Tesla Inc ($TSLA) Q2 2023 Earnings Call

Thank you, Martin. So, just a Q2 recap. In Q2, we achieved record vehicle production and deliveries and record revenue of about $25 billion in a single quarter. And Model Y became the best-selling vehicle of any kind globally in Q1, surpassing the likes of Corolla and Golf. So, it's the number one vehicle of any kind, including vehicles that are sold at a lower price. This is, I think, an incredible achievement for us in terms of team and just a huge thank you to our customers for their support. So, and this came in spite of high interest rates and a lot of macro uncertainty. And nonetheless, we managed to achieve operating margin of about 10%. We're continuing to target 1.8 million vehicle deliveries this year. Although we expect that Q3 production will be a little bit down because we've got summer shutdowns for a lot of factory upgrades. So, just a slight decrease in Q3 before sort of global factory upgrade. Now, in the long term, autonomy we think is going to just drive volume through the ceiling next level. And the future of our taxi products, the dedicated robot taxi products, we think have got quite the infinite demand. So, and the way we're going to manufacture the robot taxi is also itself revolutionary. So, it's a revolutionary design made in a revolutionary way. It'll be like my father's highest give us growth power of any vehicle production ever. So I'm excited about that. With respect to the robot, in order to build autonomy, we also need to train our neural net with the data from millions of vehicles. We have the more, I mean, this has been proven over and over again. The more training data you have, the better the results. And there are times where we see basically in a neural net, basically it's sort of at a million training examples, it barely works. At 2 million, it slightly works. At 3 million, it's like, wow, okay, we're seeing something. But they get like 10 million training examples, it's like it becomes incredible. So, you just, there's just no substitute for massive amount of data. And also Tesla has more vehicles on the road that are collecting this data than all other companies combined by I think maybe an order of magnitude. So, I think we might have 90% of all compute power is a very big number. So, yeah, the success in AI and efforts is a function of talent, unique data, and computer resources. And we have outstanding capabilities in all three arenas. And I really just don't know how anyone could do what we're doing even if they had the best software and had a computer if they did not have the training data. So, speaking of which, our Dojo training computer is designed to significantly reduce the cost of neural net training. It is designed to, it's so optimized for the kind of training that we need, which is video training. So, you know, we just see that the need for neural net training and getting the Dojo quality from the things is just enormous. So, I think having, we expect to use both NVIDIA and Dojo, to be clear. But there's, we just see the demand for really fast training resources. And we think we may reach in-house neural net training capability of 100 exaflops by the end of next year. So, to today, over 300 million miles have been driven using FSD beta. That 300 million mile number is going to seem rather small very quickly. It'll soon be billions of miles and tens of billions of miles. And the FSD will go from being as good as a human to then being vastly better than a human. We see a clear path to full self-driving being 10 times safer than the average human driver. So, and between Autopilot, those are computer inference hardware in a car which is composed, so some hardware 3, 4, you know, and it's really dedicated, it's a high efficiency inference computer that's in the car. And our Optimus robot, Tesla is really at the cutting edge of AI development. Okay. To which we are just inside, which we continue to build these cameras for the Cybertruck on our final production line in Austin. I'm actually here in Austin at the Gigafactory. This is the first truck that we're aware of that will have a full with over 6 ft bed and will fit into a 20 ft garage. So, it's the biggest on the outside but it's even bigger on the inside. So, it's a, I got to, one of the elements of good design is it should feel big on the inside than it looks on the outside. And this is no small car but we really cared about the exterior dimensions of the Cybertruck down to the last millimeter. So, it's just equally trying to random the car and all that. Not too big, not too small. And then really maximize the utility of our. And we can't wait to start delivering it later this year. Some other highlights are global supercharging network now stands over 50,000, roughly 42,000 connectors and over 5,000 locations. As I think a lot of people are aware, the Tesla charging standard, which we made open source and it's now called the North American charging standard. We're deeply honored that Ford, GM, Mercedes, and many other brands have signed up to use our connector and gain access to our charging network. Yeah, we strongly believe in helping other car companies to accelerate the EV revolution and just trying to do the right thing in general. So, that's all I'll go there. Then, I mean, I think I want to emphasize like very strongly, this is a very important point, is that Tesla, just as with the North American charging standard, although we're not licensing, I guess I'm not licensing, we're just making it available. But we are very open to licensing our full self-driving software to hardware to other car companies. And we're already in discussions with, I mean, early discussions with a major OEM about using Tesla FSD. So, we're not trying to keep this to ourselves. We're more than happy to license it to others. And that's where our new lithium refinery and cathode facility are progressing well. Now, in conclusion, we continue to focus on making as many cars as we can while maintaining healthy financials. Our artificial intelligence development is obviously entering a new era. And we're incredibly excited about what's to come. Our other businesses, such as Megapack, supercharging, service, and whatnot, also are starting to become a meaningful contributor to overall profitability this quarter. And then lastly, I'd just like to profusely thank all of our important employees for making it a lot of extra effort during uncertain times. Thank you very much for your hard work and impact you're making.

