivered with level
9:48
potonomy. So I was just wondering if you can elaborate a little bit more
9:54
on kind of the new vehicles. I think we've settled we will on that
9:58
front. So what to follow? Maybe you can just talk about where your
10:01
heart is at. As a constitute's majority of my worktime and I work pretty
10:05
much every day of the week, It's rare for me to take a Sunday
10:07
afternoon to make sure Tesla is prosperous, and it is like it is prosperous
10:13
and it will be very much so in the future. What's your team's degree
10:15
of confidence on growth above zero percent? No, I think we will have
10:18
higher sales this year than last year. How long would it take your best
10:22
Chinese competitors to copy a cheaper and better vehicle that you could offer a couple
10:26
of years from now. I don't know what our competitors could do, except
10:31
we've done relatively better than they have. You know, if you look at
10:33
the drop and our competitors in China sales versus our drop and sales our drop
10:37
is less than theirs, So we're doing well. But I think Kathy would
10:41
said it best, Like, really we should be thought of as an AI
10:43
robotics company. If you value Tesla as just like an order company, you
10:48
would just have to fundamentally, it's just the wrong framework will count. If
10:52
you ask the wrong question, then the right answer is impossible. So I
10:56
mean, if somebody doesn't believe Tesla's going to solve autonomy, I think they
10:58
should not be an investor in the company, Like that is what we will
11:03
and we are. And then you have a car that goes from ten hours
11:07
abuse a week, like an hour and a half a day to probably fifty
11:09
but it costs the same. If you've not tried the embassy twelve point three
11:13
and like I said, twel point four is going to be significantly better in
11:16
twelve point five, even better than that, and we have visibility into those
11:20
things, then you really don't understand what's going on. It's not possible.
11:24
We're putting the actual auto in automobile. Tell us about future horse carriages you're
11:28
making, I'm like, actually, it does need a horse. That's the
11:30
whole point. That's really the whole point. You've spoken about your desire to
11:35
obtain twenty five percent vote and control of the company. I think no matter
11:37
what. Tell the even if I got kidnapped by aliens tomorrow, Tesla will
11:41
solve autonomy. Maybe a little slower, but it would solve autonomy for vehicles
11:46
at least. I don't know if it would win on with respect to optimists
11:48
or with respect to future products, but it would. There's not momentum but
11:52
tells the tosolve autonomy even if I disappeared for vehicles. Now, there's a
11:56
whole range of things we can do in the future. Beyond that, I'd
11:58
be more reticent with respect to optimists. If we have a super sentient humanoid
12:03
robot that can body reindorse and that you can't escape, we're totally terminator level
12:07
risk, then yeah, I'd be uncomfortable with if there's not some meaningful level
12:13
of influence over how that is deployed, and if there's sureholders have an opportunity
12:18
to ratify or reratify the sort of competition. I guess I can't say that,
12:22
but that is a fact, they have an opportunity. Okay, very
12:26
good, and yeah, we'll see. If the company generates a lot of
12:28
positive cash flow, we could obsoustely buy back shares. What are the types
12:31
of activities that you're presumably sacrificing as a result of parting ways with these folks.
12:37
We're not giving up anything that it's significant that I'm aware of. We've
12:41
just had a long period of prosperity from twenty nineteen to now. And so
12:43
if a company organizationally is five percent wrong per year, that accumulates to twenty
12:50
five to thirty percent of inefficiency. We've made some corrections along the way,
12:54
but it is time to reorganize the company for the next phase of growth.
12:56
And you really need to reorganize it, just like a human when you restart
13:01
off us one cell and kind a zygote and lass assist, and when you
13:03
start throwing arms and legs and briefly you have a tail, and what you
13:07
showed the tail, you show the tail hopefully and then your baby and you
13:11
basically you have to be the organism. A company is a creature growing,
13:13
and if you don't reorganize it for different phases of growth, it will fail.
