Great. Well, thank you, everybody. My name is Mark Delaney, and I cover Aurora for Goldman Sachs. Aurora, as many of you know, is a leader in AV technology with an initial focus on deploying in the Class 8 trucking market. I'm very pleased to have Chris Urmson, the co-founder and CEO of Aurora, with us today. Thanks again for joining us this year. Oh, really glad to be here. Thank you. I thought to kick things off, Chris, would love to better understand Aurora's view on developing AI technology. I think the company has a bit of a hybrid approach, where there are certain specific rules of the road that you hard code in, but also use a lot of AI and AI technology- Mm-hmm ... in order to develop your Aurora Driver. So maybe talk a little bit more around what you're doing in that area. Yeah, I think we were one of the first companies to recognize, certainly in the AV space, recognize the importance of this type of modern AI approaches, technologies like the Transformer, which have been key to the advancement of LLMs. That's been part of what we've baked into our approach. You know, you can look to a blog post we put out in 2017 talking about the importance of data, the importance of trainability in the system. But one of the things we realized was really important was making sure the system was verifiable and controllable, if you will. So, what do I mean by that? If you use an LLM today, I'm sure everyone here has used ChatGPT or whatever else, it's incredible how lifelike and realistic it is, and every once in a while, it has a what I technically call a brain fart, right? Something where it says just something completely outlandish. And it turns out if you have the equivalent of that on the road, it can lead to tragedy. And so for Aurora, our approach has been one where we recognize that risk, and we combined the best of these modern machine learning or AI techniques with guardrails that constrain the behavior and encode things that we really think are important, like stopping at stop signs. So it turns out that if you model driving behavior off what people do, there's this great study that shows that 11% of people actually stop at stop signs. And so if you just observe human driving and use that as the input to train the system, then you would rightfully expect that the truck is going to stop at stop signs 11% of the time, and we prefer that it actually obey the law and behave appropriately. And so that's a place where we can encode explicitly what we think should happen or needs to happen in a very obvious situation. And then we use the AI to actually generate the human-like and the much more dynamic kind of behavior. So when a vehicle cuts in front of you on the freeway, it's easy to tell your learning driver, "You should leave three chevrons of space between you and that vehicle that cut in front of you." But how quickly do you hit the brakes? Do you kinda let it coast through it a little bit because they're going way faster, or do you have to hammer the brakes aggressively because they're cutting in more slowly in front of you? There's not a really trivial way or a straightforward way to say, you know, to write an if-then-else statement for that. It just doesn't work. This is a place where we can gather data from both our expert drivers on the road and what we observe of other drivers on the road and use that to train, "Okay, here's the right behavior for those dynamic situations," and really understand naturalistic driving in that way. This combination of machine-learned, or AI, if you will, approaches with a verifiable set of guardrails we think is absolutely essential to actually delivering this product in a way that you can, you know, as a customer, you could trust it, as the general public, you'd be safe and comfortable with it on the road around you, and as a regulator, importantly, that you can understand it well enough that you'd be okay with it on the road. Super interesting, super important topic. It's influencing the way we think about solving the problem, and we've been fortunate to kind of catch that wave early and been riding it for the last chunk of time. That all makes a lot of sense. I think one of the arguments in favor of a different approach and fully end-to-end is that is what's required to get through some of these edge cases that have been- Mm-hmm ... a limiting factor on AV development. What’s your thought on those edge cases? Yeah, I think that's kind of nonsense. Right, so there's two parts to it. So one is that these models are fundamentally estimators or interpolators. What that means is that they take things that they've seen in the world, and they say, "Okay, this kinda looks like that. Let me do something similar to that," at least if they're implemented well. It turns out that you don't get a whole lot of experiences of some of these crazy things, like a person driving down the freeway at you on the wrong side of the road. You know, we've seen it, I think, twice. But that's nowhere near enough data to actually train the model. And so for some of those boundaries, you actually do want to encode explicitly the correct behavior when it's straightforward to do so. You can also generate an understanding of what the edge cases are. Like, people talk a lot about edge cases, but in practice, when you're driving a vehicle down the road, the thing you wanna do is get where you're going safely without bumping into stuff. And the variability is in kind of what the stuff looks like and exactly how it's going to move. And so we've been able to generate guardrails that keep us in the right place, and then