RJ Scaringe on Rivian's in-house 800-TOPS autonomy chip and path to mass market with the $45K R2

R1S in the driveway and they're very happy with it. And I'm very happy that we get to talk to R.J. from Riven here in the studio.

Welcome to the show. How are you doing?

Thanks so much for taking the time to come chat with us today.

It's great to have you.

Yeah. Nice to be on.

Yeah. Um I would love to I mean first set the table on you know like how the year went where the business is and then of course I want to go into the new chip and uh and just self-driving broadly but um but how how are you describing the scale of Rivian at at this point in time?

Well I mean we're we've launched as you as you just referenced we launched a set of flagship products. We have an R1S which is the bestselling premium SUV in the US and then um electric SUV in the US and then we have a truck R1T.

Yep.

And make commercial vehicles

but we haven't launched our mass market product yet which um which is coming in the form of what we call R2

R2.

And so that will dramatically expand the scale of the business and I think a lot of the uh a lot of the in some ways um challenge we've had is we're investing to be a very large business. So investing in vertically integrated technology to enable us to be a you know multi-million vehicle a year business and the technology investment has to happen before the products come and so we have you know huge fixed costs to develop all this tech whether it's you referenced it already whether it's a in-house silicon or

all of our compute platforms our operating system you know of course all the high voltage work

but that all comes together really nicely with the launch of R2 which is what really gives us this scale uh to support all this R&D.

Let's start with uh R1T uh electric truck felt very contrarian. It feels still a little contrarian truck buyers, you know, not to generalize, but it's a very much like I like my truck reliable with gas. I don't like this new thing. Uh did it seem like a risk to you at the time when you put when you said, "Okay, we're going to do the truck first. We're going to do that so early." uh like what what gave you confidence that it would deliver and you could actually win that market over?

You mean the truck the truck market specifically is such a it's such a large market and it's um

and I think we often think of it as like one singular type of buyer which is like the person who's loading up, you know, 2,000 pounds of concrete in the back.

Yep.

Vast majority of trucks used in the United States are actually used more like cars. So they're, you know, they're used as a daily driver.

The the the bed is used more for lifestyle activities. So, you know, put a dirt bike in the back or

got it,

you know, kids toys in the back, that kind of thing. And so we said, we're not as a company trying to address the contractor use case. So if you're if you want to load concrete and cement blocks in the back of the truck, this is probably not the vehicle you want to do it in. It could do that. It's capable of,

but that's not the brand aspiration. And it was one of the reasons we also made it really clear that we are had both the truck and then this sibling vehicle, the SUV. And there's so much shared content between the two. But the other thing we wanted to do is just eliminate any questions of whether or not it was capable. Um, you know, you'll laugh, but in 20 I guess 2019, I would get questions like, well, what if it gets wet? Can it drive through deep puddles? and like so there's just like a very um very low level of understanding I'd say generally around electric electrification what that can do in terms of off-road and capabilities and so with R1 we made it really capable uh so it's you know it can go extreme rock crawling it can go drive through 3 ft of water it can um you know it can accelerate faster than you know most hypercars so our our current quad the 0 to 60 in like three and a half seconds It's sort of

unnecessarily silly in terms of its capabilities, but it's a it's a flagship product, so it's there to make a statement. It's there for for building the brand.

Yeah. On on the SUV side, have you been surprised by how how much runway you've had in that market for full-size SUVs? um because it feels like a lot of people went after trucks very quickly, but the the fast followers haven't really come after the R1S. And from I mean, I grew up in a family with a Ford Explorer, and I've always been full size. Feels it feels like one of those things that in in 20 years people will look back at this window of time that Rivian had to just dominate that full size SUV category and and feel like this was the biggest possible blessing that you guys couldn't necessarily plan for, but it kind of worked out nicely.

Yeah, I mean it's uh to your point, I mean in the space of electric vehicles priced over $70,000, these are all depending on how you draw the boundary diagram. The way we look at it, we have about 35% market share. And so it's the bestselling electric uh premium SUV in the United States by a significant degree. You know, significantly outs selling something like a Model X, which is

uh also in that same category.

