Buddy, we have Kodiak AI with us. The CEO, Don Burnette, is up here with us. We did a future mobility panel earlier today, but we're gonna dig a little bit deeper into the Kodiak story specifically. So thanks for joining us. He's the founder and CEO, and I'll let you maybe just give us a few minute overview of Kodiak.
Sure, yeah. It's, it's great to be here. Thanks for, thanks for coming. I've been working in the self-driving space for over 17 years. It's been a long journey, and Kodiak has been around for almost the last 8 years. We are a self-driving company. We're working on an AI solution to driving vehicles, and we're focused on the commercial market. So we're in three different verticals. We play in the self-driving trucking space for long-haul deliveries.
We work on AI technology for driving in industrial applications, so think about off-road or unstructured environments, and that's where we actually have a driverless deployment today. And then we also supply self-driving technology for military and defense applications. We, we have, as of Q3, which was our latest report, 10 driverless trucks that are out there in the hands of customers operating today.
That means nobody in the cab, no required remote monitor. So these trucks are actually out there providing value. They are owned and operated by the customer, so that's something that we can definitely dig into a little bit here. They work around the clock, so they're driving 24/7, day and night, rain or shine. A lot of dust storms in Texas, where they're based. We're looking forward to expanding that deployment and getting to a driverless over-the-road solution in the second half of 2026.
Great. When we think about what we're seeing on the passenger side, right? We think Tesla, we think cameras. Your system has a lot more robustness to it. But you mentioned, you know, Texas, right? And you're working in the Permian Basin. Anybody here who's seen Landman, there's a lot of s and and dust and a lot of other things going on. But just tell me about, like, sort of the importance of the redundancies in the system and how that applies to an environment which is probably suboptimal in the Permian.
That's right. So the more challenging environment, the more diverse environment you're gonna operate in, the more it helps to have multiple modalities of sensing and perception capability. There's a big conversation in the market right now between LiDAR versus camera-only solutions. And Kodiak is squarely of the mindset that adding more sensing capability to the solution, if the value you're producing can afford the extra cost, increases the safety of the overall system.
And on the commercial side, we're not as cost-sensitive, obviously, as a customer-facing technology, and so we've chosen to use radar, camera, and LiDAR in our solution. I think that having multiple modalities of sensing increases safety, especially today in the early stages of this deployment, safety and redundancy are ultimately what we want to maximize. It's not just redundancy in the sensing, by the way, it's throughout the entire system.
So we have redundancy in sensors, we have redundancy in our computers, we have redundancy in the platform itself, like steering, braking, power, those types of critical aspects of the system that if you lost any one of them, you suddenly wouldn't have control over the vehicle, and you don't have a human on board as a, as a safety backup. So redundancy, I think, is the name of the game when it comes to safety of these systems, and we like to incorporate as much redundancy throughout the entire stack as it is, you know, as it makes sense to do so and is prudent to do so from a cost perspective.
One of the things we hear is that when you have multiple sensors, you gotta figure out what to do if they tell you different things, right? That's sort of a part of the, honestly, part of the Tesla story. It's like, well, if we have LiDAR, radar, and cameras, we gotta process. So tell me about what you guys do to sort of solve that problem.
Well, I think it's really building up confidence through consensus. And if you only have two systems, if you had two sensors and/or two processing pipelines, and one said, there's nothing in front of you, and the other said, there's something in front of you, it can be difficult to reconcile that. And that's generally how that argument is framed. In practice, though, it isn't just one or two. We have many, many, many different parallel pathways.
I'm always hesitant to quote a number because it's always changing as we add new systems and new AI pathways, but it's, you know, 20-plus pathways through the system. Meaning learned, train-learn models that are actually making determinations about what the vehicle is seeing and how it's understanding and how it's reasoning about the environment. And so if you have one that misses something, you have a consensus of many, many, many others to be able to understand what is ultimately there.
And so it's not just about, oh, this sensor says, "Go," and this sensor says, don't go, and you're like: Now, what do I do? That's how it's framed in the argument, in the conversation, but that's not how these systems are actually implemented in practice. So we do do a multi-sensor fusion, end-to-end AI solution, but then we back that up with many, many different neural pathways that are reasoning about various aspects of the system, various inputs, like camera only, like LiDAR only, like radar only.
