Okay, so those are really cool ones. Our cars drive themselves from the end of line to the logistics yard now. They also go over the bump, squeak, and rattle track, listen to the noises, and document them back to tell people what to fix. We call it full self-hearing.
Why not? Wow. That is cool. All these things talk to each other, but they don't think to each other. And now they're thinking to each other, saying hold on, this is happening downstream, we need changes upstream. In terms of tolerancing and definition, we scan all the cars at the end of the line. Immediately the body guys know we need to go change the standings. I just saw JD Power came out with this two days ago. We're number three in IQS now. I think the amount of real world AI that's going to be around you because of us is going to be mind-blowing.
All right, so in a week from Tuesday, there'll be some cool news about things happening on the campus of Texas, and I can't talk about it right now. Today we're joined by Lars Moravey, Tesla's vice president of vehicle engineering and one of the key people turning Elon Musk's vision into real products. Lars shares what's really happening behind the scenes on CyberCab and Optimus. Plus, he gives real examples of Tesla's use of AI and why he believes people are dramatically underestimating the scale Tesla's preparing for. He also said a major announcement is coming Tuesday, July 7th around scaling big. Joining us is Jeff Lutz, a former Fortune 100 supply chain executive. There's a lot to unpack, so let's jump right into our conversation with Lars Moravey.
Thank you so much, Lars. Appreciate this very much. First of all, I think it's a good time to just pause for a second and thank you and the Tesla team because you guys didn't just reinvent the car, but you've completely changed the automotive industry and you guys achieved the mission of accelerating the path to increasing sustainable energy. So, thank you very much for doing that.
I want to thank you for the appreciation, and I want to say thanks to all the teams that do it. A lot of times I get in front of people like yourselves John and Kelvin behind the cameras and you get to talk, but there are thousands of people that make the change. The hardest thing in the world is to change something. Change yourself, change manufacturing, but they do the hard work day in and day out, so your thanks through me to them.
Very much appreciated. So we've definitely made a change to the new book now. I've got to ask you about the CyberCab. We are watching CyberCabs coming out of Giga Texas. We're seeing 150 of them in Giga Texas. Can you tell us how is the production of the CyberCab ramp going?
How do I answer that? Good question. We built CyberCab for scale. The reason we built it was primarily because part of the original mission was to accelerate the world's adoption of sustainable transport, and one of the things we realized was to do that we had to make the most affordable transportation possible. We believe CyberCab does that. It's built for scale, and in order to scale at the level we want, you have to build things and test them repeatedly. So where we're at now: the production lines are installed, we're running them. All the autonomous ride vehicles you see are part of that test fleet, and the majority have gone into training with Autopilot. Our models are great but they still need to train. You've seen the training vehicles with steering wheels and pedals, ironically from Cybertruck. We've just started the burn-in process on the burn-in track we built on the edge of the campus. Early manufacturing, early product always has issues – production quality, supplier quality, design, infant life. Every unit we build is a production unit and also a test unit. We're trying to learn as much as we can as we get ready to scale to levels that I don't think people have fully absorbed yet. In a week from Tuesday, there'll be some cool news about things happening on the campus of Texas, and I can't talk about it right now, but that's part of the scaling effort.
Wow, okay. Thank you for sharing that. So just a follow up on this: would you say that the Cyber ramp is as expected? Are you ahead of schedule?
I think it's where we thought we'd be. Production is hard. CyberCab is built totally different. It's not like building a Model Y off a Model 3 or updating a Model Y or Model 3. Everything we built and the way we built it is new – from the way the harnesses are built to the way the car is assembled to the way the battery is put together. A lot of new processes. The automation is ahead of schedule actually. I'm impressed with the rates we're able to achieve already. We maybe made 150. I'll stick with that number. But it's not like you built them at pace; you also want to make sure the automation is running as fast as you want it to. We continue to speed up the automation. Lines are over 90% automated all in. I expected us to have more issues, and the teams have come a long way since production hell into figuring out how to make things work.
I was just going to ask because if you compare the last everything new vehicle Cybertruck even Model 3 at the time, what's different about how the teams approached CyberCab? Is it too early to say or is it really getting out better?
Fear. It's just fear. Those of us in leadership that were here for production hell never want to go back. Scars, deep unhealable scars. That's actually very effective. I think most of the leaders in the teams have been there. Two big things are different. One, fear is different. Cybertruck we ramped up pretty quickly considering how new it was. Even the turnover of Model Y last year – three or four weeks, four factories all at once, that was insane. When Ford did it in 1914 with the Model T and Model A, it took them six months. That was the next biggest automated changeover at that level I can think of. Fear has gotten us in terms of being prepared. We do a lot more virtual commissioning now. We build everything virtually before we put it online. We do a lot of coding within the virtual commissioning process, so when you turn the machine on the first time, it works much better. We don't ship machines without testing them anymore. A lot of rigor has gone in. Agility is key to Tesla, but there's a lot more certainty in making sure it works the first time before you get it there.
Is CyberCab a platform for you now? You think of the new manufacturing operation unboxed, is it a platform?