But it but interestingly, in the state of California, it's the best selling premium SUV, electric or non-electric. So really, if you for those that are spending time in California, if you're driving around the Bay Area, driving around SoCal, yeah, they're just everywhere. So, um, you know, our hope is that with R2, it's a price point that starts at 45, it captures a lot of the magic and the essence that's in R1.

Yeah.

But at a much lower price point. So, if we can have even a fraction of the market share that we have with R1 with R2.

Yeah.

We'll be we'll be very very very happy.

Where did the headlight design come from? When I first saw it, I was like, it looks like cartoon eyes. It was not for me. And now it's completely grown on me and it's just completely been like, "Oh, that's just the" It's it's become not only like I I I think of it as cool. I don't think of it as like cartoony or childish, but also it's simultaneously become iconic because I see it from a mile away and I'm like, "Oh, that Rivian." And I feel like that's exactly what you how you wanted it to play out. But what was the thinking with that? Because it does feel bold and different.

Yeah, definitely. And you know getting to something that starts to feel iconic. It's always hard cuz

you can't design something and it immediately becomes iconic. So

I think it has to start with a really strong point of view and a strong conviction around a a certain direction. So in the case of the front end I mean no in 2018 2019 no one knew what a riv looked like. We didn't know what a looked like. Uh so we were figuring out so there's lots of different faces of the vehicle and one of the things we arrived at was the realization we wanted it to look technically capable but also friendly. And so often if you go friendly it starts to look like nice but not tough. And so we wanted like nice but still tough. So it was a really hard balance.

The original Whimos from Google were like a little bit too nice and they were like these goofy things and then they finally Yeah.

But yeah continue. So, it was like it was a lot of iteration. And

one of the I grew up a car enthusiast and so I' I've always been drawn to cars that now we would consider be iconic, but if you look at the history of those cars, often in the beginning,

they're made fun of. And so, you know, if you look at the Volkswagen Beetle, uh, as an example, uh, that was thought of as heck, even Volkswagen called it ugly. Uh, they leaned into it.

Well, even on the on the other end of the spectrum, the F40 was controversial.

Yeah. And now it's obviously, you know, a car that a lot of

Yeah. I mean, a lot of super cars start out like kind of mock and then we're not being daring enough or something. Yeah.

Yeah. So, we wanted to push push some boundaries, but we also wanted to use really clean forms.

Mhm.

And the benefit of a clean form like

uh you know like primitive shapes in this case like half circles with you know like a like a um like a stadium. It doesn't age. It doesn't, you know, it's a form that has been around for ages and so it feels very timeless and it the way we integrated it very seamlessly. The hope was is that it would become something that's very recognizable. You see on the road, you said it's a Rivian and then that it could scale to different sized vehicles, different vehicle segments. So yeah, there's still people that that don't love it, but I'd say the vast majority of people who have been around it start have either liked it from the beginning or as you as you just described have grown to really like it. I want to talk about R2, but first I want to ask you about R0. Are we going bigger? I mean, if the numbers go when the numbers go up, the vehicles get smaller. What do I have to do to get something that competes with like a Cadillac Escalade ESV out of you? Because I I have a big family. I have two huge dogs and I want the absolute monster land yacht. And I'm just my my my serious question is uh like is there something about the physics of electric vehicles where it gets harder to make something that's really really long or huge or big or is it more just like that's more of a niche thing and maybe you'll get to it or the market will will get to it eventually but it's not the highest priority right now?

Yeah. Yeah. Well, I I love this question because we have um

we as a company have so many different ideas and so you can imagine there's like 50 different concepts or ideas we might have for a vehicle, we have to pick a few things to do. And so when you then look at the electrified space and you say what's what's going to sell at volume for a flagship product we started with something that was larger but as we move into mass market the the the size of the market that exists for these very very large vehicles things that are bigger than R1 is very very small. So there's not a lot of customers you know it's like maybe 10,000 units a year if that. Um, and then you look at just the the biggest segment by far is a two row, five passenger SUV, and I'd say there's probably like one highly compelling choice today under $50,000. That's the Tesla Model Y.

