2024 Tesla Shareholder Meeting: Part 2 Q&A

[00:00:02] Hello everyone and welcome to Kilowatt, a podcast about electric vehicles, renewable energy, autonomous driving and much, much more. My name is Bodhi and I am your host. On today's episode, we are going to cover Tesla's 2024 shareholder meeting part Q&A part two.

[00:00:38] So just so everybody is aware, I had to break this up into small little pieces so that it's more digestible and not so overwhelming. I don't want anybody to look at the podcast episode and go, oh my goodness, this thing's an hour and 45 minutes long.

[00:00:54] I'm not sure if I'm going to be able to get through this or not. This is part two of the Q&A. Before we get to the Q&A though, and the second part of the Q&A was actually better than the first part in my opinion.

[00:01:06] But before we get to the Q&A, we have to thank some Patreon members and we have a new Patreon supporter, Patrick. Patrick signed up today. I haven't even seen this come in yet and I don't know why because I was just checking my email before I recorded.

[00:01:21] Patrick, thank you so much for taking the time and becoming a Patreon member. That really means a lot. That's cool. So thank you very much for doing that. Let's see. We got to thank some other members.

[00:01:36] By the way, Patrick, if you are listening, I will send you an email here shortly. Let's see today or tomorrow. All right. Continuing our thanks to our other Patreon members. I'm pretty sure I mentioned Bruce W last episode, but just in case it didn't,

[00:01:55] Bruce W, Isaiah Howard, Bruce C, Michael Muskie, Cybertruck, Rolando Tomaso Chip, Chris B, Sierra Dale, Dawn, Cameron, Nate, Mark and James. Thank you all so much for going to patreon.com forward slash kilowatt or support kilowatt.com and signing up and becoming patrons. I really appreciate it.

[00:02:23] And a special again, thank you to Patrick. Patrick, thanks so much for doing that. I apologize. I just kind of taken off guard. I did not get an email for this and Patreon has been doing some weird things.

[00:02:37] So if I don't get if you send me a message on Patreon and I'm not getting back to you, just shoot me an email. Bodie, B-O-D-I-E at 918digital.com because there's a lot of really good things happening

[00:02:53] with Patreon right now and there are some things in terms of communication that I feel like can be improved. But thank you everyone for signing up and becoming patrons. And that includes new members and free members. Thank you all.

[00:03:08] I really appreciate it and also includes all of our ACAST plus members. All right, let's go ahead and jump in to our Q&A session. Our first question, let me pull it up here, is about the 46 battery cells and really

[00:03:27] what the person wants is an update on how the 4680 battery cells are going, how that whole process is going. And what he gets is kind of a more in-depth answer. So let's listen in. Thanks for hosting today, Elon.

[00:03:41] It's about four years since you did Battery Day and grew out those very ambitious targets of getting to a terawatt hour of batteries. And earlier in your presentation, you alluded to the great difficulty of scaling up this new technology.

[00:03:54] I think as shareholders, we're most interested, can you tell us a little bit more because in terms of affordability, which you also mentioned as this huge burden, these 4680s get manufacturing credits and also on the consumer side get large credits as well.

[00:04:09] If you could scale that so that we could produce full level of Cybertrux and also start to put them back in the Y and the three, potentially it would greatly improve affordability and be great in terms of scale. Sure.

[00:04:23] So achieving affordability is actually at this point more about the rest of the car than it is actually about the battery pack. So the price per kilowatt hour of cells from suppliers has dropped dramatically.

[00:04:44] So there was a massive spike for a few years in the cost of lithium ion cells from suppliers because every car manufacturer put in gigantic orders for battery packs, like giant orders. So the price of lithium and all the raw materials and the battery cells went crazy.

[00:05:04] It doubled. The price of cells doubled. Price of lithium I think went up at almost like fivefold or like 500%. Like it went up crazy numbers. So since then, the suppliers, battery cell suppliers have increased their supply and the orders from other car manufacturers have declined.