learn the correct behavior around or within the constraints of those guardrails. That's all very interesting, and the purpose of doing all this R&D is commercial deployment. Yeah. The company has a target to launch commercially at the end of 2024. Maybe talk about how the company is tracking toward that and what the company still needs to complete in order to achieve that. Yeah, it's one of the things. I feel incredibly fortunate. I get to lead a company where we're going after the space that's a trillion-dollar market in the US on the back of two hundred billion miles traveled each year in trucks. And we have this incredible set of partners. We have this incredible technology, technological underpinning, and we're making incredible progress towards commercial launch. We're working hard towards that. We've talked about, you know, the measures of progress for that, which are our on-road performance indicator, API, and our Safety Case, Autonomy Readiness Measure. We, you know, feel excited about the progress we're making towards commercial launch at the end of the year. In your 2Q shareholder letter, the company highlighted the percentage of loads that achieved a 100% Autonomy Performance Indicator, and thus didn't need any- Yeah On-site support increased 80% in the month of June, up from 75% earlier that year. Maybe talk a little bit around how that metric is developing. Yeah. So first, just to decode it, what API means is, you know, what fraction of miles did a person have to touch the truck for us to be able to, you know, for it to drive safely, to its destination? And then, at some point, that got high enough that it was hard to understand what it meant, right? We're at 99-something%. And so we started to look at what really matters to customers, which is what fraction of trips would the truck have gone from A to B without having to have somebody touch it? And that's this 100% API number. And so, you know, that progress is continuing the way we would hope and expect towards getting to this, a high enough level that customers could actually benefit from this technology on the road. It ultimately won't get to a 100% because things happen, tires blow out, engines break, et cetera. So at some point, folks will have to go and touch these trucks to rescue them, but we're making strong progress there as well. Yeah, and I think it's been a little bit nonlinear, right? Yeah. There's a software release, and then there's maybe a step function change in. Yeah - performance. So one of the things that you, it's important to understand is that today, we actually operate our trucks a bit differently than they will when they're actually in production. And that's that we're almost on a daily basis, pushing new software through to the trucks that is passing a stringent set of tests, but not all possible tests. And so that means that we see some temporary regressions while we're maximizing development speed. As we get to launch, what we'll move to is actually released software, where we're stabilizing that software and kind of getting the performance up on these kind of nuisance issues. And so, you know, yes, you see this kind of nonlinear performance. Then we expect that as we grind those, we have the experience of every time we take a release version and kind of deal with the minor nuisance issues, which is a very common process in any software development and deployment. We get really good results out of it. You mentioned this safety a few times already in our- Yeah Conversation. In July, Aurora announced the results of an evaluation by a third-party organization. Yeah that audited your safety approach. Tell us what they examined, what did they conclude? Yeah - and any areas for improvement they suggested? Yeah. So this organization is called TÜV SÜD. It's one of the big European safety auditors. So if you're in Europe and you want to ship a car, these folks come in and audit you. We had them come and audit our, what we call our safety management system, and this is really how we think about, monitoring and evaluating risk at the company, so particular safety risk, and it was awesome, right? They're very professional, they're very well-respected. They came in and looked at how we perform relative to the AVSC standard for this, and it turns out we passed with flying colors. You know, I've talked with auditors for a variety of reasons. These folks came in, they spent about... They want to check with the CEO to make sure he understands what's going on and answers questions in rational ways. They got partway through that, and they said to me, "Hey, okay, we're good. I just want to let you know, like, we just call up your employees randomly and ask them about this," and they, they were able to kind of quote it directly, right? They understood how to answer the question exactly. They. It It was clear to the auditor that our company really didn't just have safety kind of stamped on the outside, that it was actually core to the way we operate the business. And the reason why that's important to me, and it should be important to you, is that at some point, I'm going to be looking at a set of reports that say, "Is this ready to go?" And the team's gonna tell me, you know, their thoughts on it. I want to be able to trust those reports and be able to say, "Yep, we're good to go." And this audit gives me confidence that we have the tools in place where folks have the opportunity to raise risks, that they feel confident that they can, and that those risks are adjudicated appropriately, so that we're not gonna have some kind of lurking safety concern