NVIDIA Corporation (NVDA) Management Presents at Evercore ISI Autotech & AI Forum Conference (Transcript)
Sep. 21, 2021 2:59 PM ET
NVIDIA Corporation (NVDA)
SA Transcripts
NVIDIA Corporation (NASDAQ: NVDA)Evercore ISI Autotech & AI Forum Conference September 21, 2021 11:45 AM ET
Company Participants
C.J. Muse - Evercore ISI
Conference Call Participants
Ali Kani - VP and General Manager of Automotive
C.J. Muse
Well, good morning, good afternoon. Thank you for joining us; Evercore ISI's Autotech & AI Forum. My name is C.J. Muse, Semiconductor and Semiconductor Equipment analyst here at Evercore ISI.
Very pleased to have NVIDIA with us today. Specifically, Ali Kani, VP and General Manager of Automotive. We've got roughly 35 to hear from Ali. He's going to present briefly and then we'll move to Q&A. If you have any questions in the audience, please feel free to email them to me and I'll try to put them in.
And with that, thank you, Ali. Great to have you here and I'll turn it over to you.
Ali Kani
Thank you very much. It's a pleasure to be here today. We can jump in. So these are just going to be our forward-looking statements.
Okay, so just wanted to start by talking about our automotive strategy at NVIDIA. And I think where we really differentiate ourselves is that we're very focused on building a platform that's open and modular such that partners can work with us at any level of the stack. So, we have some customers who just buy our computers in the car and they've built the full autonomous driving stack, do the training, simulation build their own AV application and cockpit application themselves and, that's a great partnership for us.
But we have cases where customers also ask us to build their autonomous driving application and do the training and simulation for them and we design this stack in a modular way, such that you can actually pick and choose what you want from NVIDIA and the fact that we sort of build this stack end to end ourselves makes our product better across the stack and just to note that we also have customers who don't use our computers in the car, but they work with us in the cloud for training and simulation. And so, having that open and modular platform that happens to be end to end is really fundamental to our strategy and a big, big reason of the success that NVIDIA is having in the automotive business.
So one of the things we do to make self-driving development easier for our partners is we try to build a platform that makes it easy for people to build their own applications. So even though we're building an autonomous driving application and a cockpit application, many of our customers and actually most of our customers build it themselves. And so we try to build a platform that is really attractive for those partners. So we have our computer architecture and a sensor set that we create that actually is designed to be functional and safe and secure and we provide the infrastructure and tools for an OEM to build their own AV application and cockpit application and the fact that we built the platform in this sort of open, modular way, we design it to be upgradeable across generations.
It makes it so much easier for our partners to develop on our platform. We have customers who are, for example, using us on Xavier platform and since our platform is designed to be upgradable, it's easy to take open Appen [ph] or I'm sorry, point [ph] computer and upgraded from a Xavier computer and we designed our platform to be software compatible. So the applications you built on Xavier can just run on Orent [ph] when you get it and designing the platform that way saves customers so much time and wasted investment, right. And that's a big part of our strategy is just make it easy for people to want to build their own AV development and the only reason why we can do this as we're building the platform ourselves, and we recognize those issues and so we try to enable the things that help us to develop faster to our partners in the ecosystem.
Okay. So, we talk about making the platform upgradable. The way I think about this as similar to how Tesla builds their cars is, you could buy a Tesla with a Hardware 2, but then you can upgrade it with the Hardware 2.5, and then you can upgrade it with a Hardware 3. Part of what we're doing is building this equivalent sort of TDP computer form factor and TDP compatible computer that you can then just plug in from your vehicle and so we keep giving you these upgrades as an OEM.
Take your Xavier computer, upgrade it to an Orient [ph] computer, upgraded to Allen computer and the pace of innovation that NVIDIA offers OEMs is really significant. And because its software compatible, all the investment you make on us on our Xavier platform can be pulled over to the Orion platform. And this is why the ecosystem and our platform is growing because once you develop on our platform, there's a lot of advantage for you to stay on it. Not just because the roadmap keeps getting better, but we make the platform so easy to go from one generation to the next.
