General Motors Company (GM) Earnings Call Transcript & Summary

Good morning, everyone. My name is Billy Kovanis. I'm the VP on the Autos and Shared Mobility at Morgan Stanley. Here is my colleague, Joe, who's the lead semis analyst. Very excited to have Kyle here from Cruise. Kyle is the CTO, CEO, Founder and CEO of Cruise. Kyle, thank you for joining us.

Thanks for having me.

Kyle, if you can start off with maybe a brief overview of your background and the mission of Cruise.

Sure. Yes. So I've been working on self-driving cars a lot. I started actually when I was 13, built one of those kiddy cars that kids can ride around, put a webcam and a Pentium 233 and tried to make it drive. So I've had this idea in my head for a long time. And then with MIT and worked on self-driving cars there, which was back in 2004 or so, part of the DARPA Grand Challenge; left, started a company that became Twitch, which is now owned by Amazon, with 3 other guys. And then left Twitch, started Cruise the next day back in 2013, been working on that ever since. And obviously, it's very exciting times for us now, 8 years into the journey, still a lot ahead, but it's a pretty good time to be in the space.

Great. And given the recent management change where you became CEO, can you talk to us a little bit about maybe the different vision or strategy that you have potentially versus prior management?

Yes. Well, I think every CEO has a different style of leadership, and I think we all bring something to the table. But in our case, the mission is exactly the same. We still want to deploy driverless cars at scale. And really both our short-term and long-term objectives haven't changed. We're still doing ride share in urban environments where the majority of the potential impact is, majority of the customers, the highest willingness to pay, and then further out in time, still intend to expand geographic areas to lots of cities, lots more customers and then lead into personal AVs that in partnership with GM, which we think will really explode the number of vehicles out there and the size of the market and the revenue we can generate.

Great. And maybe on the market, how do you define the TAM of Cruise? And what are the key assumptions underpinning that TAM in terms of dollar per mile and number of addressable miles?

The TAM conversation is a fun one because there's not that many markets that you can say with a straight face, it's a multitrillion-dollar market, but that is the case for personal transportation. I'm sure you know all the numbers, but in the United States, there's 3 trillion passenger miles each year. And regardless of whether you look at Uber's, S1 or AAA stats, the cost of car ownership per mile is between, call it, $0.60 to $0.80. And so you can do the math there. It gets big pretty quickly. And so we see a pretty easy path with our cost down curve to $500 billion, $1 trillion market size. And that's the floor. That's assuming no one changes their behavior and it's like people don't decide that owning a car is a huge pain, which it is. You have to take your car in for maintenance, pay for insurance, find a place to park, you got to clean it, all that kind of stuff. And that's assuming people don't value their time at anything and they're perfectly fine sitting behind a wheel for an hour or two a day when they could be doing other things. So when you factor that in, we think the TAM could be much, much bigger. But multitrillion dollar TAM, we pretend it's just kind of sitting there for anyone to go after, but it's not. Really today, it's up in the clouds, and there's only -- well, I'm a little biased, but basically, there's very few companies that I think are going to have a chance to actually -- or have earned the right to actually go after that TAM. I mean, where we are today is really just the tip of the iceberg, companies like Cruise, which we're operating driverless service in San Francisco, real users right now. And to get there, that's essentially 8 years of effort, lots of money, lots of energy to go into it to enable us to go after that TAM. And so what you see today is very small beginning stage of this. We just did our first rides to the general public 6 weeks ago. So very, very early stages. But now that we've crossed that line and got to the point where we can operate driverless cars in a major U.S. city, we're literally the only company in the world that's doing that. What you're going to see now is us enter into the rapid scaling phase, where it becomes, can we get our early users to love the product and come back again and again. And then once they love it, the tech barriers to going from 1 city to 3 cities to 10 cities are very, very small in proportion to the amount of work it took to get to city number one. And so we're really excited about the fact that there is an enormous TAM. And I think we're kind of ahead and uniquely positioned to go after that. It's not like there are 20 players in this industry. It's -- and you can count them on one hand.