B ut also then producing outputs that allow us to validate the end-to-end AI models. So when we talk about b lack box systems, it's important to frame the conversation in terms of: How do you validate that? And one of the challenges of a big AI end-to-end network is that it's kind of a black box. You don't get to peer into it, you don't get to understand how it works, and you don't get to evaluate why it makes certain decisions.
But if you have multiple of these systems, and some of them are more general and some of them are more specific, then you can back those up with certain understandings of things that you can actually probe, you can actually query. So it's about finding a combination of having the right number of pathways, and then building up that consensus through redundancy and validation.
What can you tell me about what you've learned in the Permian thus far? And, and I think alongside of that, how that applies, so we talked a little bit about this earlier, to the on-road network. 'Cause I think-
Yeah.
What we're doing in the Permian, it's still all private roads, if I'm correct?
Yeah, so Atlas operates the trucks primarily on private roads. The main reason to do that is so that you can actually carry a lot more load. Public roads are restricted to 80,000 pounds, call it 40 tons. The tractor itself is about 10 tons, and so you've got about 30 tons of capacity in your trailer. That's not even one full trailer load. And so if you want to deliver efficiently, and with Atlas, we've said, we've already announced that we're actually pulling doubles.
So these are full, double, fully loaded trailers behind our tractors, and we announced that we're working on triples, right? If you wanna pull that kind of load and get that kind of efficiency, you can only do that on private roads. You can't do that on public roads. And so there's no technical limitation. Our trucks are not technically limited to only private roads. Our trucks are very capable of operating. In fact, they don't know if it's private or public, right? A road is a road to the robot.
But when it comes to the legality of how you operate with heavy loads, and the customer, of course, wants to carry as much product as they can on a given run, because that's how you maximize ROI, that's gonna be done on the private roads. Atlas, in particular, has set up their infrastructure such that they have built this conveyor belt. It's called the Dune Express.
It's a 42-mile conveyor belt across the desert, where they move sand above and around the public roads, so that they can get into kind of the back, the heart of the well site territory, so that they can then utilize the trucks to maximum efficiency by carrying, you know, single, double, and even triples in that case.
Yeah, their delivered cost to the wellhead is significantly cheaper.
That's right. If you can.
Part of it is the Dune Express, and I think part of it is your vehicles.
Yeah, exactly. So for every run, if you can deliver three times as much product for one truckload use, then you're tripling your efficiency, right? So this is why they're being utilized there. But your original question was about what have we learned? And I talked about this earlier. There's what I call three pillars of autonomy. There's the technology pillar, there's the safety case pillar, and there's the product pillar. And really, where the learnings are in that final pillar, the product.
Because in the commercial space, if you just hand somebody a self-driving vehicle and say, here's your self-driving vehicle. Go forth, use it. They're gonna be like: Well, how do I use it? What's the interface? How do I control it? Even if you give them an app, they're like: What do I do with the app? You have to train them. We're not talking about company executives here, right?
We may agree at the executive level that this is the future of the business, but folks who are at the sites, or at the pickup and drop-off locations, that are at the docks, that are at the depots, they're just trying to get their job done on a day in and day out basis. So you really need to make the product seamless from the user's perspective, if you're gonna be effective in business. This is something I think the AV industry hasn't really addressed head-on. We haven't. They haven't talked about it, mainly because we hadn't gotten there.
We haven't actually gotten to a deployment where we actually needed to worry about the nitty-gritty details of things like: How do you turn it on, right? At the end of the day, these trucks need to be turned on, and, you know, if you're a research company, you might have, you know, command lines that you have to enter in a terminal at a computer, and, like, that doesn't work for a production system. You need to press a button. Whole thing needs to just turn on and be ready to go.
So there's a lot of industrialization, there's a lot of maturity involved in actually making the product usable by the customer, and you don't know what you don't know. We're technologists, and our AI is some of the most cutting-edge and advanced algorithms that you find anywhere out there, and we've designed them to run at ultra-low power settings, like in a truck. But none of that matters if ultimately the user doesn't have a good experience, and it's not efficient.
