Thank you, everybody, for joining us. My name is Mark Delaney, and I cover Rivian for Goldman Sachs. I am very pleased to have with me today James Philbin, SVP of Autonomy and AI, Vidya Rajagopalan, SVP of Electrical Hardware; and Chip Newcom, the VP of Investor Relations. Thank you all for coming. Great to be here. Happy to be here. Yeah. Well, one of the really exciting things out of Rivian recently has been R2, and the company recently launched that product, which, as many of you know, is a two-row SUV at a more affordable price point than the three-row and more premium R1 line that has been out previously. Chip, maybe we can start with you for this one. Can you share more on what Rivian is seeing with respect to R2 demand and feedback now that the vehicle has launched? Well, first, Mark, thanks for hosting us today and great to speak with many of the investors here. Hopefully, many out there have already gotten a chance to go out and demo drive an R2. Still early days in terms of what we're seeing on the R2. Most importantly, from a media perspective, the early feedback we've gotten, it's been absolutely phenomenal. This weekend, The New York Times posted their review. We've seen any number of comparisons against both ICE vehicles, EVs, and across the board, it's been really great feedback. So I think from our perspective as a company, it's showing that from a product market fit, we're absolutely hitting the right points for what the consumer is going to be looking for. Now, of course, it's all about execution and continuing to ramp our supply chain, continuing to ramp within our plant in Normal, but very excited about the early feedback we've seen. One of the metrics the company shared on the last earnings call was that conversions from the pre-order is roughly tracking ahead of what you had anticipated. I don't know if you can share more on how you're measuring that, any context you can give on that front. Yeah. For a little bit of context for everyone out there. We've been taking $100 refundable deposits for our potential customers if they'd like to purchase an R2, then we've been inviting folks in waves to then be able to come into our systems, configure their vehicle, and put in their intent to order and purchase, then starting to deliver against that. As we've been inviting those waves in, seeing increasingly a very good conversion rate of folks actually saying, "Okay, yes, I do want to buy the vehicle." It's interesting, as we're now introducing new trims, for example, Coastal Cloud, we just launched, seeing as we introduce each of those new waves, more interest coming through from customers. Continuing to see good progress there. That is great. In order to meet that demand, the company has been scaling up production. You had already started a single shift in Normal to make the R2. The plan was to add a second by the end of 3Q, so this current quarter, just a few weeks left. How is that ramp progressing? Yeah, still planning to add a second shift here in the third quarter, so working through the ramp process as you would expect. Any key bottlenecks the company needs to solve? Nothing that we have talked about. Continuing to work through, as with any ramp, working through the process of all of our suppliers. We can only move as fast as the slowest supplier in our supply chain, so working through that process. Great. Well, that's a helpful overview of the R2. We are very privileged to have James and Vidya from the tech team. I know you have a lot of various projects you're working on, and a great opportunity for us to dive into some of those. Well, one of the key themes, of course, at this conference is AI and applicable for Rivian in a few ways, but including some of the autonomy products. James, maybe I can direct this one to you. Where does the company stand with its plan to have supervised point-to-point driving available later this year and then eyes off for 2027? Yeah. So I think the team's been very hard at work on point-to-point recently. So scaling up the model, scaling up the data, and the validation of that model, and then actually putting the vehicles on the road in a test framework so that we can get feedback and feed it back in. So I think we're seeing very good progress. Actually, we recently gave demos to the Uber CEO and CFO. I think one of those folks posted about it on Twitter. So yeah, great feedback, and I'm very excited to get that feature out towards the end of the year. And what are the biggest technical challenges remaining? I think mostly around sort of dialing in a good humanistic behavior in all cases. So specifically around making sure the model is not too conservative, that it's sort of tracking speed as you would expect. I think one thing we've seen is actually very good behavior around relatively complex scenarios like construction or narrow negotiation. So very excited about the progress that's being made. Obviously, the validation needs to happen so that we can make sure it's safe before we ship it. Similar, as you think about moving from the point-to-point supervised to the eyes off in 2027, is it same sorts of things you're working through? Yeah. It's essentially the same system. Just imagine the continued development of this end-to-end model, where we gain confidence, we can validate on more long-tail cases. We can scale up the number of miles that that system sees. Eventually, you get to a confidence level together with the system architecture that can allow you to take that eyes off. One of the things we've heard so far today at the