Definitely, and you're trying to solve one of the most complex problems out there in terms of some of the legal, technological even moral dilemmas out there. And with that, what do you think are the most significant challenges to achieving autonomy at scale?

Well, it's a good way to phrase it. I mean, you kind of touched on it. There are lots of angles to this. You have to excel at operations. It takes a lot of capital. This is a capital-intensive business, at least early on, takes the engineering side, execution and then there's the safety and regulatory piece, which is uncharted territory. If you will, a lot of the regulations in the U.S., you read them, the Federal Motor Vehicle Safety Standards, they make a lot of assumptions about a human driver sitting there behind the wheel. So thinking about what that means when you take the driver out is difficult for regulators to work through, but also difficult for private companies to figure out the right approach to deploy responsibly. So for us though, the biggest challenge, and I can't overemphasize this, is actually the -- all the work not just to make a ride that's smooth and safe, which is kind of the table stakes, but then do all the work to handle the fact that there's no driver in the car. So you can put a human supervisor in a test vehicle and have it drive around the block or drive through a city, but you're kind of papering over all the weird stuff that can happen when there's no driver. So for example -- a simple example is if there's a police officer behind an AV and it flashes its lights, that AV needs to detect that, pull over and then the police officer needs to have some way to interact with AV. And that whole interaction has to work and it has to work smoothly. And that's like just one of the many long-tail things. Another -- there are so many examples of that, like a school bus puts its stop sign out. You don't want the AV to go around that even though it can find a path it actually has to obey and figure that out. And then on the safety side, like we've built these cars to be redundant. It can't be like your cellphone or your laptop, where it freezes and locks up and you just power cycle it, right? This is a vehicle on the road. And so what that means is every hardware system has diagnostics and failover modes and backups so that no matter what happens, whether it's a software issue or a hardware issue in the vehicle, which we know they will happen at some point, and you need to be ready for those and ensure that the vehicle can pull over or do what it needs to do to keep the passenger safe every single time. And that is sort of the hidden and less glamorous work required to take the driver out of the car. And those are the things that you have to do to even earn the right to go after this giant TAM.

Got it. So the CPUC just granted Cruise and Waymo, the ability to operate passenger services but with a safety driver. Kyle, when do you think we can remove the safety driver from the vehicle?

Yes. The California permitting requirements are a little bit complicated. You did -- I think you -- we got a pretty good understanding of it there. I would say on the -- first of all, we have 5 out of the 6 permits we need to charge customers for driverless rides. Right now, just to put it in perspective, every night, dozens of AVs, literally dozens, go out and are picking up passengers, members of the general public. The only thing that separates us, that last permit does, is it lets our engineers flip the switch on the back end, so it starts accepting people's credit cards for rides. From a technical standpoint, there's basically zero incremental work to get to revenue. On the regulatory side, we've been working with the CPUC, the California DMV, other regulators, and they're pretty interested in seeing this happen. There's obvious benefits that self-driving cars bring in terms of safety and accessibility of transportation. And so they're incentivized to do that. And there's a lot of advocates that see it the same way. But the whole issue of permitting is unique to California. There are a lot of -- 49 other states, many of them are open arms for this and also very ephemeral. I don't think we're going to be talking about permits a few months from now because hopefully, we'll have all the ones we need to start charging customers and that will sort of be a thing of the past. But yes, and so I think we're in a good shape there on permitting. And we're getting all the development we need to do with the driverless cars and having real people in them, even without that last permit, like it's not holding back the development or the acceleration of the business. Obviously, if we're having the same conversation next year on the stage, and we still don't have a permit, that's a different story, but we don't see that being an issue.

Got it. And how do you foresee the vehicle ramp out to 2030 playing out in terms of number of cars?

