Aurora Innovation, Inc. (AUR) Earnings Call Transcript & Summary

Okay. Great. Thank you, everybody, for joining. My name is Mark Delaney, and I cover Aurora for Goldman Sachs. I'm very pleased to have with us yet again this year Chris Urmson, who is the Co-Founder and CEO of Aurora. Thanks for coming.

Glad to be here. Glad to be here. Thanks for having me.

Well, let's jump right into it. I mean a lot of interesting stuff that Aurora has been working on, several exciting announcements on your journey and now launch of autonomous trucking, so really exciting stuff. And maybe we can start right there because the company announced actually just this July that you started to driverless operations without an observer. Maybe you can talk more around how that launch has gone. Any key learnings so far?

Yes. Now that we've woken everybody up, so glad to have everyone here. Yes, it's gone kind of incredibly, right? So it's been a long journey. A year ago, we proved that the technology worked with our launch then. This year, we've been able to launch our second-generation hardware. AV is really hard, let me remind you. And this is really a platform that will allow us to scale up to 1,500 units and then put us in a position to go scale with our third-generation hardware that comes online at the end of next year. So it's been great to see it on the road. It's been another proof point that's enabled customer enthusiasm and excitement and kind of help move those conversations forward. And it's really just -- it's kind of the starting line for us building and scaling the business.

And any surprises or things you didn't expect to have that you've learned just since you've taken the observer out?

It feels good, right? It's a lot of work for the team. It really -- yes, we launched last year, began operating drivers, but we have the kind of the optics challenges. There was a person who sat there is this observer. And we knew the bar we were holding it to. We knew that it was safe and operating. But it's very different for the outside world to have that visceral image of, no, there's nobody there. The Aurora driver is fully operating this. So I think that's been very powerful. And even the most cynical observers can look at that and say, "Oh, this is real."

No, very exciting stuff. In order to get those vehicles out, you up-fitted a truck. You're using Rauch as a partner to do to do some of that upfitting as well as some in-house up-fitting capabilities. Maybe just talk about some of the costs and implications of doing the upfit model here in the -- at least the near term?

So our second generation hardware kit is a major step function both in performance but also in cost reduction. That first generation was really to prove the technology work. The second generation was about actually having something that was commercially viable in market that we can ultimately be profitable with. And we're seeing that in practice. Rauch is great. If you heard me talk about our strategy in the past, it's really been one of let us focus on what we do best in world and then go work with folks who are best in world on what they do. And that's what Rauch brings to the table. And they are a very well-known quantity in the automotive space, where if you're an OEM and you need something modified and scaled up or repaired, they're the folks who do it. If you buy an F-150 Raptor, Rauch made it a Raptor instead of an F150. So they know how to do volume. They're working with us now on mass production of this or large, I guess, modest scale in OEMs terms for automated vehicles, large-scale production. And we're off the races. We're in that period of kind of production hell, I guess, some people might call it as we kind of get the ramp up, but we're expecting 20 to 25 trucks on the road at the end of this month and we'll be building 20 trucks a week in the not distant future here.

That's great. Well, the trucks need to drive somewhere and one of the key initiatives of the company has been expanding the operating design demand in terms of where you can drive, weather conditions. One of the interesting things I thought Aurora has been focused on has been driving right to customer sites, not just on highway. Tell us a little bit more around how you're doing with that direct-to-customer endpoint part of the ODD.

Yes. And this is just one of the things that just makes sense, right? Customers want their stuff to go from A to B, not from some other C to D. And so this has been core to how we've been thinking about the business for a very long time, and we're seeing this in action. We've been operating from customer endpoints for some time now. We're now with our customer Demar operating from sand mine to the distribution point, just running that many times a day in and out, and we're continuing to operate more places for customers as well. So it's operating exactly the way we expected. It's important to remember that we've been driving between our terminals off-highway to these places, right? Our location in Houston is a 5-mile drive from the freeway to that terminal. And so it's not really surprising that we drive to these points for customers as well.

So you're already doing this today, not just for the off-highway, but even sort of on public road, you're doing that direct...

Yes. When I say -- sorry, when I meant off-highway, I meant on public roads. Like we -- it turns out that they don't just put the sand mine connected to the freeway, right? You have to drive through the town and get to the place. And similarly, when we talk with our customers in Phoenix, our customers in Dallas, right, these are places where we're going to endpoint locations for them today.

Okay. That makes sense. Well, another aspect of expanding the ODD is about lanes. And you've got a number of different routes you've been adding on. Talk a little bit more around where the company is expanding to.