Okay. So once we move up into the application software, that's just a lot of work for OEMs to do. And, one of the areas that we're differentiated is we have the full self-driving application, that's available for our partners. And so if a OEM right now wants to invest in electrification and the base vehicle OS, they can partner with us to do the parking application, the self-driving application, the end cap application for them and this is more about do OEMs want that help or not, but having that capability one, gives us the opportunity to help OEMs in more ways than just the standard component supplier, but it also all that learning that we have in developing this application feeds into our next generation chip and our chips are actually better next generation because we're delivering all this application. We realized, well, if we put this into the chip, it'll even be better for someone building a self-driving computer next time and so our products actually are increasingly differentiated for people who are actually building self-driving cars.
So, we talk about that end to end data flow from training to simulation, the real differentiator for anyone building AG in our mind is the speed of that development flow. And the reason why I say this is building an ML ops development flow that has to be safe and secure has never been done in the industry. This isn't like you're building software in a cluster, or it's an Android tablet or iOS tablet that, if you do something wrong, it's okay. You just later do a software update and it can get better over time.
Building a self-driving computer, you have to make sure that it's safe and secure. So you want to be able to test software in simulation before you roll it out to the car. You want to be able to test the software in shadow mode, inside the car, before you update it in the car to make sure you're not causing any negative experiences for drivers. And to do that, you need a software development flow that has a bunch of interconnected AI loops happening at the same time. The way you build a map is AI loop. You have to take data from the road. You need to train networks off of that data. You need to build a map and then you need to OTA it into the car.
You take that mapping network, but you also feed that mapping network into your perception network for the car, because a map is actually a good way to train your path perception network. And so those loops are interconnected and everything you're doing also gets tested in a SIM loop, as well as a self-driving loop and making that self-driving loop essentially be identical to that digital twin simulator is a lot of work. How you build that software development flow, such that all these pieces are interconnected, safe and secure, and actually designed so that you could quickly find an issue, root cause it, fix it and then OTA it in a car. The speed of doing that is a core value proposition for anyone building autonomous driving software and that's where we're investing as a company is that end to end flow.
We have a DGX systems that we sell to help people with training. We have constellation systems that we help people with simulation and we have the AV software. All of those things have to intertwine together to form out software development flow that's efficient. That's where we're focused on innovating and differentiating and I really believe that that actually is the core source of advantage for one AV company versus another.
So I think here, we just wanted to point out that NVIDIA ecosystem is going through significant growth. There's traditional OEMs, like a Mercedes and a Volvo and a Hyundai, but there's a whole slew of any of these most of which were building a car the way you would want to build it if you could reimagine what a future car would be. Most all of those companies are developing on the NVIDIA platform, whether it's Neo, Li Auto, Xpeng, Faraday Futures, and then all those companies also building robo taxis or autonomous trucks. The vast majority of them also developing on our platform. And I think that ecosystem is also a source of advantage.
And the reason is, is when people develop on our platform, we learn from them. They tell us, hey, there's something here that you could do next time. Can you make your software better in this way? Could you make your hardware better in this way? And we take those learnings and we incorporate them into our roadmap. And, I'm particularly proud of the fact that there's hundreds of companies developing AV and it's not just an OEM strategy. It's a platform strategy. This slide show you guys, OEM partners develop our own platform, but we actually just work with camera vendors and radar and LIDAR and mapping partners and tier twos, tier threes, all of those OEMs, all those ISVs and start-ups, building software for hardware vendors, and ecosystem. They're also bringing up their products on our developer platform and we think that's great.
It's not -- we're not just trying to service OEMs. Anyone's building an application or a product that connects to our computer, we want to work with them to make their application run best on our platform or their sensor work best on our platform. And I think that's a core part of our advantage. And we always talk to you guys about it. But, I wanted to give you guys the background of why we do it and why we think it really helps us differentiate.
And I think we've mentioned this before, but our pipeline is growing and it's across the board, it's not just passenger vehicles, it's robo taxis, it's trucks and it's not just in terms of hardware, right. There's customers who now increasingly want to work with us on software. And so, when we put all those opportunities together, our pipeline is growing significantly and we have more than an $8 billion pipeline now.