behind. And that means that when the product does get to the road, we're gonna in 2027. That hardware will ultimately get line side installed on our OEM partner's truck. So, you know, as a customer, you'll call up PACCAR, you'll say: "I want a Peterbilt 579. I want it to roll off the line with the Aurora Driver on it." They'll put it on, and you'll get it from the factory that way. With both what we do today and with the hardware that'll come from Fabrinet, the OEM will create a truck that is ready to accept the Aurora Driver hardware. Then we will upfit it at our shop, taking both first with our initial version, so it'll commercial launch that we're working to commercial launch with initially, at our shop, and then the hardware that comes back from Fabrinet, we'll upfit that for our customers. Maybe you can talk a little bit more on Remote Assistance? Yeah. -and help us understand what that even is, but also the ratio of people per truck doing remote assistance. Yeah, so the world is complicated, and the way we've approached driving is that the truck itself is responsible for keeping itself and others around it safe on the road, but if it gets into a situation where either it needs some help addressing some internal limitation or it encounters a situation that it doesn't understand, it'll take a safe action, and then it'll call base and ask for help, and you can think of this not as someone with a Logitech steering wheel behind a desk that's gonna jump in and drive the truck down the road, but much more like a trucker on the road who's gonna call a dispatcher and say, "Hey, this is happening. What should I do?", and so that's the role that our remote assist operators play in our vehicles. So they don't drive the trucks, but they support the trucks with complicated decisions, where it may not be obvious on board what to do. You can think of it as kind of, abstracting intelligence, right? So these are problems that are either we don't understand how to solve algorithmically, or algorithmically would be too expensive to run on the hardware on the truck, and so we outsource that to a person off-board. When we launch our product, we expect that to be one operator, one remote assistance person, to a small number or a few trucks, but then we expect to get that to many trucks, and apologize for the imprecision, but, you know, that's the way we're talking about it right now. We shared at our Analyst Day a curve that shows kind of the knee in the curve, that once you get to one to ten, you've got most of the savings, and as you get further down, you know, we can continue to drive cost out of the system. So we're excited about the progress there. And then maybe you can help us better understand if Aurora needs to add incremental infrastructure in terms of people or services? Yeah ... in order to enable commercial launch. No, not to enable commercial launch, and we've really intentionally built our business to focus on what we can do best in the world, and that is delivering the driving capability, and that's why you see us partnering with companies like PACCAR and Volvo. We don't want to build trucks. That's a really hard business. These are companies who've built, you know, decades of experience and know how to do that well. You know, we don't want to manufacture things at scale because that's hard and complicated, and the skills people have, and that's why we work with Fabrinet and Continental. You know, we don't want to become FedEx, Werner, Schneider. Again, complicated, difficult business with a hard-won experience. Let us amplify each of our partners' skills and help create value for our customers. And so we've tried to keep a model where we'll be able to kind of manage, you know, our internal expenses and end up with this really nice, contained business going forward. Maybe talk about how much investment may be needed in order to add new lanes beyond the initial- Yeah ... Dallas to Houston, like? Yeah. Yeah, so as we look at expanding from our launch lane, Dallas to Houston, there's really two types of work we need to do. One is we need to map the area, and this turns out to be not very expensive. We drive the trucks down it a couple of times. We have automated tools that then generate maps, and then we have the QA team go through and make sure that map meets our quality standards. So... And what's great is that then becomes a competitive advantage for us because we can maintain it from the vehicles operating on it, and then we have a model. Think of it, like, just I guess intuitively, when you drive your car in your neighborhood, I would bet that you feel more comfortable than when you drive in some random place elsewhere in the world, even if you're a very good driver. Right? I can tell you that when I drive around the Bay Area, I'm a much safer driver than when I drive in Dallas or Houston. I have to stay clear because I just don't understand the road network. I don't understand where to expect traffic. The map gives that benefit to our truck, and it leans into what computers do best, which is recall information, and it does that with very little cost to us. And it becomes, like I said, this perpetuating advantage we will have. The other thing is that we need to develop any new skills that are required for that lane. And so the next lane that we expect to expand onto is Fort Worth to El Paso. You know, I've kind of talked about that. There's, you know, most of freeway driving looks the same. Freeway going Dallas to Houston looks mostly the same as the freeway going Fort Worth to El Paso. There's a big hill. That means that