Yeah.

And so it's like wildly underserved. You

imagine there's 300 different choices in the in the combustion world and you've got like maybe one

there's many choices but there's one that's highly compelling which

how uh molded like how how people wake up every day and they they tell themselves they're buying the car that that they want but in reality there's like a regulatory environment that that forces manufacturers to make certain decisions and trade-offs and and push different vehicles

uh with the new Don't want seat belts in your car, right?

No. Inside joke. Um the uh but but with the new with uh cafe regulations going away, people are talking about uh America being flooded with these mini trucks,

trucks.

I'm not con I'm not convinced that that uh there's actually that many buyers of of those vehicles in the US. Uh may maybe there is, maybe there isn't. How how much do you feel like the the popularity of different vehicles is really just downstream of um choices that manufacturers have had to make because of for various regulatory reasons.

Boy, uh we have this discussion all the time. I mean the the causality of demand is interesting because it's uh you don't have an infinite set of choices. So therefore demand starts to look like what supply looks like. And we we see this we we believe this is going to happen with R2 where there's very few choices in I'll take let's talk about it in detail. So Tesla launch Model Y very very successful.

It's a it's a non-traditional form factor in terms of a midsize SUV. You know it's more it's very carlike. Yeah,

that has its advantages, but you've found a lot of people that are coming out of more traditional SUV form factors into that, but they don't have another choice. And so what we saw happen is a bunch of other manufacturers create their own version of a Tesla Model Y. So the form factor is very similar to a Model Y. You know, if you look at the side view of it and draw a line over the profile of it, it looks, you know, it's a Model Y. like

different OEM's versions of that,

which was actually the wrong conclusion to draw because you didn't provide customers with better diversity of choice, but rather you gave them a less good version of a Tesla Model Y. If you want a Model Y, buy the Model Y, not some other company's version of it. And so we were very clear on the need to have a a very different point of view and not uh fall into that trap of thinking everyone is going to want that exact vehicle form factor and profile package. Mhm.

And so I think that that happens all the time in automotive and and it's sort of shocking, but if you were to think about take like the sedan space, you have the BMW 3 series, the Mercedes C-Class, the Audi A4, they're all like dimensionally very very similar. And then you have, you know, the BMW 5 series and the Mercedes E-Class and the Audi A6. So you have like these interesting segments that are like very much uh because of this like hard to trace causality, you just have a lot of people building very similar things with different fronts, different rears, maybe slightly different surfacing,

but the the specs of the vehicle are highly aligned. And so I think there is an opportunity with electrification to reset some of those uh expectations and to reset segments, sizes, performance characteristics, some of the attributes. One of the things we worked on really hard in R2 was the rear seat like comfort or space is completely unlike anything else in the segment. So it's a lot of space.

Yeah.

Um but and then you look at all the cars in space and they're all like within 15 to 25 millimeters of each other for rear seat leg room and you're like Why is why are there 30 cars that have almost identical rear seat configurations? And it's

Well, yeah. When the when the team sees something like that, are they like, are we missing some sort of like regulation that that like requires like you ever and then and then you kind of trying to figure out, wait, do we actually have flexibil I always you end up wondering like are are we missing something and then you realize

well the the mechanical nature of a car is means that everybody can buy everybody's products. So every company like Rivians are owned by every car company and they take them apart and they own all the other, you know, all the other things they're competing against. And so as a result, a lot of like the mechanical innovations, let's say welding techniques or

casting techniques, they sort of very quickly become state-of-the-art. But the the downside of that ability to buy each other's products and so easily observe them is you do have this funneling towards like uh consensus around different segments. And so we we we sort of lose some of the uniqueness between the different vehicles. And see you like for for car enthusiasts you find a lot of times car enthusiasts will say a car they all look the same. They all feel the same.

