In-house AI chips will let Rivian control its own destiny: CEO

00:00 jared

Another top trending stock is Rivian. Shares of the EV market maker rebounded from Thursday’s decline after a series of announcements focused on autonomous driving and chips. More from senior reporter Pras Subramanian here and Pras, you just met with the CEO of Rivian and we had a big event today. I don’t know if there is a correlation.

00:20 Pras Subramanyan

Yes, some Wall Street commentary, you know, maybe me, I of course me, but there’s also some bullish Wall Street commentary. But yes, Rivian’s autonomy and AI day was a big deal. The company outlined its self-driving strategy, an iterative process that would start with a universal hands-free system, similar to GM Super Cruise. It will run for 3.5 million miles in the US, and then it will reach personal level four, turning into an eyeless and hands-free self-driving product, meaning it will drive you around like a robot taxi, and even perform some tasks like picking up your kids. I spoke with CEO RJ Scorange at the event, as you mentioned Jared, and we started by talking about why custom silicon, in-house designed chips, are kind of the way to go and why it makes sense for their AI-centric approach.

01:13 RJ Scary

We made the decision years ago to build our entire vertical software platform in-house, all our computers in-house, and not rely on Tier One vendors for any of this content. It wasn’t like we said we were going to do this and then didn’t put in the work and then created teams of thousands of people to put it all together. You know, we’ve spent many, many hundreds of millions of dollars developing these systems, and as we get closer to autonomy and our overall AI roadmap,

01:54 RJ Scary

We had a very similar view that this would be the most important technical change in transportation. I would say more than anything else since the inception of the car, we wanted to control our own destiny. When we started the process of redesigning our platform and clean shooting in 2022, we said that we wanted to do this, we wanted to control the perception platform. We want to build a very robust data flywheel. We want to build this around extremely high-level inferences. So it’s a very, very smart brain.

02:37 RJ Scary

and we need to fit that into a cost-effective package, and to do that, we’re one of the biggest drivers of cost, the implication is computation. And finally we decided to bring it home. And then again, if you do that, that just means you’re signing up to spend hundreds of millions of dollars on the chip. You need a huge team for this, you need to hire a team that is incredibly talented in silicon development. You need a partnership with TSMC or equivalent, but most likely TMC TSMC is our partner in our case.

03:22 RJ Scary

And what we’ll get is better performance, which is something very specific to vision-based robotics. Here is an example tool. You can imagine that other examples in our work could benefit from an inference platform for vision-based robotics. And then we needed to create this data wheel. And a lot of offline training, which means a lot of GPUs. But this is by far our largest investment category as a company in terms of R&D.

04:10 Pras Subramanyan

You know, we hear a lot about data center plays and people investing heavily in chips and how making your own chips cuts your margins in half, right? It can be this powerful once the investment is made. So, you know, I want to mention Tesla because they’re kind of doing the same thing in terms of going completely vertical. They have their own chips, they build their own data centers. Do you think your autonomy plays out differently with different sensors, sweetie, do you think you can somehow meet and surpass where you are now in the next few years?

04:52 RJ Scary

So of course our goal is to be at world standards. I mean, obviously Tesla is a great product. They have built a truly impressive platform. We very much agree with them in believing that this should be a neural network-based approach, you know, building an end-to-end trained model that uses a fleet of vehicles, building a data flywheel that trains the model with reinforcement learning. So we’re really on the same page on this one. We have a different view on its benefits.

05:32 RJ Scary

It is a combination of various detection methods. You know, in our case the camera, we call it the primary camera because it’s still camera-dominant, but by incorporating radar and lidar, it allows us to turn our entire fleet into a ground realities fleet. So every vehicle on the road helps a lot in the training of our model. So, we think this will allow us to build the model very quickly, and then these sensors do a really good job of helping us build the model as well as covering some pretty extreme corner cases.

06:12 RJ Scary

Of course, radars have the advantage of not being affected by low visibility conditions, so they are very successful in things like fog or rain. And Lidar has the advantage of being able to actually, ah, accurately detect the three-dimensional world and can be used to train a two-dimensional image of a camera. So you can actually use it to increase the speed at which the neural network understands the world. And that’s why

06:50 RJ Scary

This is our approach, but our view moves towards the fourth level. So that’s the last thing I think everyone should be thinking about. And this is no longer the science fiction of 10 years from now, 15 years from now. This is on a fairly short horizon as we believe that within the next few years, personally owned vehicles should offer level four. This will change the way you think about your car. Suddenly your car

07:29 RJ Scary

Your car will do things almost like if you had a nanny you could pick up the kids from school with your car. I have three children and they are nine, eight, seven. So they play sports, there are many different sports, they are constantly training and dropping out, it’s like pure chaos trying to get everyone to the right place. And so, I know in my own life it would be very helpful to have a car that could transport them to different places, different practices, and events that they would go to.

08:12 Pras Subramanyan

You know, we hear about 2026, maybe there will be a turning point there. So we’ll keep an eye on it. I know your time is limited, but I want to ask you about the big picture of the EV industry in America. You know, the tax credit disappeared in Q, sorry, in Q4. How are things going for you? I know we can’t talk about metrics, but how is the market finding you right now?

08:39 Pras Subramanyan

And I’ve heard people talk about how pure EV manufacturers can really benefit from this because they will have fewer competitors in the market. I heard that. I’m not sure if this is true or not, but what is your general view on what’s going on in the US?

09:00 RJ Scary

I think it will not change materially in the long run. It does not change the current situation. I think dynamics are also interesting to think about, such as system dynamics and the causality of what others will do. I think what you suggest in your question is probably true to some extent. I think you’re seeing a lot of manufacturers pulling back. I think this is unfortunate. I think it would be healthier for the automotive industry if everyone was like that.

09:59 RJ Scary

focused on electrification. I think the withdrawal means there will be less choice, which is bad for customers. But I think having fewer options means there will be a less competitive environment. So we can’t control what others do. We can control our own business. Things we can do to encourage the adoption of electrification include creating incredible products and giving customers additional options. You know, if you look at the segment of vehicles that start under $50,000,

10:49 RJ Scary

There are very few extremely compelling options. and as a result, a single company, Tesla, had approximately 50% market share. You know, it moves from quarter to quarter, but that’s what the sub-$50,000 half of the market calls a single company. And that’s not a reflection of a healthy industry. This is a reflection of an underserved market. And if we want to see the adoption rate of electrification go from 8%, which in the United States today is 25 or 30 or eventually 100%,

11:39 RJ Scary

We need more than one great option. And I think Tesla will continue to be a great choice, but I think there should be others as well. And of course I think Rivian is one of the ones that has our R2. I think it will be extraordinary. The car itself is truly magnificent. I hope there are others as well and, you know, we’re excited to have a partnership with Volkswagen where we help implement some of our technology into their vehicles. I know the tools we make in partnership with them will be great tools as well, so there will be some options.

12:15 RJ Scary

But um, but I think there aren’t enough options in the market today.

12:18 Pras Subramanyan

So yeah, you know, the Rivian R2 with their midsize vehicle is going to be a big deal when it comes out next year and it’s going to kind of take off throughout the year because as RJ mentioned, you know Tesla dominates half the market with EVs under $50,000 and they really need that to fuel that and they’re really making a difference from a mainstream standpoint with EVs.

12:44 jared

And making your own chips is no joke. and we’ll leave it at that for time reasons, but thanks for stopping by Pras, great interview.