Yes. Again, like the -- we do see a path to 1 million-plus AVs. And I think the thing that is important to understand is that this is really a tech where -- a business where the main bottleneck is the technical capability. Getting to 1 million vehicles -- 1 million AVs with GM as a partner is a viable thing to do. The challenge is building the technology that can support driverless operation at all, can make happy customers that are going to come back and use it again and again and then can scale to enough cities and markets that you can basically put that many vehicles into supply. And so again, like for the last 8 years, we've been making pretty consistent and exponential progress on the core tech metrics, the safety of the product, the comfort of the products and other things. Right now, we're 6 weeks into our first driverless deployment with real passengers. We're just getting started. But I think what you're going to see, which is probably unlike a lot of other things in the auto industry, is that continued exponential scaling. And so you look at where we are now, we're talking dozens of vehicles. That's a long way from 1 million. But once we scale from one city to the first three, that's a very small lift relative to what we've already done, that will be interesting and sort of on a path, but then the work to get from those first 3 cities to the next 20 is even smaller. And so I believe you're going to see this exponential ramp as we knock down some of those remaining tech barriers, which again are relatively small in magnitude compared to what it took to get to where we are today.

Absolutely. And maybe just touching on the GM relationship a little bit, a couple of connected questions here. The first one is a quick one. Can you remind us what percentage of Cruise GM owns?

Over 60%. So they're a majority investor.

Got it. And how much autonomy does Cruise have to operate and make decisions independently within GM?

Well, I think the notion of autonomy or independence is important for a couple of reasons. One is it kind of sets -- it empowers the leaders inside a company like Cruise to make decisions and do what they need to do. But in practice, achieving independence is all about alignment between the parent company and in this case, Cruise. And so what's really important to me is that Mary Barra, the CEO and the GM leadership team and the GM Board, see basically our future the same way we do. And that's what we've achieved. And so it's -- they've been very supportive, helped us go after, do the things we need to do to attract and retain the best talent to capitalize the company. So it's been a pretty good relationship, especially since -- just in a few months, I've been in the CEO role.

Got it. And just a final question on that. We get asked a question a lot by investors, is Cruise better off consolidated under GM, some of the points you mentioned or is it better as a stand-alone entity? Just how do you respond to that?

Look, right now, building AV technology, integrating it into a production vehicle and manufacturing it requires the teams at Cruise and GM to work in lockstep. Like they're in the same meetings. They're in regular execution meetings. And I think we benefit greatly from being a part of GM. And so I don't know what the future holds exactly, but right now to get us to this critical -- to keep us on this trajectory of getting to 1 million vehicles by 2030, this is the right move for us today.

Makes sense. Maybe one more question and then I'll come to you, Joe. In your view, Kyle, what are some of the key factors that will determine the winners and losers in the autonomy race? And is it really possible to measure who's in the lead in autonomous today?

I'll answer your second question first. I mean if I were in your seat, I know you're trying to figure this out every single day, like who's in the lead and why, and even myself having like intimate knowledge of the technology and the business, it's hard to know. What I would say and what I've said in the past is that given the amount of work it takes to take the driver out of the car and put a vehicle on a public road, especially in a major city, that's a proof point, that's an existence proof this technology has reached a certain level of maturity. And I think that really is a line you can draw in the sand between companies who have reached that milestone and have the conviction behind the quality and safety of their technology to deploy it and those that are somewhere still in the R&D phase. And we don't know if they're a week or two away from that or many months away, but we do know that they haven't done all that work yet. And so that's, I think, an important point for determining like the companies that are closest to actual commercialization and turning this into a business. Beyond that, when we reach the phase where I anticipate there will be a handful of companies that do ultimately operate driverless cars in the U.S. cities, then it becomes about more traditional metrics like do the users love it, do they come back again, what's the usage, what's the revenue. And so I think we're going through this shift in the industry where the last 5 or honestly 10 years have been about if this technology will appear at all. In the last few years, it's been, well is it going to be this year or next year or in a few years. And I think now the question is, can you scale this and can you turn it into a business. And that's the question we expect to be asked, that's the thing we're focused on internally.

Got it. Joe?

Yes, I'd be curious, first, just on that question of who's in the lead today given all the different approaches people have. How much of that is a question of the hardware technology versus software technology versus the regulatory piece? I mean it seems like almost the regulatory piece at this point is the most challenging and we'll determine who gets to scale quickest. But which -- are all those things important? Is one of those kind of more important in your eyes?