Yeah. Uh, and so we we try hard to change that, but there are certain expectations. I'll give you a really good example. Um, on our R1T, we did a ton of research and we looked at what people put into the bed of the vehicle. And if you're using it for lifestyle, you want to be able to fit it in your garage. And if you're putting anything uh anything other than a motorcycle, which in which case you or dirt bike where you'd have the tailgate down, the bed length doesn't need to be longer than 4 and 1/2 ft.

Just doesn't. There's nothing that goes in that requires unless you're doing like construction and you need to put something. And so we made the bed shorter with this articulating tailgate that goosesenecks out. So it gives you it's actually longer with the tailgate down. So it fits a really long motorcycle. But when it's up, it it everything fits in the you know, it fits in the garage more easily. And so we we look at most if you drive through a neighborhood, you know how you know who has a truck in the neighborhood? It's because it's parked in the driveway. Most trucks don't in the garage.

Yeah.

Uh and so we wanted to fit in 95% of the garage in the United States. We went through all this thought process. We built it and the standard size is 5 foot. And the amount of times I've been asked, why didn't you make the bed 5t long? I said, well, of course we could have. It was just a decision. We made the decision. They wanted the truck to more easily fit in a garage. So those are like you sort of sometimes if you break the mold of like what's expected for dimensions because customers media car journalists are so trained on a certain size.

Sure.

The battle may not be worth it.

Yeah. And also there's like there's the decision to purchase a car and you might just be looking at specs and you're like five is better than four and a half but then you're live you live with a car for four years and you're like oh well this is actually way better. I never needed that extra half foot. Um yeah fascinating. What uh where do you think uh Chinese EVs are overhyped uh or underhyped?

Give us No, give us a white pill. Overhyped. I want the so recently there's been a lot of excitement around them and fortunately we're not letting them flood our country

uh with them. Uh but we had a guest on the show recently that had said he was in China a few weeks ago and and he there was one of the I I forget which manufacturer, but one of them were just like caught like side of the road burning on on fire. he felt like they were potentially overhyped because they do have some features that are kind of

scroll stopping. I'm thinking of the one that can jump uh

the Yong Wang U9.

Yeah. But uh but you know given that given that uh every manufacturer is free to assemble I mean disassemble any other vehicle and and learn from it. Uh what do you think American manufacturers can uh learn from what they're doing? H

yeah, I do think it's important to sort of uh pull the pull the curtain back on what sometimes I think gets presented as if it's magic, particularly on cost.

So I think there's two things to take away from uh the Chinese electric vehicle space. First, there's over a hundred different brands and manufacturers in China and there's only a small fraction of those that I would consider to be in the category what I'm going to talk about leading in technology and having really robust architectures. Um, but the two things with that said, so if if you say there's more than five less than 10 manufacturers that fall into the categories I'm about to describe, you have an interesting phenomena where a lot of these newer companies in China for the same reasons that Rivian or Tesla have very different software architectures, electronics architectures, and incumbent OEMs is they started with a clean sheet. Mhm.

And when you start with a clean sheet, you would very quickly arrive at a completely different technology topology than what evolved into cars over the last 50 or 60 years.

Yeah.

And what I mean by that is um prior to like 19 early 1960s, cars were 100% analog. So there are no computers in a car. And the first computer to make its way into a car was ironically was for the fuel injection system. And so for any of the car enthusiasts out there, this is like those original like Bosch Kronic fuel injection systems that we started to see emerge 1960s and early 70s.

And car companies at the time said, "Boy, we build engines, we design vehicle bodies, we assemble the cars. We don't need to make these little electronic modules." And so they pushed that work to suppliers. Companies like Bosch or Continental uh would make these little computers that run the fuel injection system. Then subsequent to that over the currents of the last 50 years a bunch of other things started to have a need for computers and your seats suddenly became smart and there was a computer that went with the seat. Your a your air conditioning became intelligent. There was a computer that went with that. your sunroof had a computer and before you knew it, the vehicle architecture was this proliferation of, you know, in some cars 100 to 150 little mini what we call electronic control units or computers that run these specific domains like the domain of an engine or the domain of a door or the domain of a seat. And it's precisely the opposite of what you'd architect if you were thinking about it as a clean sheet. We would never say I'm going to build a network architecture and software platform that has 150 different software code bases running 150 different little mini computers which communicate through this sort of klutzy can architecture

that might all be need to be independently updated at various points in the car's life cycle. It's just like it's a total disaster. So what you'd say is I'd have as few computers as possible doing as much as they can.