So we talk about this as the flywheel effect. We are learning the hard lessons that you can only get with a product in the market today, and we consider that to be a huge competitive advantage independent of the technology, right? People say they have better technology or worse technology, whatever. Independent of the technology argument, that those learnings from day-to-day usage, usability, app improvements, efficiency improvements, matter so much to our launch customer, Atlas, but those are learnings that we can then take to our other customers as well.
We've talked to many of our partners in the over-the-road space, who are actually really excited that we've been in the market with a real driverless product for well over a year now, because we're building up those chops. When we bring this to the long-haul market, we're gonna be ready to go. We're gonna be able to hit the ground running and actually move much more quickly, I think, than the competition.
One of the things that I think drove Atlas, right, was their Dune Express was in part about efficiency, it was in part about getting trucks off the road, and then obviously, you guys step in at the end of the Dune Express. But so part of it was decarbonization, I think part of it was safety, and part of it was cost. when you talk to potential customers, how do they think about the unit economics of going to an autonomous vehicle?
Yeah, it's an interesting one. There's so many factors, right? There's the classic driver shortage problem that everybody faces, and some people tell you there's no driver shortage, some people tell you there's a massive driver shortage. I think really when it comes to quality drivers, there's a shortage, and that's one of the big headaches that customers are, you know, routinely talking about. It's like hiring, retaining, you know, sign-on bonuses, turnover, churn, those are challenges that every fleet faces, whether it's an industrial or over-the-road fleet.
Getting rid of those headaches in a model where you can scale your fleet almost unboundedly because you have unlimited access to more and more autonomous trucks that do exactly what you tell it to do, day in and day out, is very, very appealing. Beyond that, you know, we are looking to provide an immediate cost discount to our customers, right? Especially those who are early adopters. This is a new technology. We recognize that.
There's gonna be some inefficiencies that we talked about earlier in the actual implementation, and so we want them to see a direct kinda unit economic savings right out of the gate. But then there's all kinds of ancillary effects to unit economics and TCO, like insurance, right? I don't have to tell people that insurance prices are skyrocketing, right? They're way higher than they've ever been, and it's a major problem. Nuclear verdicts are still a major challenge for the industry.
Well, imagine how much we can reduce nuclear verdicts when we have a system that's monitoring and storing data 360 degrees around the truck continuously throughout the drive. We can pinpoint the exact locations, velocities, accelerations, braking profiles of every other vehicle around us, and we can recreate a scene after the fact, right? Really powerful information that the industry has not had access to, and I've been told by executives, I won't name names.
But I've had conversations with trucking executives who said, even if you didn't do any of the other things that you talked about, if you could just deliver me that, it's worth it, right? So this is a compounding effect I think it's gonna have on the industry, beyond just the driver shortage, immediate cost savings, insurance, you know, reductions, nuclear verdict reductions, and then efficiency improvements. Asset utilization. Right now, the assets aren't maximally utilized.
We've built up a freight network around the country that is based on the concept of hours of service for a given driver. A lot of companies have implemented relay networks as a result of the human aspect, and we can not unwind those right away because that takes time, but we give customers flexibility to be able to utilize these fleets in ways that there was just never possible before.
And so you can think about redesigning and reoptimizing the network to take advantage of these assets that effectively can drive coast to coast without stopping, you know, if they have the appropriate fuel. So it's a long list of benefits I see, and I didn't even mention the safety aspect, right? Improving safety of our roadways, reducing congestion of our roadways is an imperative that I think we all share and we'll all find value from as well.
One of the things that we maybe should have started with this, but where does Kodiak sit in their journey as far as, you know, revenue growth, units, path to profitability? What have you sort of said about that, and what can you share with us as kind of where you stand now? You're relatively newly public.
We're newly public, so I can't actually say much. We're in our quiet period. We have our earnings coming soon for Q4. I can talk about Q3, which is admittedly a little stale at this point, but as of end of Q3, we had deployed 10 vehicles to customers that are out there running around the clock. We did guide to mid- to high-teens, so you should expect that to continue to grow throughout the year, and obviously now in 2026. Revenue is still on the small side, right?