13:18
You can't have the same organizational structure if you're ten cells versus one hundred,
13:24
versus a million, versus a billion versus a trillion, where humans are
13:26
like around thirty five trillion cells. It doesn't feel like it feels like one
13:31
person, but you're basically a working cell colony of roughly thirty five trillion depending
13:35
on your body mass, and about three times that number in backteria. Anyway,
13:37
you've got to reorganize the company for a new phase of growth or it
13:41
will fail to achieve that growth. Can you elaborate on how much the licensing
13:43
business opportunity you mentioned today has progressed. I think we just need to It
13:48
just needs to be obvious that our approach is the right approach, and I
13:50
think it is. I think we're now with twelve point three. If you
13:54
just have the car drive you around, it is obvious that our solution,
13:58
with a relatively local cost inference computer and standard cameras can achieve self driving no
14:03
light oars, no radars, no ultrasonics, nothing, just no having integration
14:07
work for vehicle manufacturers. Yeah, it's so really just be a case of
14:13
having them use the same cameras and inference computer and licensing out software. And
14:18
but it's once it becomes obvious that if you don't have this car, nobody
14:22
wants your car. It's a smart car. And I just remember, like
14:24
the back when there was King of the Hill crushing I saw come out with
14:28
a smartphone that was basically a break with limited functionality, and then the iPhone
14:33
and Android, but people still did not understand that all the phones are going
14:37
to be that way. There's not going to be any footbook, if there'll
14:39
be a niche product or not even exactly there's last time be the homebook.
14:43
And people that understand all cars will need to be smart cars there or you
14:46
will not sell the car, will not nobody will buy it. Once that
14:50
becomes obvious, I think licensing becomes not optional licensing or nobody will buy your
14:54
car. A deal sign now would result in being in a car probably three
14:58
years. That's like lightning. Basically, I wouldn't be surprised if we do
15:03
sign a I think a good chance we do sign a deal this year,
15:05
maybe more than one, but yeah, it would be probably three years.
15:07
Supports integrated with a car, even though all you need is cameras and our
15:11
Burns computer, so it's like not a massive design change. Do you still
15:15
see meaningful incremental price reductions as making sense from here for the existing products?
15:18
Yeah, I think we can be pre casually positive meaningfully in the day.
15:22
Like for any given company, if you sell a great product at a great
15:26
price, the sale if if you have a great product at a great price,
15:28
the sales will be excellent. That's true of any area. Over time,
15:33
we do need to keep making sure that it's a great product at a
15:37
great price, and moreover, that price is accessible to people, so it's
15:39
not you have to solve both the value for money and the fundamental affordability question.
15:43
The fundamental afordability question is sometimes overlooked. Somebody's earning several hundred thousand dollars
15:48
a year, they don't think of a car from a fundamental affordability standpoint,
15:52
but from the best majority of people are living paycheck to paychecks, so it
15:56
actually makes difference if the cost per month will least financing is ten dollars one
16:02
way or the other. It's important to keep improving the affordability, to keep
16:07
making the price example, yeah, exactly, the price more accessible, the
16:11
value for money better, and to keep improving that over time. But also
16:14
the kick if people want to buy, Yeah, it's got to be a
16:17
great product at a great price, and the statards for what constant's great product
16:19
at a great price keep increasing, So that you can't just be static.
16:22
You have to keep saying, keep making the car better, improving the price,
16:26
improving the cost of production. And that's what we're doing. Can you
16:29
please help us understand some of the timing of launching FSD in additional geographies,
16:34
including maybe clarifying your recent comment about China new markets. Yeah, we are
16:40
a bunch of marks where we don't currently sell cars that we should be selling
16:44
cars, and we'll see some acceleration of that. So the thing about the
16:47
all the anti end neural net based autonomy is that just like a human,
16:52
it actually works pretty well without modification in almost any market. So we plan
16:56
on with the approval of the regulators, releasing it as a supervised autonomy system
17:03
in any market that where we can get regulatory approval for that, which we
17:07
think includes China. So yeah, it's just like a human. You can
17:10
go rent a car in a foreign country and you can drive pretty well obviously
17:12
if you're if you're living that country, you'll drive better. And so we
17:15
will make a car drive better in these other countries with countries specific training.
17:19
But it can drive quite well almost everywhere and understands that it shouldn't hit things
17:23
no matter where exact whether or not you feel that supply constraints that you mentioned
17:29
throughout the release impacted the results, and maybe can you help us quantify that?
17:32
And is that why you have some confidence in growth in twenty twenty four.