Yeah. And so what you know the the fancy way we describe that now is it's a zonal architecture. It's a computer that controls a whole zone. And so Tesla of course developed their architecture like that. We of course developed our architecture like that. And a couple of the Chinese did as well. And this the the real benefit of this besides just taking a lot of cost and complexity out is that you can make updates really easily. And so if I want to change let's say the sequence of events that occur when you unlock the car, I don't have to coordinate amongst 15 different suppliers. the supplier for the horn ECU, the supplier for the door lock ECU, the supplier for the, you know, the interior lighting ECU. I can do all of that in like a matter of minutes internally because it's running on our own software platform and it's all of our own code. And so just the the the emergence of regular updates, improved features, features that respond to dynamic customer needs, I think is a really big shift. And in the west, there's two companies that have that, Rivian and Tesla. Um and then in China there's as I said more than five less than 10 companies that have that.

Mhm.

And if you don't do a software defined architecture well the ability to do like AI integrated into the vehicle or an AI defined vehicle is enormously hard. Uh so you have to have all these ingredients to be able to do like the broader architecture well.

Anyway, so that's one big difference and that actually underpins we did a $5.8 8 billion software licensing deal with Volkswagen that that technology I just described for network architecture software OS u is what we license uh as part of a big partnership with Volkswagen group but the other is that the Chinese companies have just fundamentally lower cost structure they have much lower labor costs their cost of capitals often free or more than free meaning they get paid to build a plant or they get enormous government subsidies and so you can just build a spreadsheet to look at this it's not it's not magic they're building the cars. You they're not using some new type of material from outer space. It's it's using very, you know, consistent material. It's what we use in the United States. Consistent joining and forming techniques and putting the cars together are very similar. They just have labor costs and capital costs are much much lower than the United States.

If you had to pick a gimmick that you do like, what's your favorite gimmick? Because the jumping the jumping car is a pretty good one. I showed that to my four-year-old and he was like, I need to I need

I don't think I don't think it's good unless unless they're jumping a human like a demo where they jump a human other otherwise it's just a cheap trick. I don't need to jump I don't need to jump a paperclipip.

Okay. Yeah. Or a pothole. Hey, you do need to jump a pothole. You got a you got a flat tire.

I did blow out a tire.

You blew out a tire this weekend. Anyway, favorite gimmick. If like if you had to pick one, what what do you like? You know, there's like a there's the tank turn. There's, you know, a whole bunch of these different gimmicks. Is there are there any gimmicks that you think are fun

or maybe retro gimmick?

Uh, yeah. I don't know. Um,

I mean, they're just fun to watch. I don't know if I I we haven't Again, we haven't decided to put any of them in there. They're But in vehicles, we we do have something called kick steer in our vehicles, which when you're off-roading, because we have a quad motor, you can turn the vehicle on its axis.

That's good. Yeah.

Sort of cool. Yeah. Um

I mean even even the Gear Tunnel to some to some degree like stood out. It was like like the Gear Tunnel when that launched that was definitely like a like a oh well like this is different than normal. This is not something I normally and I feel like there there's a balance like you don't want to be all all gimmick all the time but uh you need a little bit of style and a little bit of substance. You need some

we got to talk about uh we we needed clearly needed an hour 90 minutes. I know I don't know if you're we got to talk about uh some of your guys' new releases. the chip uh and some of the new technical decisions that you guys are making around self-driving.

Sure. Yeah. Sure. Yeah. So, last week we actually finally announced it. Amazingly, it didn't leak. We've been working on this since 2020, early 2022,

but we um we've rearchitected our whole self-driving platform around what we call an end toend train model. So the model is using the millions and millions of miles that are being accumulated uh through our deployed fleet of our Gen 2 vehicles which launched a little more than a year ago to build effectively a neural net or a foundation model for how to drive.