conference more broadly has been about tightening some compute capacity to train AI. As you think about trying to hit some of these milestones, are you able to get enough supply to train your AI? Yeah. Thankfully, we have obviously a very close relationship with Amazon, one of our biggest investors. Through AWS, we have a great supply of GPUs that at least for the next 6- 9 months is going to support as well. Past that, of course, we'll need to see, and I'm very interested to see how this market develops in terms of compute. And this may be for any of you, Chip, feel free to chime in on these as well. But the company's indicated that Autonomy+ adoption is trending positively. What have you learned so far about customer willingness to pay for this product? Yeah. So I think, yeah, generally tracking actually better than we expected. I think one thing I'd say is that the product is very kind of sticky in the sense that once customers have these features, they really struggle to go back. And I've actually seen this myself, testing point to point is, I really hate going back to vehicles now that don't have that feature set, even though it's at an early stage. So I think in general, those products are very sticky. People really love to use them. I think over time it's going to become increasingly table stakes for any new vehicle to have these advanced autonomy capabilities. So that's kind of what we see. And yeah, all of that is driving obviously demand for R2s and R1s and also Autonomy+. As you're engaged with customers around people purchasing R1, R2, what are they saying around whether or not those buying decisions are being influenced by what Rivian offers on its autonomy software side? And is that leading to incremental sales or maybe even any lost sales as you think about where you stand relative to competition? I think you typically see two maybe cohorts of customers. One cohort comes from the traditional ICE world, where obviously the autonomy levels are much, much lower. And there, I think, people are looking much more around the size of R2, its capability, how fun it is to drive, the stance you have in it. And that's where I think, yeah, R2 is really shining. Then on the EV side, where people may be coming from Tesla, which have higher expectations, I think Universal Hands-Free, which is what we launched last year, is going quite a long way with those customers. But then, the final set of that cohort is really looking for a point-to-point type feature. So we're hoping we can really knock it out of the park with both cohorts at the end of the year. Well, that software does need to run on hardware. So Vidya, maybe I can go to you for this next one. At the AI Day event back in December, you presented and were nice enough to take some of our questions. But you outlined a number of initiatives, including some custom silicon, the RAP1 processor. Maybe talk about why you made that choice and how important is custom silicon to Rivian's long-term competitive position in technology? Absolutely. I've always said there's three reasons we've built our own silicon, and they're performance, cost, and velocity. I'll start with the cost. It's pretty obvious. When you buy merchant silicon, there's a lot of margin stacking going on, and in this day and age, AI silicon commands quite a bit of premium. If we can shave off a big piece of it, that really helps us offer more at the same price point. So that's the number one reason. Number two reason is performance. In particular, because we are a vertically integrated company, we actually understand our vehicle top to bottom. So we understand the application really well, and we're able to actually design a piece of silicon that is tailored towards autonomy, tailored towards physical AI. It is not a piece of silicon that was designed for a data center inference product that's then migrated over here. It's really purpose-built for this purpose, and we believe that you can actually get the best performance as well as the best cost because we don't have superfluous stuff that comes from a previous vestigial piece of hardware that's sitting around there. At the same time, I've spent a long time prior to this in chip companies, and often chip companies design to benchmarks or sample applications or open source models. But actually having the real model we run, having that and co-developing the hardware along with it helps you really design the hardware optimally so you get really good performance for it. Then finally, another very important and most important part is the velocity, is James' team and my team are co-located. When we find issues, we can work through them very quickly, which it may seem like a trivial thing until you actually work at a company and work through that process. If you have to buy silicon from anybody else, you find a problem, you file a ticket, then their solutions or support team looks at it. If they're not able to solve it, they escalate to the engineering team. Then it takes usually it's a matter of weeks before it gets taken care of. When we're working with our teams, usually the teams walk over to each other, discuss it, and gets fixed much sooner. That's important because that's a virtuous cycle. You can move and iterate much faster. All three of these over generations really help you build a moat, and that's what we're after. As you think about the challenges that come with developing that silicon and some of the milestones you still need to maybe solve for the Gen 3 hardware, what does that mean for timelines, and is the Gen 3 