So we're not. With 10 trucks deployed, you're not making massive amounts of revenue, but there is revenue nonetheless, and that is something that is a huge milestone for us as a company. Again, we've been around for about eight years. The majority of the self-driving industry is pre-revenue, right? It's still a coming soon story. It's, it'll be here soon, trust me.
Any day now, we're gonna have it, and I think we've really crossed the inflection point when it comes to having a product that is demonstrated to be working and valuable for customers in the real world. So we're excited about that. But yeah, revenues are still small. Truck numbers are still small. It's still very early days. We've taken a very prudent and safe, responsible approach to rolling this out. We also needed to secure the supply chain, the partners, the ecosystem.
There's a lot that we needed to build up around the product itself, outside of just the software and the AI, to make this possible. Now, we've done that with partners like Roush. We have a manufacturing partner where we can build these at scale. We just brought on Bosch as our tier one supplier for future generations of Kodiak Driver. That gives unlocks global scale for this at, you know, with quantities going into the hundreds of thousands or even millions. Putting all the pieces in place, now I think we're in a great position to go and start to capture real market share.
Just when you talk about the Kodiak Driver, just kind of explain that for everybody.
That's a great point. So, setting the scene, what does that involve? It involves our SensorPod solution. So the physical manifestation of the AV solution is these pods that you put kind of near where the mirrors would go on the truck, and that houses all of our sensing capabilities. So all of the LiDARs, the camera, the radar, and some of the boards that run those electronics. That's where you get some of your cleaning solutions and other things that are important for the product. And then we have the compute. So we have a compute stack inside the truck.
And then we augment the truck with redundant steering, redundant braking, and redundant power solutions, right? 'Cause everything runs on power, right? It runs on electronics. If we lose power, then the truck is not controllable. So obviously, we have a battery bank, but that's not good enough. We need to have two battery banks that are fully independent, and then two entire electrical interfaces, which are completely independent. So if we lose one, we still have one to go, same with braking, same with steering.
So that's, that's the whole package of the truck, and we've built it to be modular. So we can put it on many different form factors, different makes and models of vehicles. We know that customers care about their specific flavor of truck, everything from long-haul highway trucks to industrial Class 8 trucks to passenger vehicles like Fords and Ford F-150s, and we've even put this on tanks in our military work.
Can you talk a little bit about the competitive landscape, and, you know, who else is out there with similar products and maybe the compare and contrast what you're doing, that sort of differentiate or maybe just talk about the biggest differentiator you believe Kodiak is?
Yeah, I mean, and I don't wanna, I don't wanna single people out or exclude folks that may want to be on that list or not. So I'd prefer not to, not to give an actual name list. People can build up their own list. But in terms of differentiation, one of the things that's really starting to differentiate the market is whether or not you deploy what is now being considered kind of a legacy approach to the software. For instance, using high-definition maps.
That's something that we did very early on in the self-driving ecosystem. Kodiak does not rely on high-definition maps. That's something that requires a lot of pre-mapping. Obviously, you have to go in, you have to pre-map the environment, you have to validate those maps, you have to keep them up to date. There's a lot of cost involved. There's a lot of process involved. This is, this is how Waymo does it, by the way, right? So I'm not saying that can't be successful.
I'm just saying that it's not as kind of AI modern as it, I think it should be. Whereas Kodiak takes a very kind of AI-centric approach that allows us flexibility to incorporate new algorithms and new models as they come about. There's an explosion of AI research that's happening right now. I'm sure people are following that. And what's really amazing about what Kodiak is working on is this idea of taking these algorithms and these ideas that we know, we all collectively know, at data center scale, what I call data center scale.
So this is your ChatGPTs, your Geminis, et cetera, things that run at massive power draw and huge, huge clusters of compute in a data center. How do you take those ideas, those algorithms, those architectures, and distill them down into something that runs effectively in a shoebox, right? It, it's relatively powered by a hamster wheel. That's how I like to say it. You know, we have very, very low power compute capabilities, and so, but we still wanna get the most bang for the least buck when it comes to power consumption.