Mhm.

And on our Gen 2 hardware stack, it's it's um it's using an Nvidia platform. It's got around 200 tops of compute. We've got 55 megapixels of cameras. uh nice array of radars, but it's a great platform for for building a data flywheel and for delivering ultimately this will be able to deliver pointto-oint uh you know autonomy. So you can put the address in the vehicle drive there. But to get to higher levels of autonomy, we've developed an in-house processor.

Boy, look at that timing. You guys did that perfectly in that video. Um but the in-house processor is a it's a significant step up. So it's a 800 tops. Mhm.

It's got 35 billion transistors uh on the silicon. Uh the neural net's capable of processing 5 billion pixels per second.

Uh so this is like an incredibly powerful platform and we brought it in house just to given the importance of vision. This is a vision based robot.

Yeah.

The the vehicle, but we have other things we're doing in the vision- based robotic space as well. And so

we came to the view that having our own inference was going to be really valuable.

Yeah. But it, you know, it took it took, you know, the better part of almost four years to build the team, develop it. You these are things that take a tremendous amount of capital.

Uh we're working with TSMC on making it,

but um yes, we're excited about that. And then we also include upgraded cameras and we have a new light arm

which is a another sensing modality that sits at the top of the windshield which will help us raise the the capability of the vehicle ultimately to what we call level four which is think of it as the vehicle can operate empty uh or without anybody in the driver's seat. So it could like pick your kids up from school or

you know drop you at the airport u these kinds of things.

Is LAR dangerous? I was telling Jordy that there was a lidar system that uh if you pointed your phone at it, it would damage the phone. And of course, you know, he said, "Well, if it's damaging the phone, can't be good for your eyes." How do you tell customers that uh okay, our LAR safe? Like what what are the what are the parameters to optimize around?

Yeah, I mean that that was just uh that was a poorly designed LAR system that was doing that. This lighter is Yeah. They don't burn out your phones. I certainly don't burn out your eyes.

Yeah. Okay.

As you're Boy, this is great. You showing the image here. So, I mean, the beauty of something LAR is we can use the LAR and the radar array to help train our cameras because again, this is a this is think of this as we're building a brain. We're building a neural net on driving the vehicle. And it has the ability to see very very far distances much further than the human eye or a camera could see but also um uh see in perfectly dark conditions or perfectly bright conditions. So

and and it imagine it helps with weather like extreme weather as well.

Yeah. Well, the the radar does particularly well where you have uh optical occlusion. Yeah. Like heavy fog or heavy rain or snow.

Sure. Um and then the LAR does need line of sight because it is a laser. So the LAR is very helpful for training the cameras and and training them the radar system.

Yeah.

But it's uh but it's also very helpful for long range

is uh uh is so I mean it seems like you've solved the inference problem like you're like the vehicle is ready to rock. Um are you worried about or is it is it a challenge to like accumulate enough data? Are you getting enough data off the current fleet uh that you feel like you're competitive there? Uh are you GPU rich or GPU poor in the in the training phase? Are there are is it is it about the algorithms like do you need more AI researchers like what is the most uh what is the rate limiting factor to actually delivering like something really top-notch?

Yeah. So so I mean the first is building this when you say data flywheel.

Yeah. every vehicle that we every gen two so we launched in end of 2021 and that was what we called our gen one vehicle.

Yeah.

In the summer of 2024 we updated it with a complete new set of hardware under the skin but really most importantly changed the self-driving platform

and that that's the beginning of this data accumulation platform. And what this is being used for is there's a whole host of trigger events that we identify interesting or or important information from the vehicles that are driving.

Yep.

That get sent back up through the cloud and then we process them as you said on a large um you know a large pool of of GPUs.

Yeah. And you know this is you we're talking like many thousands of GPUs um you know enormous amount of investment in GPU spend here but um but ultimately that's being used to train the model and so then we create this very large model that runs offline and then we distill it into something that's a little bit tighter

and smaller that runs real time in the vehicle on inference. And the beauty of expanding the inference capability with our Gen 3 platform is it allows us to run bigger models.