[00:05:31] So I mean it is worth noting that it's tough sledding out there for EVs. GM announced I think a 50% reduction recently in their EVs. Ford announced reductions. I think almost every manufacturer has announced reductions in their electric vehicle plans.

[00:05:49] Now Tesla, we're going to increase but it is not an easy market. But to get the cost of a car down, it's really, the battery pack is a piece of the puzzle but it's really everything else too.

[00:06:06] So I really would describe like we're making a car more affordable. It's like Game of Thrones but pennies. Game of pennies. So there's quote to first approximation 10,000 unique items in a car and if each one costs five bucks, then you've got a $50,000 car.

[00:06:26] So to actually get the cost of the car down, you've got to chisel away pennies and to get the 10,000 items to be three dollars instead of five dollars. That's kind of what it amounts to.

[00:06:43] Now with respect to our own cell production, we do see a path to cost parity by the end of this year. A very difficult path to cost parity. Currently our 4680s cost more than our suppliers.

[00:06:58] Now they cost more than the suppliers today but they cost less than the suppliers a year ago. So there's a bit of a feast famine thing with battery cell supply. It's kind of like for DRAM chips. The DRAM industry goes from oversupplied, under-supply, the price of DRAM

[00:07:18] changes like crazy. It's kind of like that for cells. So we expect to achieve cost parity with even the much lower supplier cell price today by the end of the year. So it does sound like Tesla can take some cost out of the battery cells in

[00:07:42] the battery pack because they're not cheaper than their suppliers if I'm understanding this correctly. But I thought this was a good answer. It sounds like they're as cheap as they can possibly go for the most part with the current battery technology.

[00:07:58] Now new battery technology comes in, new chemistries come in and maybe they'll get cheaper when that happens. Next up we have a Cybertruck question. Actually we have a couple so we'll just roll into those Cybertruck questions. Hi Mr. Musk. I've been a long-term investor since 2012.

[00:08:17] I voted for your compensation package in 2018 and again this year. Shout out to Tesla Boomer Mama, Alexandra. I had the privilege to talk to you in 2013 briefly and you spoke about the truck which became the Cybertruck. You're a man of your word.

[00:08:35] I have friends all over the world in Asia, Canada and Europe and I know that Cybertruck has been on tour. How's been the reception for the Cybertruck and when will it be or is it possible to sell and when will it be sold internationally?

[00:08:48] Yeah, so I think we might be able to certify it for other markets sometime next year but for sure this year it's just North America. We did design the car to North American requirements because if you start going with the superset of all international requirements it

[00:09:13] forces a lot of constraints on Cybertruck that would make the product frankly worse. So I think we'll need to make a special version that is for example China compliant or Europe compliant. But it really doesn't make sense to add that complexity until

[00:09:35] we've achieved higher volume production on the Cybertruck. There's still a lot of work to do on cost down for the Cybertruck. I generally would say the level of difficulty of going from say prototype, you go from prototype to production it's 100 times

[00:09:54] harder to go from prototype to production is 100 times harder and then once you reach production to improve the price by 20% or the cost of goods by 20% is harder than reaching production in the first place. Like it's mega pain.

[00:10:13] It's like, you know, and to be frank it's not the funnest job in the world. Like chiseling pennies is not, it's like washing the dishes frankly. Like wash a lot of dishes here. Like it's super fun to make prototypes.

[00:10:28] That's really fun and it's kind of cool to get, it's pretty cool to get the production going. Then the cost grind is a grind. That is like, you know, it's hard work. You know there's some reward to it.

[00:10:41] Like there's sort of a reward to washing the dishes I guess. You have clean dishes. But it's, that is tough. So that's real hard work. I just have to say that there's been a really big reception. Like I have friends that message me and said they

[00:11:04] love the Cybertruck design and it's definitely head-turner. Oh yeah, I think from a demand standpoint we would have plenty of demand in other parts of the world for Cybertruck. No question. Like wherever it goes anywhere in the world it gets a massive crowd.