hardware stack still on track for late 2026? Building silicon, and it is a very good point, is not for the faint of heart. It comes with a lot of risk because simple matter of fact is the wafer, a single wafer cost to tape out costs a lot of money, and then it is very expensive. Then when you put it in the oven, it comes back after four months, so the cost of mistake is very high. You have to spend a lot of time doing validation pre-tape out, which I am happy to say we are at the tail end of this. We have had silicon in-house for more than a year and a half at this point, so we feel very comfortable in the health and quality of the silicon. Because it goes into an automotive use case, we have to go through more rigorous testing, and all of that is underway and pretty much wrapping up. I should say within the next month and a half, we should be pretty much done in sort of the hardware characterization testing piece of it. We have vehicles that have the silicon that are driving around. The computer that actually uses the silicon is also built and well tested. We actually demoed that at AI Day last year, so we have had that in-house for over a year, too. More recently, it is not enough to have hardware. You have to have software running on it. So more recently, we have proven that the actual application, the autonomy application can be mapped to it. It has been mapped, and it is actually executing as expected. We are exercising the features that are released to the public, Universal Hands-Free, Lane Change on Command, Highway Assist. All of the autonomy features have been exercised and are working as expected. Not only are they working functionally, but we are meeting the performance targets we had set for ourselves. We had set a goal to at least be 4x as performant, and we are achieving that as yet without fully optimized software. Our compiler tools, all of the stuff that needs to work in order to make the silicon work, is all on track. Vidya, what should investors think about for when the Gen 3 hardware architecture can make its way onto the R1 platform? We haven't actually announced that yet, so that's something we're evaluating, but we haven't announced a timeline on that yet. The important thing, which I think is sometimes lost on people, is that when we come out with point-to-point, it will actually be available on Gen 2 and Gen 3. So what the RAP1 chip, our custom silicon chip, and LiDAR enable is really functionality at a later date. The eyes-off functionality we've talked of for later. There should be no difference at launch. Okay. For the Gen 3 hardware stack, in terms of costs, how does that compare relative to Gen 2? Part of the reason we designed it was for cost. The way to look at it is Gen 3 will enable a few different performance points. The lower performance point will be cheaper than the Gen 2 hardware, and the higher performance point will be on par or slightly higher. It gives us two different performance points and a lot of cost savings. It's cheaper dollar per top, if you look at it, or dollar per performance unit. Okay. I wanted to talk on AVs. As a part of the Uber Robotaxi announcement, the company plans to deploy AV starting in 2028 with initial deployments in San Francisco and Miami. I think the company's planning to deploy some R2 vehicles with safety drivers in San Francisco and Miami later this year. Maybe this one's for you, James. Can you share more on how this effort is going and the broader development there? Yeah. I think the progress towards that later milestone is tracking very well. We had, actually, an Uber review a few weeks ago, so I think that everyone's very pleased with the progress. As you said, part of that milestone involves testing in three urban areas by the end of the year. That's San Francisco, Miami, and Chicago. We already have expert driver teams already driving around, actually collecting data and validating the system that we want to show. Yeah, everything looks good from that point. Chip, maybe you can speak a bit on the financial implications of the Uber deal, both in terms of that business on a standalone basis, but also what it might mean in terms of selling R2 vehicles as well as the capital considerations. Yeah. A lot of benefits in terms of the partnership. First is, of course, the capital that we're getting from Uber as a partner, where in total, across the next several years, we expect to get $1.25 billion of equity capital from Uber. We already received, earlier this year, an initial $300 million of capital from them as part of the signing. To James's point, later this year, we expect to achieve the second milestone, which will then unlock another $250 million, and then we'll have three more milestones thereafter to unlock the final $700 million. Those milestones are going to be around achieving certain technical hurdles associated with an L4-capable R2 for the Robotaxi service, and then the expansion of that service to up to 25 markets around the world, including at least one in Europe. From a capital perspective, one, that helps us enable the acceleration of spend on our R&D around autonomy. That is a big part of what we need to be doing in the near term. Then over the medium to longer term, once we have an L4-capable R2, of course, we have an initial agreement for them to purchase 10,000 vehicles, either Uber or their fleet partners, to then operate on the Uber network with then another option for another 40,000 thereafter. An important part that we will get with that once these vehicles are out on the Uber network is