And so as the the model becomes larger and larger to understand all the nuances and sort of in intricacies of driving uh the need to compress that to run real time you know I should say the ceiling of what it can run and therefore the need to compress it is reduced. We don't have to compress it as much.

Sure. Sure. Yeah. That makes sense.

So but this is this is by far biggest investment category. Uh it's been interesting because we've when we first launched a lot of the investment was going into like the fundamentals of the business of ECUs base software high voltage architectures things like DCDC converters all this stuff to make the car real

now those platforms are pretty stable and we're now shifting a lot of those R&D dollars are going really almost you know entirely towards uh AI and our self-driving platform.

Yeah. Uh you saw Luminar went bankrupt today.

Uh what what do you think happens with that asset with their technology uh from here? Do do you think there's uh do you think there's I I imagine uh there's a number I can imagine a number of different types of businesses that would uh be interested in in the technology and what they built, but uh what's what's your read? Uh, get it. It's hard for me to comment. I It's It's not something we're looking at, but you know, I can't I can't comment on others. It Yeah,

there may be folks out there looking to buy it out of bankruptcy. I don't know.

Yeah, we'll see.

Convertibles.

Are we You're selling my dad. My dad really wants a convertible, maybe.

Are we doing No, I mean, uh, it correct me if I'm wrong. The number one bestselling convertible in the United States, Jeep Wrangler. Yeah, Jeep Wrangler is very popular.

Is it possible?

So, why don't we do it?

Why don't you do it?

So, it's so funny you ask this because we I I don't think I've ever said this before. Why not say it here? We um on R1, we had a version of the roof that was removable early on

and we actually toled it and we decided not to launch it because it was just a lot of complexity.

Yeah. Um, but in the fullness of time, at some point in Rivian's life, I'd love to have a version that has a top come off. If nothing else, just so that my dad will be satisfied.

Amazing.

I'll I'll buy I'll buy I'll buy one, too. I'll buy one, too.

Confirmed. Within the long arc of history, we are we are launching a convertible. Uh, no, thank you so much for coming on the show. I'm excited to see like how you guys would approach uh the removable top to do it in a very uh

I I I don't like a lot of convertible design specifically, but I if you guys were thinking about it and saying like, well, what's the actual technology we have today? What's the best way to make it? So, anyways, this was a super fun conversation. Uh and please come on again soon.

Uh we'll definitely reach out and and uh have your team reach out as well if if whenever you have news.

Yeah. Thanks so much for hopping on. This is a lot of fun. We'll talk to you soon. Cheers everyone.

Goodbye.

Let me tell you about profound get your brand mentioned in chat GBT. Reach millions of consumers who use pro who use AI to discover new products and brands. Um let me also tell you about this post from Nihilism Disrespector who says, "So is the US economic plan really just build the machine God?" Tyler, what do you think? Is the US's current economic plan

really just build the machine god?

Yes,

I I would agree. I think it is. I think that that is basically the plan. Um

we need to nationalize the labs.

National or not?

Why why are we nationalizing the labs?

That's the only way we can get we can marshall the the capital soon enough. It seems easy right now. Everyone's like, "Oh, it's so you know, you can raise so easily. Give me three years when you need 10 trillion

or even what is it?

The10 trillion dollar training run is$10 trillion cluster. There's only a couple ways you can get that.

Yeah, I don't know. I feel like there's a way that the government can can foot the bill. Well, well, not fully nationalizing, right? There's got to be a way.

Yeah, but I mean certainly, you know, maybe not nationalized, but there's a lot of ways where it's like it probably does make sense in some sense to to to uh subsidize Intel, stuff like that.

Well, react to this. The fact that Sergey Karayv Karv says it's the annual reminder the years left to escape the permanent underclass. It was infinity in 2020. It was infinity years to escape the permanent underclass in 2021. And for the last four years, it's been exactly two years to escape the permanent underclass. Feels like uh this AI thing isn't isn't actually creating a permanent underclass like it was supposed to.