Yes. So we're really going where our customers need is today. We expect to have 200 trucks on the road by the end of the year, and it turns out Texas is big, and there's a lot of places for those trucks to go right there. At the same time, we are accelerating our ability to open new lanes. And to open new lanes for us, it's really 2 parts. So one is putting in place the map. So the truck can have this kind of anticipation of what's about to happen down the road and not just what it can see over its horizon. And then the other part, of course, is the competence to drive across the different features that are in the world out there. On the mapping side, we've talked for a long time about our Atlas technology, this ability to share the map and build it efficiently. That just continues to get faster and faster. And the first lane, it took us, what, 6 years or something to open. The second lane took us 6 months, the third lane 6 weeks, and we continue to see that accelerate. And then on the capability of drive every time we open a new lane, we go through the place that we're going to operate and ask is there something novel here. And any time we find something novel, we'll go and build tests. Those test tells can the Aurora driver operate through that or not. If the answer comes back that we're not confident in the performance of it, then we added a little backlog under the engineering cycle, and then we kick it off. But every time we solve one of those things that's novel, that is not just solving it for that GPS coordinate here, right? It's solving it for every other instance of that thing that was novel there. So if, for example, it was a 3-way stop, was the thing that we couldn't handle or we hadn't got the test in place for, once we've tested 3-way stops, it's now, okay, every other 3-way stop that we would encounter is in scope and able to drive. And so you can see how very quickly with not many places that you drive that are useful, you've accumulated the skills that are necessary to really generalize and operate broadly across the sub-elements. We're very excited about seeing.

You do some of those test and validation points. How much of that is something that needs to be done with real world test and validation versus what do you think you can do with simulation? And can simulation replace a lot or all of some of those types of things?

Yes. So simulation is a very broad set of technologies, everything from replaying data and kind of a living what the actions are to generating the full synthetic perception system and everything in between. And we use a combination of all of those tools to do this validation. So this is where we're seeing some real lift from generative techniques, so we can take a real world scene and augment that in certain interesting ways. We have a process by which we will go ground that simulation with reality to have conviction that we're not just playing a video game. We're actually operating in a world, a simulated world that's reflective of reality. And then we can use that to get to the point where we're now confident that this thing is actually working and go release the 70,000 pound truck down the road. One of the things that we do not believe in is kind of one-shot YOLO, right? Like, oh, we saw it work once, cool. Let's let it go, right? That's not how you deliver a safe real product in the world. That's how you cause real damage.

Dovetails well with my next question, which is on your approach to AI more generally and obviously guided by safety, but you've written a lot and talked a lot about verifiable AI. Tell us more about what that means for Aurora.

Yes. Verifiable AI means making sure we are using the most cutting-edge techniques but ensuring that we have the tools and the architecture so that we can have conviction that they're actually going to behave the way we want, right? We -- again, when you're driving a 70,000 pound thing down the road at 70 miles an hour, it's not okay to have an oopsie moment. And if you're really just going to throw data in one end, train a model and kick it out on the road and talk about the magic of AI, you're destined to have these problems, right? And we see it constantly in the entertaining output we get from LMMs, a massive amount of value and opportunities there, but they brain fart, for lack of a better word for it, pretty regularly, and we just can't afford that. And so for us, verifiable AI is this combination of how do we take these modern techniques use them but put them in sufficient box so we can actually test and validate them and that we can put constraints on how they operate such that we know that the behavior is going to be good.

Said another way, you're not an end-to-end AI proponent.

No. I think that, again, the tools are super useful, but a black box between sensors and pedals where you just hope that the outcome is always going to be right is a recipe for disaster.

I know you and I have spoken about how important this might be for regulations and talking to regulators because you can point to specific things and why your software made the choices it did. I know you're engaged right now, I think, with some regulators, including in California. Maybe talk around some of the key regulatory things that Aurora is working through.

Yes. So we continue to be really happy with the regulatory environment in the United States. It's a giant market, of course, and it's an important market for us and continue to see really positive support, both at the state and federal level. Recently, California put in place regulations that allow automated trucks later. This has been a many year journey to get there. And so we're excited to see those. We've already received our testing permit. We expect it will take some time to move through that regulatory process. I think the DOT in California has been appropriately careful, and so we'll just continue to work through that work with them to get this technology unlocked in the State of California here. And then at the federal level, again, we're seeing tremendous support and traction in the surface reauthorization bill this year, this explicit engagement around automated vehicles and automated trucking. And so we're seeing really positive traction there. And the Secretary of Transportation, Secretary Duffy has really taken a step forward with this innovation agenda, ensuring we maintain safety but continue to allow America innovate in this space.

And I think one of the other angles there was around you potentially needing to warn other cars, if there is a breakdown and the trucks on the side of the road. I mean this is a long-standing regulation, but the existing ones, like a person has to walk and put these cones. Where are you with the warning lights and more of a digital solution?