17:36
If you've got cars that are sitting on ships, they obviously cannot be
17:38
delivered to people. And if you've got the access demand for want for model
17:44
through or model line in one market but you don't have it there, it's
17:47
like it's quite as it's extremely complex logistics situation. So you know, I'd
17:52
say also the we did over complicate the sales process, which we've just in
17:57
the past a week or so have greatly simplified. So the it just became
18:03
far too complex to buy a Tesla, whereas it should just be you can
18:06
buy the card and under a minute. So we're getting back to that you
18:10
can buy a Teslaine under a minute interface from what was quite complex and your
18:14
comments around distributed inference, can you talk about what that approach is unlocking beyond
18:18
what's happening in the vehicle right now. That's why I think it's analogous to
18:22
Amazon Web Services, where people didn't expect that AWS would be the most valuable
18:26
pot of Amazon when it started out as a bookstore, so that was on
18:30
nobody's radar. But they found that they had excess compute because the compute needs
18:34
would spike to extreme levels for brief periods of the year, and then they
18:38
had idle compute for the rest of the year. So then what should they
18:41
do with all that access compute for the rest of the year. That's moetizing,
18:45
yeah, monetized it. So it seems like no brainer to say,
18:47
Okay, if we've got millions and then tens of millions of vehicles out there
18:51
where the computers are idle most of the time, that we might as well
18:55
have them do something useful exactly. And if you get to one hundred million
18:59
vehicles level, which I think we will at some point get to, then
19:02
and you've got to kill a what of usable compute, and maybe you're on
19:04
hardware six or seven by that time, then you really I think you could
19:07
have on the order of one hundred gigawattes of usable compute, which might be
19:11
more than anyone, more than any company, probably more than any company like
19:15
technically like Apple would have the most amount of distributed compute, but you can't
19:18
use it because you can't get the you can't just run the phone and full
19:22
power and during the battery. Yep, So the for the car, even
19:26
if you're a killer what level furs computer, which is crazy power to get
19:30
to a phone if you've got to after you're a sixty erlo hour pack,
19:33
it's still not a big deal to run. You're plugging, whether whether you're
19:36
plugged in or not plugged in or not plugged in, you could run for
19:38
ten hours and use ten killer what hours if you killer what of compute?
19:44
Yeah, we're together built in liquid cold thermal management. Yeah, it's exactly
19:47
if that's for data centers, so sort of there in the car exactly.
19:49
So that's distributed power generations, just distributed access to power and distributed cooling and
19:56
that's already paidful. Looking at the forty six eighty ramp, can you talk
19:57
about how close you are to target yield and when you might start to accelerate.
20:02
So we were making good progress on that, but I don't think it's
20:04
super important for at least in the near term. As as Lara said,
20:08
we think it will be exceed the competitiveness of suppliers by the end of this
20:12
year and then we will continue to improve. The big part of the forty
20:15
six eighty IT test doing in total sales was a hedge against what would happen
20:19
with our suppliers, because for a while there it was very difficult because every
20:25
big carmaker put in massive battery orders and so the price is the price per
20:29
killed hour of a battery of latinion batteries went to crazy numbers, crazy levels,
20:33
just bonkers. Okay, we've got to have some hedge here to deal
20:37
with the cost per cold ours numbers that were double what we anticipated. If
20:41
we have an internal self production, then we have that hedge against demand trucks
20:45
with too much demand. That's really the way to think about it. It's
20:48
not like we want to take on a whole bunch of problems just for the
20:51
hell of it. We did the seal program in order to address the crazy
20:56
increase in cost per kill what hour from our suppliers due to jig antique orders
21:00
placed by every car maker on oath. And once again would just like to
21:03
strongly recommend that anyone who is thinking about the Tesla stock should really drive f
21:08
s T twelve point three. Really you can't. It is impossible to understand
21:14
the company if you do not do this. You mentioned that autocogs per unit
21:17
for the next gen vehicle would decline by fifty percent versus the current three and
21:21
y on that topic of forty six eighty cells, I know you mentioned it,
21:25
you really thought of it more as like a hedge against against the rising
21:27
battery costs from other OEMs. What seems to be happening is that in this
21:30
I'm missing something. The orders for batteries from other order makers have declined dramatically,
21:36
So we're seeing much more competitive prices for cells from our suppliers, dramatically
21:41
more competitive than in the past. It is clear that a lot of our
21:45
suppliers have access capacity. Yeah, now this is going to wax and way.
21:48
Obviously there's going to be a boom and bust in battery cell production where
21:52
production exceeds supply and then supply exceeds production and back and forth on a dram
21:56
or something. But it's like, what is true today not be true in
22:00
the future. There's going to be somewhat of a boom and bust cycle here,
22:03
and then there are additional complications by with governan centatives like the Inflation Production
22:07
Act the Irish always found like a funny name for a Yeah, this is
22:11
like the Irish Republican Army, the Internet Research Agency from Russia, Independent retirement
22:15
too, Yeah exactly roth iras a for Spider Man situation which I which IRA
22:19
wins is complicate the incentive structure so that there's there is a stronger demand for
22:26
sales that are produced in the US than outside the US. But then how
22:29
long does that The IRA