[00:11:18] So it's not a demand thing. It's really, we've got to reach volume production, go through the cost grind. I can't, it would be difficult for me to overstate the difficulty in the cost grind. It's really intense. And then we've got to recertify the car which

[00:11:36] includes making some design changes for, to meet, to be compliant in other markets. I admittedly have not been to South America. So in terms of every market that's out there, I don't know which markets will the Cybertruck play well in versus not. In North America it's fine.

[00:11:56] There's plenty of room in most Mexican cities or most Canadian cities or most US cities for the Cybertruck to move around and operate. European cities, not so much. I've been to plenty of European places and it'd be really tough to operate a Cybertruck in some of

[00:12:14] those more dense areas or some of those more like the little castle towns and things like that. And then also it would be very difficult in a lot of Asia as well. I've been to a few Asian countries and that is also difficult.

[00:12:30] Now maybe Australia this would work really well in. Maybe New Zealand Cybertruck would do really well in. But yeah, it's a tough sell for Asia and the European market for sure. Next up we're going to find out about how many more foundation Cybertrucks Tesla is planning on making.

[00:12:50] Good afternoon, Elon. I like your shoes. I don't know if everybody else can see them. Out there. Cybertruck, yeah. My question was in regards to the foundation series, do you guys have a certain number that you are planning on making or a timeline of when

[00:13:08] the non-foundation series are going to come out? Yeah, we'll start making it. We'll end the foundation series actually pretty soon and the non-foundation series sometime next quarter. Woo indeed. Woo indeed. The question is how much? How much are these non-foundation or non-Cyberbeast

[00:13:33] Cybertrucks, how much are they going to cost? I would imagine it's still going to be quite high but not 90 to 100 to what is it $130,000 high. So we'll have to see. Let's see here. Next up we have a Neuralink FSD update and

[00:13:52] this one's going to be a little bit longer. Hi Elon. Thank you for delivering shareholder value and congrats on your pay package. Thank you. And I'm Michael, I ask a lot of questions on the quarterly investor calls and I wanted to ask

[00:14:07] you can you give me a little bit more color on the neural net scaling laws that you're seeing as you're training FSD? Like how much data, how much more inference compute, how much more training compute do you need for each successive nine in reliability of

[00:14:23] the full self-driving software and how much data are you currently collecting? Thank you. Yeah, that's a great question. I mean that's something that we go through really every week at Tesla. I spend a lot of time with the Tesla AI team

[00:14:40] and I think probably like a huge portion of, I mean it's really like a big part of my job is like okay, we need to get the car to drive itself. And yeah, you need to solve things on a chip level for inference, for training, for software.

[00:15:08] There's also massive data storage challenges because you've got petabytes of video coming in. And even for say for a big NVIDIA cluster it's actually really hard to make a big NVIDIA cluster actually work in training. It doesn't just open the box and plug it in.

[00:15:27] It's not like that at all. It's like massive effort required to get a big training cluster to operate with high up time. So I mean with respect to like the various versions that we're seeing, like you go from version 12.3, just ignore the 12 and think of it like 3, 4, 5, 6.

[00:15:55] That's actually the way to think about it. Ignore the 12 part and just call it like version 3, version 4, version 5. That's version 6. That's how to think about it. Between each one there's anywhere from I don't know, a minimum of a doubling of improvement to in some cases

[00:16:15] I think a 5 to 10 fold improvement. Especially as we eliminate sort of say, I think we're close to eliminating almost all static object collisions. So like curves and medians and like we're pretty close to having it, to have almost zero static object collisions.