Yes. So back in the day, somebody thought, hey, it'd be good if people put triangles or flares behind cars. And there is no data to support it, but it seems like not a bad idea. And that got turned into regulation. And so obviously, when we don't have a person on the truck, we could find ways to do this. We can imagine a little dropper on the back of the truck that drops triangles or flares. But we prefer to do something more rational, and that is turn on blinking lights. This is what police vehicles do. This is what other emergency responds to you. This is what construction vehicles do. This is what to trucks do to indicate there's something here. Don't hit me. So we made the deep, bold, innovative step of saying, let's do that for trucks. And we've used it, we've collected the data now to show that it's actually as effective, if not more effective, than dropping cones. And so we have an exemption from the DOT to move forward with this. And we look forward to having a more permanent kind of common sense solution for this.

Very helpful. I want to talk on the ramp. You already said earlier, on track for the 20 to 25 trucks by the end of this quarter and 200 by end of the year. But you also mentioned production is difficult early stage a lot of compassion -- maybe just talk about how you're working with supply chain. I mean actually, the team at this conference has been -- certain components are pretty tight right now. So anything you'd point to as you're working through that ramp?

Yes. I think this is one of the things where we have a really strong advantage relative competitors in the space, right, is the part that gets the headlines is the very cool and technically complicated and kind of sexy AI driving stuff. And we are very good at that. We're best in world with that. But for that to actually go out into the world and do something, you need the physical parts to put it on the truck. And this is where our first generation hardware was stuff we made in-house. We knew how to test it. We were confident in the safety of it, but we could produce that at the scale of tens of things. The Generation 2 hardware that we just launched, we worked with Fabrinet, which is a global contract manufacturer, that we can produce the scale of 1,500 units. We talked about Rauch earlier in the supply chain. We have to upfit those trucks to add by wire capability and add our vehicles -- our road driver kit onto that. That allows us to get to a scale that's unprecedented in automated trucking and will allow us to bring customers along to kind of allow them to experience it to start to build this into fleets and actually really gain significant traction and ultimately generate meaningful revenue. And then our third generation hardware, we're codeveloping with AUMOVIO, and that is produced at scales of tens of thousands of units. And that is where we get to true scale and true inflection in the business. And that path that we've been on of simultaneously developing the first, second and third-generation hardware has been critical to enable us to go scale. And so for us, this is gigantic, right? We think we have a multiyear lead on kind of the hard tech AI driving capability. We think we have a perhaps even larger multiyear lead on the supply chain and hardware side of this, where if you do not have a path to produce the hardware, I don't even know what you're doing. And we're clearly well ahead on that. So super excited there.

Very helpful. And as you kind of exit this year with the 200 trucks, you ramp into 2027, as you do ramp in '27, maybe talk about the business model in terms of do you haul the cargo as well and as Transportation-as-a-Service? Or do you do more just the virtual driver and is Driver as a Service next year?

Yes. So it's going to be very exciting, right? Today, we're operating a handful of trucks. We will be operating 20 to 25 trucks driverlessly by the end of the month. And then we will be adding 20 trucks a week as we get into Q4. Just the scale of that growth in this space is, again, it's super exciting. It is turning a corner. Today, we operate in what we call Transportation as a Service, which means we own and operate the trucks. We operate them for carriers. So we don't go out and bid with shippers, right? That's not our business. We are not a trucking company. We operate under contract for companies like Werner Schneider, Hirschbach, Federal Express, Demar and several others, where we're providing capability for them. This allows them to experience the benefits of automated driving to see these trucks in action and get a way to have a relatively low-risk opportunity to learn about the technology and then commit to it. As we go into 2027, we will transition to Driver as a Service. And this is where customers go from having us own and operate the asset to leaning into their core strengths, being owner-operator to maintain those assets and pay us a subscription fee for driving it. And this is the long-term model we've been talking about for the last several years. We want to be asset light. We want to focus on what we do best at world, which is build that driving capability and allow our customers to excel and benefit from that technology but run their business in the way they're used to. And so we're -- we had a very exciting announcement earlier this year with Hirschbach, where we signed an MOU with them for 500 tractors, where we will provide the driver service. They will own and operate the tractor. And we continue to have further conversation with other customers for that all even '27 and beyond.

And from the conversations you're having or the MAUs you've been signed, is your sense that customers are willing to do the driver as a service model on upfitted trucks?

Yes. And they want this model, right? If you are a major carrier in this country, you see your ability to own, operate and maintain that asset as a competitive advantage. That is what you do best in world along with the customer service and the rest of the operational rigor that you put around it. And so for us, being able to slot into the business that they already have, their model today is they buy the asset. They pay someone to drive it for them and they do the rest of the work around that. That's what our business model will look like to them, and that's exciting and compelling. And yes, folks are interested in these Roush upfit international trucks.