CyberCab is going to scale to a level where it's its own platform. The reason we redid it and rethought it the way it was is because we believe we're going to build more of these than anything else we've ever built. Transportation for humanity will hopefully change completely. It's going to take a minute, but when you get something better, something cheaper, something that's just better on all fronts. I always think about technology in terms of innovation over time. It's ironic – Seagate over there, but hard disks to flash memory. What is worse about solid state flash memory? It's smaller, cheaper, faster, more reliable, more durable. The only problem initially was it cost more, but everything else was better. I think with CyberCab we're in a situation like that for transportation: cheaper, safer, more reliable, more efficient. Because of the way we've built it and scaled it, even the cost will be better up front. It took 10 or 15 years for hard disks to almost completely disappear. What laptop or desktop can you buy with an actual hard disk anymore? Solid state memory. I think that's going to happen to some extent with transportation. To do that, we really had to change a platform. So it's its own platform. Whether we need to build other vehicles off that will depend on the adoption rate of autonomous vehicles.
Yeah, makes a lot of sense. So I saw a person who saw a CyberCab for the first time and said, 'What is this?' And then us who follow it closely know it's the most efficient vehicle ever made. Is there anything that is being completely underestimated that people don't know about the CyberCab yet?
You said the word but I don't think you really feel it the same way I do – efficiency. It's not efficiency in terms of watt-hours per mile. That's okay. But it's efficiency in the end-to-end process from raw material in to vehicle out in terms of cost. If you think about what's most important with autonomous vehicles in the future, it's cost per mile or cost of ownership. To build something as efficient as CyberCab you need to start at a white sheet of paper. Our competitors – Zoox gets it, they started white sheet. Waymo is buying a car and putting compute on top, that's not efficient enough. When I read business outlooks from Ark or Wells Fargo about autonomous vehicles, I don't think they actually understand the efficiency. You're going to see operational costs that blow their minds. It starts with the capex up front to build a car. Every bit of energy that goes into that car and how it operates – waste in charging, recharging, supercharging – has been considered. I don't think anyone else can do that because they don't have all those things in house. We build things from metal from ingot from pellets, from sand and the earth, and turn it into vehicles at our factory. Then we build the compute, write the software, own the charging and service. If you were rewriting a playbook about Tesla, you'd think we had this idea for a long time.
I have been saying that over and over again. Did you guys know this 10 years ago or did you get lucky?
No, Elon's been pushing us since 2014. We've been planning a lot of moves over the years to get us to a place where we're ready to do this at a level that I don't think anyone else really can. What's being underestimated is the efficiency and the volume scale that's going to be out there of this product.
So when you walk the cyber factory line – the unbox model – is there anything you're really proud of that you think you guys did something special?
I'm most proud of the people that did it. We drew on a whiteboard five years ago how we thought we would build this thing. Everyone was like 'No way.' Lines feeding lines, material in one end, cars out the other. The final line is only like 20-25 cells, whereas normal GA is 200 cells. Output per square meter is ridiculous. It takes a certain amount of belief from the people that did it to actually make it happen. You can't build something like that and not think it's going to work. I walk through and I can't believe all these people thought it would work. I'm proud of the studio team for giving the engineers freedom, working with Franz's team. We changed so many things about how to build it because those teams worked together. The design teams conceded to manufacturing, and manufacturing didn't take a second seat. The automation groups, suppliers, supply chain – I walk in and think 'I can't believe we made it real.'
That's insane. So with the CyberCab coming out and it's already so low-priced – $30,000 maybe lower – the efficiency of the battery, automation. Is there an area you think you can still improve on?
Of course, everything will always improve. I think we left some vehicle efficiency on the table. Maybe 10-15% is a good goal to cut out. But I have a real passion for mass efficiency. It's a thankless job to take a gram out of a part or a watt-hour per mile. The downstream impact on unit economics is incredible. The thing I think we'll improve is the customer experience. We made our best effort at understanding what people will do in an autonomous vehicle without driving, but we don't really know yet. Will they work, be entertained, sleep? We did a good job creating an environment in CyberCab, but we'll learn more and make a better environment.
Just hoping I don't have to put a toilet in it. I saw that. Do you cover China? I love that answer so much. Let's move on, switch gears. We want to talk about Tesla's competitive advantage. There's a quote from the former Shanghai leader, Song Gan: 'Tesla's competitive advantage is manufacturing.' I always view it as that world between engineering and manufacturing together.
It's one world. I think he's right. Him and I had a good relationship, and that speaks to his comment. What head of design engineering has such intimate thought? When he was going to leave, he called me first. That personal level of integration between our factories and engineering teams – we are one engineering team: manufacturing, automation, design. There are organizational structures, but it's not over the wall. We sit in the same building, work together every day. When you have a manufacturing complexity issue, you walk across the aisle and talk about it. In my former life, I couldn't remember many manufacturing equipment process engineers; I came in late after design and validation. Here, it's from the top down. Elon tells us, 'You're all manufacturing engineers.' And because we built the first batteries for Roadster or Model S, no one knew how to build that. Design engineers had to be there because there was no machine to buy.