[00:16:46] So once you get to zero static object, now you have a gigantic improvement in miles between intervention. Then you get into classes of dynamic objects that are, there are different classes and different, things can get obviously very complicated if you've got a complex intersection with traffic lights

[00:17:07] all over the place and like somebody bumped the one traffic light, now it's pointing in the wrong direction. Well the traffic lights broken and there's someone who is not obeying the traffic laws because they're just running a red or they got confused about the traffic light.

[00:17:23] So you get into these very complex situations that are much harder to solve. So, but then like as I was saying earlier, it actually gets, as the system gets better, it gets harder to figure out which AI model is better. Like because now, you know like okay,

[00:17:55] it's thousands of miles between intervention. How do we as quickly as possible figure out which AI model is better? And when you make these different AI models, they're obviously not like super deterministic. So we have something, a new model that does,

[00:18:13] that eliminates one problem but creates another problem. So we're trying to solve this by a combination of simulation, uploading models, having them run in shadow mode. So it's actually kind of helpful that not everyone has full self driving

[00:18:35] because we can see, we can run it in shadow mode and see what would this new model have done compared to what the user did. So since we've got millions of cars that we can do this with, that gives us a delta between

[00:18:50] what the AI model predicted would do and the user would do. And if you kind of sum up the errors between them, you can see oh, there was a bigger error stack from this model versus that model when you each uploaded them into 100,000 cars.

[00:19:10] But that's the biggest limiter right now. It's not training, it's not data, it's actually testing the AI models and then figuring out clever ways to figure out if a new model is better or not. Like we know there are sort of particular intersections that are difficult.

[00:19:28] And so if we just, like all the models will be great at driving on a well painted road with no intersection. They'll all be great at that. But will they be great at the complicate? So we've identified several thousand intersections in the United States where we'll actually take

[00:19:50] the new model and try the car in those very difficult intersections. Like the, you know, it's sort of for the famous like Chuck's left turn. Yeah. So if you imagine Chuck's left turn, hey Chuck. I wonder what the neighbors think. But there's many such complicated

[00:20:22] traffic situations in the country. So we've identified several thousand which we then send cars to those locations to compare the models. And then as I mentioned, we have yet to uncork any of the ability of hardware for which we'll do later this year.

[00:20:43] So I think the exponential rate of improvement of miles between intervention will continue and will accelerate. All right, Elon mentioned, he mentioned, you know, FSD getting to the point where reduce those times when it rubs against the curb and stuff like that.

[00:21:06] My front left tire or my front left wheel got bit by the FSD curb rubbing machine. Like I was pulling away from an ATM machine. I put it in FSD and it rubbed my wheel on the curb. Now I stopped it pretty early on.

[00:21:28] It didn't do a lot of damage, but yeah, that did not make me happy. But I probably should have pulled away from the ATM machine before activating full self-driving. So I'll take partial credit on that little rub, but it was very loud and very unfortunate when it happened.

[00:21:51] So let's talk about Elon's answer though. Those complex situations that he talked about, those are the things that you need to solve before you can have RoboTaxi everywhere. Now I do agree with the better the software gets, the harder it is to solve that next problem.

[00:22:09] And by solving that problem, you might create two more problems. That's how software development works. That is not a, that's not new. That is something, that's exactly how software development works. As soon as you think you solve one problem, you create two more.

[00:22:26] It's just, you know, it's a whack, it's a game of whack-a-mole is what it is. I do like that Elon mentioned that they send cars out to problematic areas to drive those areas to help solve those problems. I think that's a bit of insight we haven't gotten before.

[00:22:42] But yeah, we still have a little bit to go. Hopefully not as long as I think, but I'm not as optimistic as Elon is with full self-driving and when it will actually be full self-driving, like actual full self-driving level five autonomy,

[00:22:59] even level four autonomy I think we're a ways away. Next up we're gonna hear about the, can you personalize or will you be able to personalize Optimus? Hey Elon. I was wondering if as you've called a Tesla, basically a robot with wheels.