So a good question. On the regulatory side, Joe, I disagree a little bit. I think the regulatory thing is not really the issue holding anyone back because there's a pathway to operate driverless cars in many states today in many cities. And so if you have the technology, you have the product, you have the operations all figured out, you'd be out there. So I don't see that as the critical bottleneck. When it comes to the hardware and software, what I really see today are 2 different approaches that are being played out by different companies, different initial approaches. And in the end, I think they converge. But I'll tell you what the 2 approaches are. At Cruise, what we do and maybe a handful of our competitors, as you start with well-equipped vehicles so that you know you have the sensing capability and the computability to enable driverless operation. And I think our approach has been let's well equip the vehicles early on, make the technology work, get it out there, operate it, make sure that customers like it, and we actually can validate all our assumptions that this is the correct hardware and then turn this into an engineering optimization problem, how do we take the cost out, how we take the power out so that we can increase the scale of the vehicles and improve the margins. It takes a science problem and turns it into a well-understood engineering problem of cost down and power down, those kind of things. The other approach I see, like some people in the AV industry, is to start with vehicles that are not well equipped and make a bet that you can write software that can somehow turn this less equipped vehicle into a driverless car. And at this phase, that is a science project. It's not proven, no one has done it. And a lot of people are betting that if they build these cars, bake in a very limited set of hardware, they will be able to invent their way into a driverless car. That carries a huge amount of risk. But if your primary goal is to sell cars and your secondary goal is to create robotaxis, that's maybe the approach you take. In our case, we are maniacally focused on making robotaxis work. And so we've taken the approach we did. And we're also very confident that now that where we are now in operating driverless cars that we can take the cost out and get this to a very competitive price point really soon.

And I'd be curious what that competitive price point means, because if you have a robotaxi that replaces 4 drivers a day, 24 hours a day, like you could spend $100,000 on the autonomous portion of the car, and it would be a very good investment. That's obviously not going to be in the GM Ultra Cruise type of model. So can you just talk a little bit about what that bill of materials is that you need to fit in?

Yes. I mean, again, so eventually, these strategies converge. And by that, I mean, we'll take so much cost out of our system that it looks at a similar price point to what the less-equipped vehicles have today. So maybe single-digit thousands in that range over time. And at that point, you're not -- you don't need to solely focus on these high-utilization businesses like ride share, like you said these can operate 20 hours a day, and therefore, you can afford to spend more on that asset upfront compared to personal car ownership where it sits idle 90% of the time. As we go further out in time and we drop the costs, the BoM cost of the technology, mainly the hardware for the AV system, it becomes more attractive for a personal car ownership model. And like I said, we're already working with GM on making that happen. And it's not with the hardware that you see on the vehicles today, the sensor's everywhere in the heavy compute systems. That was about -- those are really like our beta vehicles, making it well equipped, but not focusing on extremely low cost. The stuff we're working on with GM is 2 generations forward on our hardware that uses Cruise's custom silicon investments we've made to drive down the cost substantially and the power consumption, so it actually fits the profile of a personally owned vehicle.

Great. And I'd be curious what that sensor suite looks like. And again, I'm not trying to -- obviously, not going to tell me anything that would be competitive. But just is there a philosophy of you sort of start with solving division problems and then LiDAR is kind of a fail-safe to that? Or do you use all of the LiDAR, RADAR vision inputs as kind of one set of inputs is how would you qualitatively approach these problems?