Yes, the question we get asked a lot. So, we're excited to announce that for Q3, we will be allowing transfer of FSD. This is a one-time amnesty. So, you need to take advantage of it in Q3, but or at least place the order in Q3 within certain recovery frames. So, yeah. Yeah, I hope that actually will happen. Well, we're not going to do this, this is a one-time thing. You understand.

The demand is so far off the hook you can't even see the hook. So that's really not an issue. I do want to emphasize that the Cybertruck has a lot of new technology in it. Like a lot. It doesn't look like it. It doesn't look like any other vehicle because it is not like any other vehicle. So, the production ramp will move as fast as the slowest and least like elements of the supply chain and internal production. So, you know, I wouldn't expect it to go, I hope it's smooth. You know, we're certainly better at production ramps than we've got a lot of experience with production ramps, but you know, first order to first order approximation is like 10,000 unique parts and processes in the Cybertruck and if any one of it will go as fast as the least lucky, you know, least well-executed element of the 10,000. So, always afraid of predicting the ramp initially, but I think we'll be making them in high volume next year. And we will be delivering the car this year.

So, that must be why everyone's copying us.

Yeah, closing loop on collision repair, factoring that into design is a big deal. Crucial. I don't think anyone else can do it with the ecosystem that we have. Yeah. And we are actually able to change the details of the casting with inserts. And we actually do that all the time. So, the advances are actually world class and easy to replace anyway. So, we can actually design changes the inserts and tweak the castings, but the casting is basically cast the rear body or front body is lighter, cheaper, better for noise vibration, harshness, much easier to manufacture. And it's better in every way. And that's why so many other car companies are copying us. It's probably, they're not going to, well, they certainly put out a lot of pretty good styling. I think it's basically going to be how all cars are made in the future.

10 million. Yes, I think we're around five or six bots. Like, you know, there's, we're at 10, I guess. Which depends on which I said how many are working in what phase, but it's a, yeah, there's more than one. The, we, yes, I can, I thought I should say larger, should say about the Optimus bot. We found that there are actually no suppliers that can produce the actuators. There's no off-the-shelf actuators that work well for a humanoid robot. At any price. Certainly not a compelling humanoid robot. Yes, but not a, so, there's not a humanoid robot that can do stuff that, you know, the things that a human can do. So, we've actually had to design our own actuators that integrate the motor, the electronics, the controller, the sensors. And it really, yeah, every one of them is custom designed. So, and then of course, we'll be using the same inference hardware as the car. So, yeah, and we're all in designing these actuators, I'm designing them for fine questions. So, they're not just lighter, tighter, more and more capable than any other actuators that we're aware of that exist for a while. It's also actually manufacturable. So, it will take them a while. The first Optimus that will have all of the Tesla designed actuators, so the production candidate actuators integrated and working should be around November-ish. So, and then we'll start ramping up after that. You know, in terms of when we'll be able to do some useful things, we'll first be trying this out in our own factories and just proving out its utility, but I think we'll be able to have it do something useful in our factories sometime next year. I would be pretty confident about that. So, yeah, and it's going well. I should go another cool thing about Optimus is that, you know, there's just in the US alone there are 2 million amputees. And I was just talking to the Neuralink team and by combining a Neuralink implant and a robotic arm or leg for someone that has had their arms or legs or all arms or legs amputated, we believe we can give you a basically a cyborg body that is incredibly capable. $6 million man in real life. And it will, don't worry, cost $6 million. $60,000 man. Still doesn't impress me, but it will. So, that actually could be a really, I think it would be incredible to, you know, potentially help millions of people around the world. And I think I get from, you know, a whole lot like that is as good maybe, maybe long-term better than a biological.