the software licensing fees associated with the cars actually driving around. I think that is an important part that many investors miss, is that a really attractive part of the economics is the fact that it is going to be Rivian's driver that is going to be driving around over time. I do not know that you have, but any quantification of what that licensing fee would look like? We have not described what that licensing fee is going to look like as of yet. Okay. Maybe talk about what the technology looks like for an L4 vehicle relative to what Rivian's currently working on for consumer vehicles, both from a hardware and software perspective, and maybe how similar is it versus some differences, either on the hardware or software side? Ultimately, we think the software stack should essentially be the same. You can imagine the same stack and data loop that's ultimately powering point-to-point getting better and better and better. As you scale up in terms of the number of data, the amount of driving actually the customer fleet is doing, and you're also validating further and further over more miles, then that system gets better and better, and then you can transition it to increasingly driverless domains. Now, supporting that, you need a robust safety architecture. That's something that very much we're thinking through. It actually comes with the RAP1 silicon. There will also be some additional sensor changes for the final L4 product that we're working through the details of now. But essentially, you can think of the stack, the same stack powering the consumer fleet, ultimately also powering the Robotaxi fleet. Vidya, anything to add? I would add, yeah, I think in the hardware space, what we will be doing is really having more sensing. I would say more and more capable is the way to look at it, in order to make sure that all the corner cases are fully covered. In particular, Robotaxi has to start from a stationary spot with nobody in the vehicle, which is not typically a problem you have to solve for a consumer vehicle. That means there's a higher need for sensing around. There will be additional sensing to augment the cases that are not covered in a normal consumer vehicle. Then we're just going to provision for obviously more compute, because you want to really provision for really having all the compute you might need as the models get more sophisticated. James made an important point here for investors, is that the future end state is not just purely about commercial applications like Robotaxi. It really is around this concept of personal L4 over time. So the idea that all of us could collectively get our time back so that your commute now, you can be productive in ways that is not focusing on the road. For a lot of consumers, there's a lot of value to that over time. Yeah, we see actually Robotaxi is not the end state for us. We really think personal level 4 is the product that we want to get to. Actually, Robotaxi is a step along the way towards that product. So really a vehicle that you own, you can direct it to go where you want, it can go and pick things up, drop things off, and really kind of work for you rather than being stuck to you. Very interesting. Across the board, you would think about the cases where there's nobody in the vehicle to handle certain cases. So there's just a higher bar on some of those things. So we look at the entire hardware architecture and look at, what if XYZ and there's nobody, what do you need to do? So I would say everything in general is made more robust to deal with that scenario. Okay. It sounds you guys have a fair amount of work to come on all these different efforts. Another one I wanted to touch on briefly is robotics, another physical AI application. Rivian spun off Mind Robotics. How much of what Rivian is working on is applicable to the field of robotics? Is that something that your teams are engaged in? I think the jury's still out a little bit on how much these models transfer directly to the physical AI space, or not. I would say that the idea of using human-derived ground truth is very powerful. The imitation learning, that is something that our models are trained to do. In the robotics spaces, two additional problems. One is the actuation is much more complex. There is a lot more degrees of freedom, and it is actually not clear how to get the human-derived ground truth from that. Then it is also, you have some more ambiguity in the classes and the types of problems you have to solve. There is more generalization problems you have to solve versus driving, which is really the one task. I think, the teams talk closely together on what are the best techniques. I do not know that they will be the same right now. I would say on the hardware side, the chip we built is really, the first application was autonomy, but it is a very capable chip for physical AI. If anybody out there is interested in using our chip for robots to reach out. It definitely is an area of potential engagement. Of course, we cannot speak for Mind, they are another company. Okay. That makes a lot of sense. I wanted to talk about the collaboration with VW, and you have the joint venture with them. Maybe talk about how the integration of Rivian's electrical architecture is going with respect to VW, and the timeline for when the first VW vehicles may launch using Rivian technology. Yes. So in terms of VW has discussed this externally. They expect to roll out their first vehicle that includes the zonal electrical architecture and software next year, which will actually unlock