[00:23:26] And so I'm wondering kind of as that, as that logic transfers over to Optimus, what kind of, will Tesla use any large language models on top of the current Tesla robot's capabilities and also if so, will Optimus have any kind of personality and would that be something

[00:23:48] that the end user might be able to have control over? Yes to all of those questions. Yes to all. So like we're headed for a wild future. I mean, wild, wild, wild. You know, cars will drive themselves. People will have super helpful humanoid help robots.

[00:24:17] And yeah, you'll be able to customize the personality, customize the voice and really kind of the robot will kind of get to know you as well and know your preferences. You won't even have to say things because it'll just kind of understand you.

[00:24:32] I think we'll get to this point where it's like, it'll be odd not to have a robot buddy. Like, you know, sort of not to have your C3PO and you know, R2D2. Now, like much of everything that I've talked about for this particular shareholder meeting,

[00:24:51] I don't think we're as close to what Elon says we are. We are not two years away from having a robot that can teach itself or you teach it one time and it can actually do it. I don't think we're that close.

[00:25:07] Boy, I hope I'm wrong because I am getting older and I would like a robot to do all of the menial tasks that I don't wanna do. I would also like a robot who would be able to co-host this podcast.

[00:25:18] And I would like that robot to like be kind of like a Don Rickles to everybody they meet. In the sweetest way, Don Rickles. I don't think we're gonna get that for a while. Maybe in the first 10 versions. I don't think we're gonna get that.

[00:25:35] But you know, I will be pleasantly surprised if it happens. Next up, we're gonna hear about Optimus in the workplace. Hi, Elon. Max here, huge fan. Thank you for hosting us. I wanted to ask you, how much integration work do you think will be needed to deploy Optimus

[00:26:00] into different types of factories, industries, like technicians or just have them go watch? And how do you foresee that? Well, I think if you fast forward a few years, it really will be quite a generalized robot. Like two years from now.

[00:26:20] Like next year, like this year it's still quite specific. You know, task specific training. So to do a particular task, we have to kind of do a fair bit of training to solve that task. But over time, I think you sort of accumulate

[00:26:40] these kind of training primitives where like, calling like walking for example is a training primitive. Picking something, identifying an object and grabbing it would be a primitive. And these primitives are then generalized to anything. So you could say pick up a cup, pick up a shirt,

[00:27:02] fold the shirt. These things will generalize quite quickly. So I expect probably not next year but like in 2026, it's really the kind of thing where you could tell it to do kind of a wide range of tasks, a wide range of tasks including tasks

[00:27:27] where it has not specifically done that exact thing before probably in I'm guessing 2026. Sure. And do you foresee a cheaper version where tasks don't require like walking for example? So the hardware will become a lot simpler or cheaper? Well, we're gonna focus on making just a fantastic

[00:27:54] humanoid robot at scale and really achieving scale economies and building that humanoid robot. At very high volume, if you're really good at manufacturing, you can get the cost of almost anything down to its constituent elements. Essentially the weight and if you just look at the weight

[00:28:20] of the raw materials times the value of those raw materials excluding circuit boards, you really can say like if it weighs less, it's gonna cost less. And what I'm saying is that it's better for us to achieve very high volume with optimists

[00:28:40] than to have a wide range of robots that have lower volume. Yeah. Yeah, I agree 100% with Elon. Like Tesla tried to build the factory with all these other robots that weren't humanoid and they already do build, Tesla does build robots that build cars so for optimists

[00:29:03] it makes the most amount of sense for that to be a humanoid robot or at least in the way that they want it to work now. Otherwise you can just get some robot arm to do all of the other tasks.

[00:29:17] So we're gonna move on to our next question which is all about lithium refining. So let's go ahead and listen to that. Thanks. All right, I think so this will have to be the last question I think or maybe the last two. Yeah, all right.

[00:29:36] So last two questions, you and your guys. Hey Elon my name is Steven. I'm truly thankful for all you've done for humanity and watching your companies and what you've done has been the most exciting, hope inspiring movie that there's ever been so thank you so much. Thank you.