Yeah, I guess demand is roughly tracked production. So, which is what we aim for. It's something that we have that really I think no other car maker has is that we have real-time demand and real-time production. Like so, 7 days a week, you know, I get an email when orders are in. Email shows output from all factories and orders globally. So, it's like a real-time thing on the

Pulse of Earth, basically. And we adjust costs according to what the mood of the public is. Buying a new car is a big decision for most people, so anytime there's economic uncertainty, people try to pause on new car buying at least to see what happens. And then also another challenge is the interest rate environment. As interest rates rise, the affordability of anything bought with that decreases, effectively increasing the price of the car. So when interest rates rise dramatically, we actually have to reduce the price of the car because the interest payments increase the price of the car. And this is, at least up until recently, I believe the sharpest interest rate rise in history. So we have to do something about that. And it's not good for the global economy.

Yeah, exactly. Yeah, I'm sure of that. It should be another Twitter. So, I mean, one day it seems like the world economy is falling apart and the next day everything's fine. I don't know what the hell's going on, to be totally frank. I wish I did. So, I mean, that's why I feel like, yeah, I always say, you know, on Twitter I posted like, you know, just really advising, I care a lot about the sort of small shareholders, especially the ones that have stuck with us through thick and thin. I love you guys. And so, we can't control these macro shocks, you know, or the manic-depressive nature of the stock market. So that's why I recommend against margin loans in times that are turbulent. You know, in fact, if times are not that turbulent, actually margin loans can be a smart move within reason. But we're in, I would call it, turbulent times. I have very high confidence in the long-term value of Tesla. Like, I see a path to a 10X. Or maybe I'm being completely disingenuous, but I'll call it a 5X. You know, in the value of the company. It may be a 10X. But where things go along the way, the trials and tribulations and the mood of the market, you won't count predict. And so, you know, I mean, the old adage of buy and hold is right, you know. For investment advice, I feel like identify companies whose products you love. See if they, you know, do they seem like they'll continue to make good products or great products. Buy that stock and hold it. That's it. You will win. And the reason companies exist is to make goods and services, ideally great goods and services. They don't exist for any other reason. They shouldn't. So, that's why I like to talk about a company that makes good products and has a great future pipeline. It's common sense, actually. And then, in general, if you buy that company for what that company's products or services are, when the market panics, buy. And when the market is, you know, overly exuberant, you know, you can sell. I'm not recommending you sell those as well, but yeah, that's it. Buy low, sell high. You know, Warren Buffett actually has a saying, I'm paraphrasing him, but you know, a publicly traded company is like, it's like you're living in your house and some crazy manic depressive guy comes and stands at the outside of your house and yells property prices at you. You know, and it's a different price every day. But the house is still the same house. So it's like the stock market. You know, you credit that to Warren Buffett.

What was that? What was that question again? Say it again.

Great. Great. First of all, I think it's ridiculous that we have positive free cash flow in a capital intensive business while investing massive amounts of money in new technology. That is super hard.

Yeah, we actually make up... Yeah.

So, are you saying how much we can spend on Dojo or...?