the last equity investment by VW. Overall, the partnership continues to be going very well and looking forward to having their first vehicle out on the road. What is the company's current thought around potentially licensing the EE architecture to other auto OEMs? Look, I think, always open to have conversations, and I think RJ being a very prominent member of the automotive leadership community is talking with any number of other OEMs. I think part of what the industry has learned over time is that this shift to a software-defined vehicle is where consumers want things to head over time. So the legacy architecture of outsourcing every single black box in the software to a tier 1 or even a tier 2 provider. As you think about that software-defined vehicle that we're developing now and have now, there are limitations with that that you can't offer with what we can now do. So certainly it's a latent opportunity, but our biggest focus right now is continuing to execute both in terms of our roadmap with R2, but then also executing with our roadmap with our partners at Volkswagen. I do want to spend a couple of minutes around battery technology. I think it's a really important part of the long-term opportunity with electric vehicles and some really fascinating stuff that's underway. For R2, using 4695 cells, maybe talk about why Rivian made the choice to use 4695. Yeah. So one, it's about higher power density, but then also from a structure of the R2, we now have a structural battery pack. So you've got fewer battery cells within the battery as well. A lot of it, as we think about driving over the long haul, it's about how do you drive down costs for manufacturing R2. Any update on when Rivian will source those from the U.S.? The plan is still with LG to start sourcing that next year. There's also innovation beyond the cell form factors in terms of ultra-fast charging cells, and there's been some products that have been announced recently where batteries can charge in 3-5 minutes, which I think could be game-changing for the EV market, including in the U.S., where people like to road trip. So I'm curious what Rivian's thoughts are about potentially using that kind of ultra-fast charging battery technology over the medium to longer term. Well, look, and certainly either of you jump in here as well. I think the question is the use case for customers and what's the necessity there. We were talking about this earlier today in some of our conversations, but for the vast majority of owners of EVs in the U.S. are plugging their cars in overnight and charging things up overnight. Now, certainly there are use cases where you're doing the long distance drive going across the country, but current technology can cover the vast majority of those use cases. Even, I will say personally, when I'm driving with my family in our EV, going on that longer road trip, it's actually nice to be able to stop and get out for 20 minutes and let the kids run around, go get a Starbucks, go get a sandwich. I think part of what a lot of consumers, until you've owned an electric vehicle, don't realize is it's just a different behavior pattern for how you're going to treat a road trip. It's no longer, "Okay, we got to fill up the gas tank as quickly as we can and get back on the road." It might be slightly more leisurely in that you're waiting to charge for 15 or 20 minutes. But at least in the case of my family, I actually think it's a very nice and different way of approaching a long-term drive. Vidya, just from an engineering perspective on the hardware side and what would have to happen to integrate those types of cells and how easy or hard might that be? I look at the analogies like LiDAR. LiDAR technology existed for a while, but the question is always, when are you integrating it? For the longest time, it was too big and too expensive, right? I think when we just integrated now for R2, it was both of them were right. The price was right, and the form factor was right. I think the same with alternate technologies when the price is right and the use case is right, sort of justifies it. I think we can definitely look at it. I think we've had a lot of experience dealing with battery management systems. That's where all the complexities move to the BMS, and I don't think that's easily a solvable problem. I think it's the availability of the technology at the right price, to just say. Very interesting. A handful of financial questions and maybe Chip, I'll direct these at you. The company's expecting a 50% reduction in bill of materials for R2 relative to R1. Maybe talk about to what extent that's on track, especially with all of the inflation around materials, semiconductors, and other components. Yeah, that goal to get there is as we've fully ramped up R2 at our plant in Normal. Still relatively early days in terms of that ramp, as we talked about on our most recent earnings call. In the second quarter, we did incur about $100 million of incremental expense through COGS associated with the ramp of R2. As we said on the call, expect to continue to see more ramp costs flowing through here in the third quarter before in the fourth quarter, then flipping the other way and being on our way towards achieving automotive gross profit positive on an exit rate basis for the year. So that still remains the goal, but we're still early in our ramp. As you just alluded to, the company does have that positive automotive gross profit target exiting this year. What are the key things that need to happen to get there? Well, we haven't