we need to use engine braking and not just foundation brakes. We do that today, we just need to validate it. We need to go through a Customs Border Patrol station. We're super proud of the fact that actually we're the first company that's built a process to do that in autonomy with the U.S. Customs and Border Protection, and we're now working with them to expand that to other places, and so we'll need to validate that. At that point, we basically can drive Fort Worth to El Paso, and so as we open new lanes, there'll be small incremental features we have to add, potentially, that will go less and less the more places we operate, so really excited about that and expect to be able to grow the business relatively quickly, as we get to that, that time. Have you shared any time frames around when to think about that El Paso leg and sort of the time to go? Yeah. Yeah, incremental lanes beyond that? I don't know that we have. I think we've guided to kind of next year, we expect to begin operating Fort Worth, El Paso, and begin to pilot. I think we just announced Phoenix, you know, El Paso to Phoenix as well. Yep, that makes sense. The company announced the Premier Autonomy program with Uber Freight- Yeah ... last quarter. Can you talk about what that program is and- Yeah - The business and financial implications for Aurora? Yeah, this is a first of a kind partnership with Uber Freight. It, for customers, the big benefit is that if you're a small or medium-sized customer, it's gonna be hard for us to reach out to you, and engage with you, whereas Uber already has that relationship. For us, the benefit is that, we don't have to do that outreach, and we get connected with qualified customers that will want to use, the Aurora Driver. And so you can think of it as a, a really high-quality lead generation, arm for us, without us having to grow a massive sales force. So we're really proud and excited for that. And then for Uber Freight, it creates a unique advantage for them in that they can now unlock for their customers access to a technology where we don't see a whole lot of competition in the landscape, and that's pretty valuable. So for all parties, we see this as a huge opportunity and win. Longer term, Aurora has talked about moving to a more asset-light and revenue-per-mile business model. Yeah ... that's in partnership with Continental. You spoke a little bit about it already- Yeah - but maybe expand on that, if you could, and- Absolutely Any updates on how it's progressing? Yeah, kind of in that spirit of let's do the things we're best in world at, and let's not spend money on things that we aren't, you know, we want to be an asset-light business. And so the one of the big pillars of that is working with Continental, and we've created, again, a first of its kind type of partnership with them, where they're providing hardware as a service. So what does that mean? That means today, hardware, Continental is spending upward over $300 million co-developing with us a generation of automotive hardened Aurora Driver hardware kit that customers will pay for through us, and we will pay to Continental on a per mile basis. Think of it as cents per mile for that hardware. This allows Continental to get to a world where they are operating in value-based pricing rather than should-cost pricing, which has been kind of a goal for tier ones for decades. It allows our customers to get access to the truck and the Aurora Driver without having to pay, you know, a prohibitively high upfront cost for that or without increasing or changing the way they structure their business. Right, they're used to buying a truck and paying a driver. In this world, they'll be able to buy a truck and pay a driver, it just happens to be us. And then for us, we don't have to sink a whole lot of capital into owning and, you know, amortizing these assets. We're able to pay that through Continental on this per mile basis. It also has the benefit of aligning incentives all the way through our stack. The more our customers drive, the more money they make. The more we drive for our customers, the more money we make, and the more the hardware drives for us, the more money Continental makes, so that's, you know, it's rare a time in life where everybody's incentives are fully aligned, and we've been able to set that up here, and as I said earlier, that Continental hardware, one of the benefits of working with them, is they know how to be in an OEM supply chain, so the hardware can ship from their docks to our OEM partner's docks. They know how to be integrated into that. They know how to put... Like, silly things, like putting it in the right kind of crate, right, and be able to track the inventory through it. All stuff that we no longer have to learn because we have amazing partners who know how to do that, and it's those kind of operational things that get in the way of execution that we just aren't going to trip us up because of this relationship we have. Yeah, very exciting. Maybe- Sorry, maybe it didn't sound exciting, but it is really freaking cool. Yeah. No, no, it is. Yeah, on some financial topics, the company commented it had contracted a large portion of its 2025 volume- Yeah ... and was in the process of finalizing the rest. What does this mean for how much revenue investors should expect next year? Yeah, so we have not provided, and I'm not gonna be breaking news here about how much revenue to expect next year, sorry, but it is exciting that we have customers that are already willing to commit with us to operating trucks driverless on the road. That's one of the