So that's really where the cutting edge is, and I think that's where Kodiak has a competitive advantage. Our ability to take these new architectures, these transformer-based networks, effectively LLMs, not quite 'cause we're not doing LLM translation, but effectively that same kind of architecture, and putting it at the edge, putting it on a truck, and running it successfully in a safety-critical environment, that's something that I think we have a distinct advantage in the market on.
If anybody has questions, just raise your hand and I'll call on you. The, you mentioned insurance earlier. Are there enough data points out there to see, like, when do you think we'll start to see the impact of insurance? 'Cause, I mean, obviously, you need more vehicles out there.
Yeah. I think so soon is the answer. So, for instance, already we're seeing with the Atlas deployed fleet, insurance that's comparable to human drivers. So on the same. So it's not massively cheaper, it's not, but it's not more expensive. So we're already seeing parity there, and I think that you're gonna start to see those prices come down over time. In the trucking market, in the over-the-road market, we just don't yet have the data because when you have a safety driver, you're doing R&D, you're doing testing, that's not as useful to insurance companies as the real deal.
They need to see the product out there in reasonable scale, driving reasonable number of miles before they can really make a determination. But if you look at the way the passenger car market has gone, if you look at the kind of the safety statistics from robotaxis, Waymo, and others, you know, the numbers are there. It's clearly much, much safer than human drivers, and it's only gonna get better as time passes, right?
I think we all can probably all agree that, as a society, our driving quality is on the way down, as modern distractions become more prolific to all of us, and the autonomy capabilities and the AI systems are getting better and better and better. And so you've got this like, you know, this inflection point. I think once you're past that hurdle, the capability of the system is gonna take off, and that should result and will result in significantly lower insurance costs.
Where do you stand on the over-the-road side? I know you talked a little bit about that earlier, but where do you stand on that? How is that developing?
So we deliberately spent the last few years focused on unstructured environments, industrial environments, and a lot of this is born out of our defense work. So we decided back in 2021 that we were gonna play in the defense space. We saw a huge opportunity in that market. I still see a huge opportunity in that market, and Kodiak is the only commercially mature solution for the defense application. But what that meant is we needed to push into unstructured environments. Defense branches are not so interested as much in highway operations.
They are actually interested in highway operations, but they also want you to go to the edge. You gotta be able to go through the forest, through the dirt paths, through the trails, and into the bushes. And so we really were pushing on that aspect of it, and that's where our off-road capabilities with Atlas started to come along, and we worked through those three pillars: the technology first, the safety case, and then ultimately the product.
Now that we have all of that experience under our belt, we're bringing that back to the over-the-road market, which of course is the largest market of the three, and people will be most familiar with. So we've said that we're working to pull the driver on over the road, and we're planning to do that in the second half of 2026. And so the team is actively working very hard on that.
It's, it's the number one priority for the company, and once we've crossed that hurdle, we have all the partners and supply chain in place that we should be able to hit the ground running and start to scale the business.
From a customer perspective, are you seeing interest in that service?
They want it yesterday. I think there was a time where trucking companies were incredibly skeptical of this technology, that they, you know, didn't really see how it was gonna have an impact on their business. We've done. I mean, we've run freight for these companies on a daily basis, just with a safety driver, so we're already replicating the product, just with a safety driver, safety observer behind the wheel. And I think the fleets and the trucking companies that we work with have gotten to a point now where they're ready for it.
We're fully integrated into their TMS solution. They are already able to dispatch self-driving vehicles. We do it on a daily basis. Now they're like: Okay, when does the driver get pulled? Because it's not economically viable until you're actually able to pull the observer. To make it an actual compelling product, it needs to truly be driverless. Once we get to driverless, I think people are really excited to jump on this and scale quickly.
Is the on-road environment an easier application than what you're doing off-road? I mean, obviously, the speeds are higher. And you also just to tie in, you talked about mapping. I think it's more difficult when you're on highways than, say, like in cities, to map. I'm not sure how that plays a role, but is your technology more suited for one or the other?