Well, we're not going to be open loop on our Dojo expenditures, so... but I mean, I think we will be spending, you know, something north of a billion for the next year on, you know, the fourth to third year next year. Well over a billion dollars in Dojo. And yeah. So, I mean, we've got a truly staggering amount of video data to do training on. And this is another thing. I don't like, in order to copy us, you also need to spend billions of dollars on training computers. I mean, it's like... And it's also hard to... You know, you need the data and you need the training computers. Like, I think all these things needed to actually achieve this at scale for a generalized solution for autonomy is... this is one of the hardest problems ever. You know, you see a lot of AI companies doing, you know, LLMs and whatnot. And I'll say, if they're so great, why can't they make us all great driving cars? Because it's harder. That's why. So, but I do think there's some great AI companies out there, but this fundamentally, the staggering amount of data we've got to process or start to process somehow. And custom silicon is the best way to do that. So, that's what Dojo is designed to do, is optimized for video training. It's not optimized for LLMs, it's optimized for video training. And with video training, you have a much higher ratio of compute to memory bandwidth. So, and you know, with LLMs, like, it's typically memory bandwidth choked. So, that's everything. But like, we're also, we have some... We're using a lot of the video hardware and continue to... Yeah, we're actually taking video hardware as fast as they're able to deliver it to us. Tremendous, tremendous respect for Jensen and Nvidia. They've done an incredible job. And frankly, I don't know if I think if they could deliver us enough GPUs, we might not be able to do what they can. So, you know, they got so many customers. They've been kind enough to, you know, put us on a priority list for our GPU orders. But yeah, this magnitude of failures changes things because like I said, we're not trying to just get as good as human, we want to get to, you know, 100 times better than human, maybe 100 times better than human. Right now, I believe there's something on the order of a million automotive deaths per year. And then if you say permanent serious injuries, I think it's probably closer to 10 million per year. And yes, it matters if you're, you know, twice as good as human, 10 times, you know, like 10 times better than human would still mean 100,000 deaths and a moment so severe permanent injuries. So, it's like, okay, well, we'd probably be 100 times better. So, there's really, you know, it's a matter of minds and everyone want to achieve as perfect safety as possible. And it's a vast, truly mind-boggling amount of video and computer needed for that. So, and then I just, you know, I just think there's other applications for data, but we just have what we needed for video training.

Yeah. I would say the fundamental rate limiter on the progress of full self-driving is training. But that's if we had more training compute, we would get it done faster. So, that's a... And it's difficult to project out quickly. Yeah, we can actually do that.

It sounds like getting more share, it's just that you can think of every car that we sell or produce that has full autonomy capability as actually something that's in the future maybe worth as much as five times what it is today. Because, you know, if average passenger vehicle is doing like maybe 10 hours of driving a week, you know, if it's one and a half hours a day on average, that's 10 hours a week. If you've got autonomous, if the vehicles have to be driven autonomously and used in some either dedicated autonomous or partially autonomous like Airbnb, like something maybe sometimes you allow your car to be used by others. Sometimes you want to use it exclusively just like, you know, Airbnb, you know, doing Airbnb with a room in your house. You know, the value is just tremendous. So, I think it's sort of, it would be, I think it does make sense to sacrifice margins in favor of making more vehicles because we think in the not so distant future they will have a dramatic valuation increase. I think the Tesla fleet value increase to the point where we're going to upload a full self-driving and it's approved by regulators, it will be the single biggest step change in asset value maybe in history.