given a specific volume number, but it really is around continuing to see our supply chain mature, continuing to have them ramp so then we can ramp within the plant in Normal. Because a big part of what is going to help our automotive gross profitability is both delivering more R2s, but then also seeing that fixed cost leverage within the plant as well, where as we start then sharing the shared cost across our paint shop and our stamping press across a much broader set of manufacturing, across R1, R2, and EDV. That helps the profitability not just of R2, but also the other vehicle lines. Scale kind of being one of those main drivers to getting there. You guys aren't making any unusual assumptions around pricing or cost reductions or anything like that? No, correct. It's all about scaling and building more R2s. Well, maybe bridge us from that target of positive auto gross margin exiting this year to the medium to longer term positive EBITDA. I know the formal target there, net with some of the extra spending you're making on autonomy has been removed, but what are some of the key parameters investors should think about as Rivian ultimately works toward that positive EBITDA number? Yeah, look, I think it's going to be driven by several factors. One is our ramp of R2 and how successful and how fast we are on the ramp of R2, plus then the demand for R2. Second then is our continued investments in R&D and SG&A over time and what are those going to look like. We haven't provided a new long-term target, but it's going to be based off of those key drivers. Okay. On the balance sheet at the end of 2Q, the company had $14 billion of cash and available liquidity. Talk a bit about how the company is thinking about managing the balance sheet. It's not cash and available liquidity, that's cash and then the investments that we expect to receive, plus then our Department of Energy loan gets us up to the $14 billion. But we do have that line of sight for funding the business. In terms of managing the balance sheet on a go-forward basis, first and foremost, it's going to be continuing to execute on the ramp of R2 because that's then going to drive what our profitability ends up looking like and the funding needs associated with that. Second then, and part of the reason why we raised capital back in July is around funding the equity commitments and then any reserves around the plant in Georgia. We feel as though we'll continue to remain opportunistic should we need to raise additional capital, but have a good line of sight for a significant amount of capital for the next good period of time here. A couple other questions I wanted to ask related to R2. The company talked about selling that in Europe eventually. Can you share more on your current plans of when that might happen, and does the penetration of the Chinese companies into Europe alter that calculus at all? Well, look, we know R2, and even R3 eventually, will be very well-priced for the European market and are, given the size and form factor for what they're looking for in Europe, could be a very good opportunity. Our focus right now remains on continuing to ramp here in North America first and foremost, recognizing that it is a competitive landscape and there are some very interesting vehicles coming into the European market, but we want to make sure we're executing here first. James, as you think about the software side potentially selling into other geographies like Europe, obviously different regulations, maybe different driving patterns. What would you have to do if you're going to offer some of these autonomy features in Europe from your team? How similar or different might it be? Yeah. The current approach is we take maybe 10 to one customer data to expert driving data. The reason we need that expert data is to kind of dial in better behavior. So customers are often not stopping at stop signs fully. They may be speeding slightly in some cases. So we actually want that expert data to kind of fine-tune, hone in a good, safe Rivian driving experience. So we would deploy in a similar fashion, that expert team in Europe or wherever the new country is, collect a certain amount of data there. When we've done that in the past, we've actually seen good transferability. Then we add that to the pool, and then you fine-tune. So that would be essentially the approach. We think that with relatively modest coverage of new road conditions and new road rules, that we can actually adapt the system fairly quickly to those new cases. One other thing I wanted to close on was around autonomy for commercial vehicles and just some of the broader things maybe your team's working on there. I know the EDV business has been growing pretty nicely, but to what extent are you working on new products for that use case? Yeah. I think in general, again, we see that stack as being very general. We've built flexibility into the system kind of from the get-go. So we can move sensors around. We can add and remove different modalities as needed. The system largely stays the same. So now we would, in a similar way, need to get a certain amount of fine-tuning data for new form factors. But we think ultimately the driving task is pretty similar in all these cases, and the system should generalize well. Have you guys given any timeframe for autonomy products in the commercial space? No, we haven't. But it's something we're always discussing internally. Great. Well, I learned a lot. Thank you all for taking the time to speak with us. Appreciate it. Thanks, Mark. Thank you. Thank you.