critical enablers to actually having a business, is having customers who want your product, and so that's very exciting for us. We will definitely take a crawl, walk, run approach to this, right? We want to make sure that as this kind of groundbreaking, kind of technology comes to market, that people can come along with us on that, and that we don't kind of create problems for ourselves by moving too quickly. We're gonna move as rapidly as we can while bringing folks along with us. Tying the technology with the financials, Aurora has talked about a 50% reduction in cost per mile- Yeah ... between 2024 and 2025 with its next generation hardware kit. Can you talk a bit more on this? Yeah, sure. So if you think about the costs that drive our business, there's really three core elements to that. So one is that remote support that we just talked about a few minutes ago. The second is when the truck breaks down, having somebody go out and service it. And then the third is the depreciation of the hardware on the vehicle. And so today we have, as I mentioned, hardware that we build in-house. It's kind of expensive. As we move to working with Fabrinet, we see dramatic reductions in cost. So things like the computer that drives the system is gonna be about 40% lower weight, 40% lower cost to deliver. Things like First Light, our proprietary lidar, that's the special technology that allows us to see a long way down the road. We see another significant cost reduction in that as well, and then we see the lifetime of the hardware increasing. So that reliability and robustness reduces how often you have to change parts out, which reduces the operating cost of it. So we see a significant step change in the cost there, and then a significant additional step change when we move from the Fabrinet-produced parts to the Continental-produced hardware kit. The company is targeting a positive gross profit in 2026. Yeah. What would have to happen for this to occur? Yeah, the things that we have in plan already, right? That we look out at the world and we see, okay, we need to take the hardware that we're already bringing up from Fabrinet, and we need to get that onto our fleet and operating. We see that as kind of in plan and moving forward. We need to move that few to a few more on the remote assistance so that we're kinda safely through that curve, through the knee of that curve. And then we need to be driving up the operating life through this reliability program that we're putting in place with the hardware we're developing for Fabrinet. So, we look at that and see some elbow grease that's needed, but it's a pretty straight shot. That's great. Aurora recently raised a little over $480 million of incremental capital. Yeah. With this raise, how much additional capital- Yeah ... does Aurora still think it needs in order to reach free cash flow positive? Yeah. So when we did our analyst day earlier this year, we said we thought we needed, we estimated $850 million. If you take out the fees to our bankers, I guess you're not one of them, our bankers. You know, we raised about half of that, so you can do the math. Yep. Okay, very helpful. Maybe we talk on the competitive landscape and how that's changed, because I think a couple- Yeah ... years ago when you and I were having one of these fireside chats, there was a number of competitors. Yeah. Some of them have actually now exited. So where does the competitive landscape stand at this point? Yeah, from my perspective, it's awesome. Right, that as we look at the landscape, most of the folks have left. Right, so a few years ago, you would have looked out and said, "Well, there's Embark, there's TuSimple, and there's Waymo." Two of them don't exist in the U.S. market, and the third is focusing on robotaxis instead of trucking. So as we look at this landscape, we see none of our competitors have the OEM partnerships, the tier one partnerships, the scale that's necessary, the capital to really deliver this. And so we feel like we're, you know, in a place where this is our market to go win. And again, I'm just incredibly fortunate to be in that position with the team we have. I think you and I have also spoken, or maybe it was one of your earnings calls, that this is a market that it's very large, and presumably- Yeah ... there's room for multiple providers as well. Yeah, I think that's right. And I think ultimately, there will be multiple providers in the market, right? You know, when you talk about a trillion-dollar market, no one gets to have that to themselves. But we're excited to have what we see as a pretty good lead and a pretty good, strong position to begin from. And then we'll just keep executing and, you know, we expect this will create a lot of value for our customers, reducing both their costs and improving their revenue. And so we think that's a pretty good recipe for building a business that wins. Maybe talk on the regulatory environment and whether or not you expect any significant regulations to be put in place. At a federal level, I'm pretty convinced there's nothing gonna happen over this year, just given where we are politically. But I think that the good news is, this is one of those questions I think is somewhat misunderstood in the investment community. That in the vast majority of the United States today, if we are confident in the safety of the truck, we can put it on the road and operate it. I think twenty-five states is explicit, fifteen states is implicit, and that includes our launch markets and much of the southern freight belt that we intend to operate on initially. So, you