Well, again, our approach is focused on what we know works and what we think is very, very likely to work over the next few years. And so while I've seen a lot of improvements in pure vision-based systems, but it seems like making the problem harder than it needs to be, like the kind of LiDARs we're working on and that are going to be available in a few years' time are basically the same cost as a camera. So it seems to me like it would be kind of silly not to use them at that point. And regardless of whether your vision system functions pretty well, one of the things we've seen great benefits with is having multiple sensors and using ML models, not heuristic based fusion, but ML models to fuse them. And these models kind of learn the different types of mistakes the different sensor modalities make and fuse them together to perform far greater than anyone could on their own. And that's especially true for LiDAR and camera and particularly in the short to medium range around the vehicle. If you think about it as a human driver, when things are really far away, like an oncoming car, you don't really need to know it's precise heading or how far away it is or how fast it's going. But you need to know that it's there and coming towards you. But as you get closer to the vehicle, the sensing requirements for precision really skyrocket because you really didn't know if that person 2 meters away from the AV is about to take a step out or are they going to stay where they are. And that's where you can really benefit a lot from the precision of a LiDAR-based system and even high-resolution radars to get really precise velocity estimations because in urban driving, it can be pretty chaotic and very fine movements and very subtle cues are necessary to have safe driverless performance.

Great. I just have a couple of more questions. I guess in terms of thinking about custom silicon, obviously, there are a lot of different approaches. There's people using programmable GPUs and FPGAs to implement these things. You have other car companies developing custom solutions. You have merchant guys that are trying to sort of start to focus on solving these problems. And I guess the challenge is you're designing custom silicon for 1 million cars. It's not a huge number relative to the upfront cost of developing an ASICs. How do you kind of play the trade-offs between thinking about those merchant market solutions versus what you're doing?

Well, from a cost perspective, the cost savings compared to what we'd have to do without custom silicon, it doesn't take 1 million cars to see a payback. It's much, much less than that in terms of ROI. But the big thing is, since we started with the strategy of building vehicles that are well equipped, we know the different sensing and compute components that are necessary to enable driverless operation. But those are assembled with a lot of off-the-shelf silicon and parts from other vendors, but you've got a crazy amount of sensor data flowing through the vehicle, so we're talking like high-speed networking. You've got the sensing components themselves, the preprocessing, the signal processors, the FPGAs, the ML accelerators. There's a lot of components that today have to be put together from a lot of discrete chips. And so the biggest benefit we're seeing is the integration of all of that into a system on chip. We don't need a 10x or 100x improvement in ML performance. We just need that to be consolidated to drive down the cost by not having as much third-party silicon. And then also, when we put all those things together in one chip, we can squeeze a little bit of extra performance out of it, because they're physically sitting on the same silicon with less need for interconnects and other things. And so for us, when we do the custom silicon, we save a lot of money, and that helps us get towards the very low BoM cost we want for retail vehicles in the future. We also simplified the architecture substantially. There are fewer components and chips everywhere, and we dropped the power consumption a lot because we're literally only putting the transistors on that chip that we need -- we know we need to support driverless operation.

That's a good segue to my last question, which was about power consumption. Is this going to be an internal combustion vehicle in 5 years? Or do you need to think about fitting this into a power consumption footprint of a hybrid or a battery powered vehicle?

Well, we think that we're building the cars of the future. This is the way we're going to get around in the future. We don't think combustion engines are part of that at all. And so all of our systems are optimized for electric vehicles. And that actually helps a lot when you think about the redundancy requirements and the power consumption for these systems in their early days. So yes, we're going to be EV for all these.

That's very helpful. Maybe if we have any questions from the audience?

In terms of getting Cruise systems into passenger cars, do you see that as being closer to a 3-year goal or a 10-year goal? And in terms of how that actually looks, say, for a GM car, are you going to try to completely displace the existing Super Cruise system? Are you going to combine aspects? What does that process actually look like?

So it's a good question. The timeline, I'm not going to give you any specific numbers there, mainly because we really want to make this ride share business work. We are working on personal autonomous vehicles with GM. And I have my own ambitions, but I'm not going to put out any numbers today there. As it comes to Ultra Cruise and Cruise, which unfortunate had naming collision there, I think that in the next few years, Ultra Cruise is going to be a really compelling product. And I think as consumers are exposed to things like Ultra Cruise or similar products, it's going to be a big factor in when they go out and buy a car. And so I think it's important for a company like GM to have that. And as the driverless technology from Cruise comes online, we're going to -- we'll have to learn what consumers prefer. Some people may not want that or may not be willing to pay for it. My own view is that based on what I've seen from early users of our technology, it is -- it just comes across as pure magic you get in one of these cars and it -- like you see the steering wheel move, it drives away, like that initial reaction is just disbelief and amazement. And I think when it clicks for people that they don't have to drive, like that's an option for the first time, I think you're going to see some really big changes in consumer behavior. And I do think that if we're talking about price points we are and relatively low BoM costs, in a few years' time, it will be hard to sell a car that can't drive itself.