Just two more years, bro. Then something different. Yeah.

Just two more years. Uh anyway, let me tell you about turbo puffer serverless vector and full text search built from first principles on object storage fast 10x cheaper and extremely scalable.

Okay, so John actually I had a question. Um so so

I'm always curious like it seems like in self-driving there's like Tesla and then there's KA

and I think of those as kind of the leaders, right? Yes.

But uh you know Kama there's like the open source, there's open pilot.

Yes. So why it seems like there just aren't that many manufacturers that are that are using open pilot. Is there like a reason are they is it like George hot just like hates big manufacturers and he wants to keep it

well so so I mean I'm not exactly sure the open source license I imagine that the open source license says you can't resell this. Now the question is if you bake it into a car I would imagine most most companies like Tesla sells self-driving as an upsell. So, you buy a Tesla, it doesn't just give you the Tesla full self-driving system for free. You have to pay an upgrade. I would imagine that Rivian matches that pricing model eventually. Uh, and so if you build on open source software that says you can reuse it, you can you can use this open source software, but you cannot sell it. Uh, I believe that isn't that the Apache license? Um, yeah,

there's a few different licenses. One, uh, there's I think MIT license you can do whatever you want with. Um, but there's one license that says you you you can use it, but you can't resell it and

not for commercial use.

Exactly. Exactly. Uh, yeah. Like the license transfers down the chain and so if you're giving it away, you have to get uh if they give it to you, you you have to give it to your customer. So you can't monetize that way. So there there's that. There's also the possibility, again, this is just off the top of my head, but uh I I would imagine that the is it the NHTSA, the National uh Highway Transport Authority, uh I believe that if it's coming from an OEM, from a vehicle manufacturer, and it's making claims about this is a level two system, the level three system, you can take your eyes off the road, you can take your hands off the wheel. that probably has to go through a certification that comma doesn't necessarily have to go through the certification for. So you're in this scenario where if you're effectively white labeling comma maybe you don't like you're you would be subject to more a higher level of regulation because you're not a third-party product. So I would imagine that you still like take a peek at the open at the open pilot repo and see how they're doing it. Uh but then ultimately you go and build your own sort of proprietary stack.

Yeah. I guess also for I mean Rivian they're they're using lidar so it's like just not compatible.

Yeah. Yeah. Riv uh lidar also just like way more cameras too like the the open pilot system is built on two cameras on the on the front of the device and then one on the on the back that looks at the the driver's face. And so you have a you have a wide angle and then sort of a telephoto that will show you the road far away and then the full surroundings. But, uh, most cars, Tesla, I'm sure the Rivian, uh, will typically have cameras on both, uh, both side mirrors. They'll also have cameras on the back, cameras, you know, the multiple on the front. So, you just get way more data. And that's why I think he was pushing this concept of like we're like the the autonomy compute module 3, ACM3, is capable of processing five billion pixels per second. It's like why do you need so many pixels? because you have multiple cameras that are feeding you 30 frames a second, 60 frames a second. Um, and so you need to be processing all of that. Whereas Open Pilot's not designed for a camera that's mounted on the bumper because it doesn't have a wire that goes to the bumper. It's just a phone that's stuck to the windshield effectively.

Uh, one more note. Uh, I believe Riven uses Luminar as a supplier.

Oh, interesting. And so hence why he couldn't

he couldn't comment but he said he wasn't looking at buying which is crazy

which is but I mean they supply they might be using a different one

going forward

who knows but they just made that announcement so potentially but I'm but I'm sure they have a variety of vendors that they're speaking with but

interesting. Oh well well let me tell you about graphite.dev code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. Um, how did you sleep last night? Did you Are Are you back in the game fully? Holiday season's upon us.

I slept.

Holiday parties are here.

Brutal.

I got a 93 though. Caught up. Saturday was rough. The The holiday party came came through. I made it out with a 69. Not good.

Avoid them. Avoid the VC holiday party entirely. If you care about your sleep, go to aidsleep.com. Exceptional sleep without exception.

I got a 54.

54 on the night of the VC holiday party. Rough.