That gets to this next question. In the Bay Area, other tech companies with high volume products are extremely risk-averse. Tesla introduces clean sheet designs like Cybertruck and CyberCab for high volume. How does Tesla approach this differently?
I think 'risk' is a misused word. Investors use it a lot. We look at probability of success or failure – calculated risk. When you know the outcome or understand the impact, you can make better judgments. It starts at the top. I had one of my best engineers leave today, and he said he doesn't think there's a leadership team he'll ever have where he can explain a complex problem in two to three minutes, get options, and get direction. That's because we promote engineers to lead. I don't have an MBA, never took a finance course. The people in charge making decisions about what risk to take understand it at a level that doesn't exist everywhere. We don't have layers of program managers and product managers. An engineer in vehicle dynamics emailed Silio and me with ideas about how to help people use FSD more, and within 20 minutes both Silio and I replied with thoughts. Where else does an engineer email two VPs and they both reply with actionable feedback? That level of flat integration and servant leadership is often misunderstood about Tesla.
So you brought up FSD awareness. What happened on Monday – the world needs to understand the capability of these cars. They're not just cars anymore, they're self-driving robots. Do you feel this will be a solved issue where people understand, or do you have plans to increase awareness?
Accidents are tragic however they happen, and we work our best to prevent them. Safety always comes first – it's our priority. Public sentiment has to come with experience. People are scared of flying until they fly and land. Autonomous driving is similar. You can tell your neighbor, but until they get in and try, they'll remain skeptical. It's unfortunate when media says misleading things, but we take the high ground and stick to the facts. We've never aggressively rebutted, and that stands for where we are as a company.
From a safety perspective and belief standpoint, keep calm and carry on. That's where we go with that. I think in the end, the product will speak for itself. It already has, ever since 2014. Adoption rates in the US are crazy. It's a slog, but advertising won't change it. You have to experience it, and we'll get people to experience it through cyber cabs that take them from point A to point B quicker, more efficiently, and safer than ever before.
Can you answer this question? Elon made a bet with MKBHD who said he didn't think you'd have a cyber cab sold to a consumer by the end of this year. Elon said, 'I'll take that bet.'
Is that something you can comment on?
I don't know what Elon's doing these days in terms of bets and...
You're getting them made and into the lots.
Yeah, that's my job—to make a product. We made a product that's safe, functionally safe, built on there. The deployment will happen. I find humor in social media, but it's not my place. I try not to get involved in the politics.
You still don't even have a profile pick. What's going on with that?
Can you just like, do you guys not know it's me? If I put one on there, it'd be like a fake account. Elon gives me crap about that too. He's like, 'Why don't you just get them?' It's like, 'It's too hard. You have to click like six buttons.'
Oh yeah, he's got to work on optimizing that. Too many buttons.
Can we talk about AI? You talked earlier about automation in the factories. Can you speak about how you specifically are using AI in the factories or other areas of Tesla's operations?
Our software teams have massively benefited from code generation by AI. They use it to get base code out of the way before doing the hard work. Our neural nets and FSD have used AI generation for a while. Beyond that, there are three or four key areas. One is transitioning experience quicker to the engineering team. We have mounds of owner data about lessons learned, and we've coagulated that into our own agent so engineers can search faster. When I was a young engineer, it was a lot harder to get that information.
Used to be a library. You had to go to the library.
I literally remember we used to call it the orange files room because all the files were orange. You had to pull out the paper and put it under the microfiche.
Where were those lessons learned?
That's one key way. The other is equipment automation KPIs—anomaly detection and trending. We have KPIs, KPCs, and SPK charts on everything. We've had anomaly detection with big data for a long time.
Obviously we have KPIs and KPCs and SPK charts on everything. We're always watching it, and we have anomaly detection for a long time with big data.
But sometimes your spec limits are not the problem.
Yeah. Sometimes it's just a small trend this way or that.
It's a confluence of two things, right? Or this happened and that happened.
We have a lot of that. Jimmy's team on the quality data side does a bunch of work to integrate tools to check torque traces and make sure there are no anomalies. We've implemented a lot of anomaly detection and searching from AI into various things. Another thing is linking systems—manufacturing operating system, torque database, inline control system—they all talk to each other but now they're thinking to each other. For example, body fit: we scan all the cars at the line, and the body guys immediately go change the standings. JD Power came out with a report two days ago; we're number three in IQS now. It's not just AI, but getting information to the people who need it is paramount. We've unleashed that in the factory and will continue. The last bit is on the service and end product side—similar to lessons learned. If Herbert fixed a bearing problem in Toronto, now Jeff in San Diego knows. Service times are going down. AI has been an advantage in speed of iteration, information sharing, and quality. We'll continue to push that. It's empowering our team. And of course, we have big products coming with Cybercab and Optimus—real-world AI into the market.
That's good to hear. It's so obvious that this is what you guys are doing. Tesla is known as the leader in manufacturing and AI. Thank you for sharing that you're integrated more than other companies. Can you do a cool one?