I have a question. You mentioned it's less difficult to expand to different cities. What about different modalities driving like commercial vehicles?

Yes. It's a good question. They're actually similar problems. So today, we operate in San Francisco, limited geofence in Phoenix and a little bit in Michigan. So we've got -- we're able to see already that the technology we built in San Francisco to handle this really complex urban environment, big diversity of scenes, it scales well. It works out of the box, like we make 0 changes to the software that's running in Phoenix, as an example. That won't be the case for all cities. There will be some things that are unique in each city. And we have developed an approach to basically figure out what those differences are, make sure that the software is robust to those and validated to the extent that's necessary. But I'm sure we're going to learn a lot when we actually take this driverless operation we have in San Francisco and try to put in other cities. I do think the learning curve is steep. And after we go through the first set of cities, we'll find that now the software that has adapted to what we see in San Francisco and the first couple of cities, the incremental differences for city #4 and 5 will get smaller and smaller. And so the risk sort of actually declines after the first few cities. The other side of this vehicle modality is similar in that when we even go from the Bolt-based vehicles that we have on the road in San Francisco, to the Cruise Origin, which is an awesome vehicle for ride sharing, seats face each other, million-mile lifetime, we optimize for really low cost, the sensors are in different places. It's a different wheelbase, all those kind of things. And so in a similar way, we have to take this core technology, this core software, validate with different sensors, it still achieves the right performance and different physical footprint and other things, but it still has the same performance. So these are all flavors of the same problem. And since we have the Origin going into production early next year, we've been working in the background on all these problems because it's necessary to get the Origin, like even out of the gate and start using it at all. And that same approach that we've developed to adapt from the Bolt to the Origin is the same approach we're using to adapt from San Francisco to other cities.

Just one more, just clarifying your comments. It sounds like your vision on the passenger vehicle is a vehicle that's not one that can go between someone driving themselves and being driven. It sounds like it's pure full self-driving passenger vehicle, in the sense that what you envision the passenger vehicle being in however many years. Is that the right one?

Well, that I don't know. That I don't know. I think a lot of people do like driving some of the time. And there are certain use cases where our vehicles are level 4. We don't think there's much point in level 5 vehicles because you don't necessarily need a vehicle that can drive on an ice highway in Alaska or something. I think that rare edge case, maybe if you have like a ranch you want to drive on or you want to drive through a private property that's not, whatever it is, there may be some situations where people still want to drive. And I think ultimately, like what we want to do is have happy customers. And so that's a choice that we want to leave up to the customer.

But the system you're developing is not a Level 3, Level 2 plus type system?

Yes. Importantly, it's either going to drive itself or you're going to drive it, there won't be anything in between because quite frankly, at the point when the system is ready and safe enough to drive itself, having a person supervising it actually makes it worse than better, which is hard to comprehend when most of us have only been exposed to the stuff on the market today, but...

We have driven a little bit, it's not that hard to comprehend.

For the cars that operate here, we have plexiglass making it so that people can't interfere with the steering wheel because there's these moments where there's something about to happen and someone gets nervous, but the car sees that and is reacting simultaneously. And if you imagine you're -- there's something in front of you, the AV is going to avoid it by going left and you reach up and try to make it go right, you're going to hit that thing. And so that's why we -- we work to validate the system to make sure it's safe enough that it does not ever need your intervention. And then at that point, maybe a steering wheel disappears or whatever it is, so that there's no conflict between who's in charge of the vehicle.

Great. Well, I think we'll leave it there. Kyle, thank you so much for coming out here. It's exciting times, and thank you for coming.

That's really cool stuff. Thank you.