You know, it's, I don't think it's easier or harder, one or the other. There's different challenges in both. So yes, on highways, the speeds are higher, and anytime speeds are higher, the consequences are higher, and therefore, the risk is higher. At the end of the day, our safety case is about evaluating risk, risk relative to human driving performance. So the risk goes up as the speed goes up. Yes, so that's the default.
However, highways, and particularly freeways, right, controlled access environments, are by definition separated environments where all the traffic is generally moving in the same direction. And if there's stopped traffic up ahead, we have so many sensors, and we have an elevated perspective on the traffic, that we can see things way down the road. We can see, perceive 500, 600, 700, 800, 900 meters to a kilometer out in front of the truck, so we can easily come to a smooth, safe stop if traffic is stopped ahead.
And so the relative speeds tend to be not very high, and ultimately, when it comes to incidents, it's relative speeds that matter. So cases where you've got vehicles pulled over to the side, particularly where pedestrians are involved, somebody might be walking around a car that's broken down on the side, those are the kind of more critical situations that we need to really focus on and ensure that we give adequate space, buffer, detection, et cetera, and those are the types of things we focus on.
The flip side is that in industrial settings, we're in off-road environments where there's no separation. We have bi-directional traffic on very narrow roads, and so you've got oncoming vehicles. Even if they're only going 20, 25 miles per hour, that can be a closing speed of about 50 miles per hour, with the potential for a head-on collision if things go poorly. And so that's very hard. That's very, very challenging from a safety case perspective.
That's incredible amount of risk, arguably more risk than you would find on the highway. But we don't have to contend with complicated merges. There's no merges out there. There's lots of traffic. There's this common misconception that there's not a lot of traffic. We see tons of traffic. There's all kinds of vehicles out there moving, especially during shift changes, right? The entire workforce is going in and out of these locations, multiple times a day.
So there's lots of traffic on the road. You get opposing traffic, very narrow corridors, and that's really challenging. But we don't have to deal with things like merges. We have intersections, we have cows. So it's just a different environment that you have to contend with. I wouldn't say one is easier or harder. At the end of the day, I think having lower speed ultimately helps, but there's other challenges that you have to face.
I've driven around the Permian. It doesn't ever feel safe.
By the way, just to give you human-level performance data, you are 6 times more likely to get into a severe accident in the Permian than you are anywhere else in the country, on average. Right? It is a huge need. There's a huge need for safety systems in this environment, and people's lives are at much higher risk. We're all already at high risk when we get in cars and go on the roads, but significantly higher risk in these environments.
So one more question. If we're sitting here, let's say we're three to five years out, right? Jets finally won a Super Bowl, so I'm happy. Where do you think-
Right after the Dolphins.
Okay.
But yeah.
I'll take it. Where do you think we sit in, like, we go out, or is it gonna be the norm to see driverless trucks all over the roads in three to five? Like, how do you think it plays out?
I think three to five years, I mean, for the difference between three and five.
Okay.
I t's gonna be pretty significant, right? But I think, let's say, the five-year point, you're gonna go out on the road, and you're gonna be able to see autonomous trucks. It's not gonna be every truck, it's not gonna be every car, but they're going to be around for sure. In the same way that if you live in San Francisco today and you walk out on the streets, you see Waymos everywhere, right? They're now. And what are we, like, four years into the deployment of Waymos or so?
Right, so they become ubiquitous on certain lanes and certain corridors and certain places. But one of the great things about long-haul trucking is that we're really trying to focus on the remote places. We're trying to focus on the interstates. And so you, in your daily lives, are probably not gonna be seeing a lot of autonomous trucks out there. If you go on a road trip or if you're traveling to another state, then yes, you're gonna be, you're gonna be driving along, and you're gonna see self-driving vehicles.
They blend in, so it's not like they're super shiny. It's not quite like a Waymo, where you really see, you know, the spinning bucket on the top. You have to know what you're looking for, but absolutely, five years, they're gonna be ubiquitous, especially across the southern portion of the United States.
Great. Thanks. Thanks for your time.
All right, thank you.
Great conversation. Thanks. Thanks, this a wesome.
Yeah, this was great.