[00:29:50] Yeah, thank you. And my question today is about lithium refining. In the past you stated that would be a limiting factor in the future and I know Tesla has built the refinery in Texas. I was just kind of curious your current opinion about the refining.

[00:30:04] Do you think there's enough capacity coming online or will Tesla need to expand past Texas in that regard? There's enough lithium refining if you count China. If there's a geopolitical challenge on that front then there may be a shortage of lithium. But China's very good at refining.

[00:30:31] But I think just Tesla figuring out lithium refining, understanding what it takes and then once we understand how it works then we can scale it up from there. And I think it's pretty important to be careful with geopolitical risk and just make sure

[00:30:50] that if in a worst case situation that Tesla is okay. I also agree with Elon on this. The world is a crazy place right now. I don't think we're gonna go to war with China but I do think we are going to have little economic spats with China.

[00:31:12] They're raising the tariffs on Chinese EVs, the USs. Then in return China's like you can't use these parts and the US is like you can't use these chips from China and this is all going back and forth and back and forth between our government and their government.

[00:31:30] Do I think we're gonna start launching bombs and send soldiers to China or they're gonna do the same to us? No, I don't think that's gonna be the case. However, I do think that we're in a slap fight with China and we're just gonna keep slapping each other

[00:31:45] until somebody cries and then eventually we'll either make fists and punch each other in the face instead of slap or somebody's gonna cry and we're gonna be like you know what? We should probably go to counseling and work this out. Hopefully it's the counseling side of things.

[00:32:02] It's kind of irritating to watch governments do this kind of thing from a citizen's standpoint. It doesn't make friends and it makes it difficult for all the people living in those countries to be nice to one another and I'm not blaming the United States

[00:32:20] more than I'm blaming China. I think they're both pretty, it's an election year. Let's just put it that way. It's an election year so we'll move on from there but yeah, that is it for this episode. I do have one more clip though before we go

[00:32:39] and I'm gonna play that at the very end so you'll hear the outro music and then you'll hear one more clip and that is friend of the show Jessica Kirsch. I wanted to give Jessica one, the final word on the podcast. Jessica went to the shareholder meeting.

[00:32:56] She was able to ask a question. If you want to know more about what Jessica Kirsch is doing because she is down at Starbase doing all things SpaceX, you can go to youtube.com forward slash Jessica Kirsch and check out what she is up to.

[00:33:14] All right everybody if you wanna email me it's Bodie B-O-D-I-E at 918digital.com. You can find me on X at 918digital. I have a podcast called Beyond the Post. You can find links in the show notes for that if you're interested in becoming a digital creator

[00:33:29] or if you are one and you just kind of want a different perspective on things. And yeah hope everybody has had a nice Father's Day if you celebrated and I will talk to you all on Friday where we'll cover the actual business of the shareholder meeting. Hi everyone.

[00:33:58] Jessica or my name is Jessica and I have lived full time RV for about three years in South Texas and I just wanted to thank you for your Starship or for the Starship program and everything you're doing there. In addition to Tesla.

[00:34:14] Do you work at SpaceX or something? No. Okay. Okay but you live in an RV in South Texas? Yeah. Okay. Hi and in my RV park they don't allow car camping for Teslas or electric vehicles at all if you're not pulling a trailer

[00:34:32] and I feel that there's a huge opportunity there but also for just for example people who are coming into town for Starship launches in Teslas if they could park somewhere overnight in trickle charge. Okay. And maybe have access to bathroom facility or just something simple like that. Okay.

[00:34:57] Just something I wanted to just put forward because I know that a lot of people come into the park and are disappointed they can't stay so. Okay. Got it. Just a thought. All right. I'll inquire and see if there's like maybe some place

[00:35:11] that would allow people to stay in their cars which in a Tesla you really can often stay quite comfortably in a car and yes and charge. All right.