Well, obviously, you know, my prediction that full self-driving has been optimistic in the past. The reason I've been optimistic is what it just look like there's the, we'll make rapid progress with the new version of FSD. But then the curve curves over logarithmically. So first, like a logarithmic curve looks like, you know, just a fairly straight upward line going up. And so if you extrapolate that, then you have a great thing but then the actual logarithmic curve goes over and then there's a series of stacked logarithmic curves. Now, I know I'm the boy who cried FSD. But I think we'll be better than human by the end of this year. That must be where we are with our regulators. And I'm saying that that would be in the US because we got to focus on one market first. But I think we'll be better than human by the end of this year. I've been wrong in the past. I might be wrong this time. And for the price of FSD, so the way to think of the price of FSD is actually very low. It's not high. We go back to what I said earlier. If the value of the car increases dramatically if it is actually autonomous, you know, if $15,000 is actually a low price. It's not a high price. And now we will offer and, you know, and we do sort of offer FSD as a sort of monthly subscription. Although I think most people are not that... so I recommend like maybe trying it out as a monthly subscription so you don't have to go with the $15,000 thing but I think... Yeah, obviously if a car is worth so much money it should be really priced... $15,000 is actually a low price for FSD.

Yeah, I think it'll actually enhance the value of Tesla. There were just some of the world's best AI engineers and scientists that were willing to join a startup but they were not willing to join a large sort of relatively established company like Tesla so that's actually how it got started. I was interviewing a few people and they're like, 'No, we want startup.' I was like, 'I know, it's all I can't convince them to join.' So that's why I said, 'So I was like, okay, well, you know, better at the startup that I run this than they go somewhere else.' That's kind of the genesis of xAI. And xAI is focused on AGI. Yeah. So it's... But I think there will be some value that xAI brings to Tesla. You know, I also have some of the best, for the very best people in the world that they just want to work on interesting problems. If you take, say, you know, my pure science group, you know, really what convinced the chief colonel to leave Apple where he was very happy and well compensated. And both at Google... Yes. We need to... Yeah, the what we think is the best material science group in the world. With that he got to work at both Tesla and SpaceX. He wasn't going to leave Apple if it was just Tesla, but he was going to go if it was Tesla and SpaceX. So sometimes you get the best talent in the world. That's the kind of thing, you know, you need to do. And that actually has been very beneficial to Tesla. So yeah.

Sure. You know, we have a really... So, we started a referral program, which I think will be quite effective. But, you know, as I was saying earlier, we don't control macro economic conditions. So, if interest rates continue to rise, that reduces the affordability of cars. You know, and for a lot of people, they're really carved out that's, you know, barely breaking even every month. In fact, if you look at the rise in current car debt, they are in fact not breaking even every month. Current car debt is looking not scary. So, you know, we just don't control macro conditions. If macro conditions stable, I think prices will be stable. If they're not stable, then we would have to lower prices then.

This is a custom software stack. So, but it is designed so that you can run at a high level PyTorch and JAX. So, but then, you know, we have to customize it to actually run on our computer. Custom software. So, the software stack is a combination of open source software and then an internal software all the way to the bare silicon, which is the case for the first computer in the car.

Well, we... You know, this is a period of unusual geopolitical risk. So, I think we're the best we can do is, you know, have factories in many possible worlds such that if things get difficult in one part of the world, we, you know, I guess we'll keep things going in the rest of the world.

Yeah, and I totally agree. Look, prices went absolutely insane there for a while, but... Yeah, and they're recovering now. How about the second half of what you're seeing? Yeah. Yeah, and you know, we're still early in the ramps. Well, not early in the ramp, but early in the cost down curve of Austin and Berlin. Yeah. And so, it takes time to work the cost down at first to get focused on ramps. Ramp brings cost down. Yeah, quality and then yeah. Ramp brings quality cost down. Yeah, and then once that stabilizes, we can divert bandwidth to cost reduction. And so, Austin and Berlin saw quite a decent amount of cost reduction on a fundamental basis from Q1 to Q2. We'll continue to do that work that will be helpful. And so, we're just going to keep chipping away at it.

Yeah, so the saying goes like that. Amateurs win tactics, professionals win logistics. And we've made tremendous improvements in cost and on-time. You know, express cost, we have gone up to pre-pandemic express cost levels now. And our goal is to go further down.

Yeah, it's a game of Thrones. Like Game of Thrones with tennis.