know, from our perspective, there's always opportunity to improve regulation, but there's no big movement that's needed. We continue to work with folks like California, where light vehicles are allowed and heavy vehicles are not currently allowed, but we saw a really promising movement last week, I think it was, or maybe the week before, when the California DMV released their early draft regulations for heavy vehicles, so we're really excited about the enthusiasm we're seeing from the regulatory community and the lawmaking community, and I think a lot of that is due to hard work and education that we and others in the community have been doing. Maybe talk about how you respond with some of those local regulators, permitting officials, to the extent there is an accident, that- Yeah ... unfortunately, maybe Aurora Driver is at fault in one of these situations. What steps do you guys have to take so you- Yeah ... don't run the risk of a change in the political environment, losing permits or- Yeah ... or things like that? So we have not been. We've been involved in a number of collisions, unfortunately. And when we look at the tape of those, the Aurora Driver was doing the things that you would expect it to, and the other driver was not. You know, we've had situations where we had a very gentle collision, where somebody in a box truck driving along the side of us and just touches the trailer. Turns out in that situation, we were able to see that they were looking down at their phone, and just not paying attention while driving. We had another situation where somebody came screaming up behind us on a wet road, tried to make a swerve around the truck and hydroplaned and tapped the back of it. We see these things on the road. It's one of the benefits that we think is gonna be really valuable from having things like the Aurora Driver out in the world, is that we can improve safety. There's about 6,000 fatalities related to truck accidents every year, and we can drive that down. As it turns to your question directly, we both believe it's very important to build relationships with the parties that are on the regulatory side of this, and so we spend effort making sure they understand our intentions, they understand how the technology works, they understand the risks and opportunities, so they can make informed decisions, right? Their job is to keep the public safe, and we see that as one of the opportunities and part of our mission as well. So that aligns really well. We're proactive in communication, so when we have an event, we make sure that whether it's our customer or whether it's a, you know, regulatory agent, entity, that they know what's happened and we can give them visibility into it. We also internally practice this, right? So we, on a regular basis, turn the crank on mock incidents so that we know what will happen when a real event occurs. And we think that's important, right? You don't want the first time you deal with this to be in the heat of battle. You want to have turned the crank, practiced it, you know, much like a football team doesn't show up on, you know, first day of the regular season and go out there and hope all the blocking schemes work, right? You actually set this up ahead of time so that when the real event occurs, you're ready to go. That makes a lot of sense. So the company's focus today is on the Class 8 trucking market- Yeah. - but you've spoken about ambition to do robotaxis in the longer term. Help us better understand what type of timeframe we can? Yeah ... think about for that and how similar the technologies are. I, I think this is one of the things that's just incredibly cool about what we're building, right? When we get to the point where we have trucks operating on the road, doing that at scale, in a sustainable business, that's incredibly powerful. The tools we will have built internally, the intelligence that we've built for that truck, is gonna be broadly applicable. The obvious next steps are in things like robotaxi, things like local goods delivery. More broadly, over time, this kind of general environmental understanding and intelligence is gonna become an incredibly interesting asset for us. Now, we need to keep our eye on the ball for the next few years to make sure that we get this product out the door, and focus on trucking, because we think that's an incredible opportunity, strong unit economics, huge market scale, all of the things we talk about. But from there, the opportunity for this business to go and have a broad impact on the world is amazing. Bringing it back to a more personal note for all of those in the audience who are listening in on the webcast, we'd love to get your views on when you think consumers will be able to get a personal Level 4-capable vehicle? Yeah, I don't know. I don't see it happening soon. I think the opportunity to use these vehicles as parts of fleets is incredibly powerful. I'm really excited about what the Waymo folks are doing in robotaxi and seeing that, you know, in San Francisco here. Like, I drove in this morning, and they're just all over the place. It's just awesome. So I think that that is actually the right way to introduce this technology to consumers for a variety of reasons, and so I'm excited to see that. I don't have a lot of confidence that there's anything happening in the near term, where someone's gonna truly own a real Level four system. Great. Lots of exciting stuff the company is working on. Hopefully, we can get an update again next year. Thanks for joining us. Yeah. Thanks so much for having me. Look forward to it. Thank you.