Question-and-Answer Session
Q - C.J. Muse
Great. Ali, thank you for the presentation. I guess to start you know, I'd love to kind of level set, where NVIDIA's automotive business is today, $500 million plus in fiscal '21 to what you kind of see in the future, particularly as you look at that pipeline of $8 billion extending out to fiscal '27. So is there a way where we can kind of level set what is kind of legacy infotainment, smart cockpit, L2 plus autopilot other, and how we should kind of see that ramping in the coming years?
Ali Kani
Yeah, we'd mentioned legacy infotainment is less than a third of our business and we'll continue to dwindle away and combined with the smart cockpit, together legacy and entertainment, it's not cockpit is, a little over half of our business today. But, I think what you'll find is that the mix of AV and self-driving will ramp and grow over time. We'll start to see an inflection later next year and, we'll really try to scale it in calendar year '23 and '24.
C.J. Muse
And in terms of that inflection in the second half of '22 what's principally driving that confidence?
Ali Kani
It's really just the wins we've announced. There are a good number of NEVs that had announced products to go to production in 2022. A lot of those China NEV's that we talked about as well as the emerging ramp of a couple of OEMs both Volvo, we've announced is going to be going to production in 2022 and SAIC. So as all of those vehicles start to ramp up, our business is going to start to see that inflection later next year. And then you'll really start it and hit as they ramp up in '23 and '24.
C.J. Muse
Very helpful. I guess just to take a step back trying to get my arms around the hardware compute side of things, at the heart of NVIDIA is compute although clearly software's becoming bigger and bigger part of the company, but we'd love to hear the benefits of Orion coming into production in '22. And then I guess perhaps more importantly, would drive that one scheduled to come online in 2024 what that roadmap we'll be able to support?
Ali Kani
So, I think first, let me say, I think Orion is a home run. I think that just the amount of traction and success we've had with Orion is far larger than what we have with Xavier, and we're really excited about the product. And I think it's because the industry has started to truly appreciate something we've been preaching for a long time. As you know, you really should build a software defined computer in the car. It's not just easier for OEMs to update and support and upgrade that experience over time. But, if you think of it in terms of a software and services business, it's transformative. It transforms the business model of OEMs and, now that we have all these supply shortages and the chip industry OEMs also realize that it's a better strategy for their supply management.
So, there's so many fundamental areas that a software defined computer like Orion unlocks for OEMs. I think that's what's driving the success and so now when you talk about Allen [ph], when you put our strategy together is let's make sure the platform is compatible. Let's build an Allen computer that's one factor and power compatible to Orion, and let's build our software stack such that the API is that people develop on is compatible from one generation to the next.
Now you see how exciting for us Allen is because we let our customers upgrade their vehicles, but not have to throw away their software. And so what you can find is, is faster generation advancements from OEMs, where he's before they build one architecture and they have to keep it for six to eight years because they didn't want to change the architecture, but our products are architecture compatible. And so what they can really do is just unlock new features and functionality because there's a form factor in architecture compatible upgrade available with Allen.
C.J. Muse
Very helpful. I guess maybe to, to extend on that and certainly mobility, you focused on speed of development flow is a key differentiator for the AV market. You spoke to it again today. Was hoping maybe you could dig a little bit deeper into the meaning of that? Is that strictly on the hardware side, software side? Does it play a role and how your customers are choosing kind of tier one, tier two, tier three suppliers. Would love to hear I guess more thoughts around speed and how that is so vital for this emerging market.
Ali Kani
Okay. So let me sort of explain what I'm talking about. I'm talking about the speed of AV development flow. And what I mean by that is, is let's take Tesla as an example. We've been reading about how there've been all these accidents that Tesla has had. If you go analyze those accidents, there's more than 10 of them, almost every one of them is an emergency vehicle with its sirens on. It's a parked emergency vehicle. Like every one of them is a parked emergency vehicle. It could be ambulance, it could be a police car, but every single case, it's a parked emergency vehicle and the car just hits it. And the first accident was two and a half years ago. And the most recent one was a couple of months ago. And so -- and there is no existence proof of any company who is building AV code, autonomous driving code that they upgrade and update that experience over the life of the car in a safe and secure way.
Tesla is doing, and Tesla is doing OTs, right? But I'm saying that there's all these accidents over the last two and a half years, they're the same accident and it's still not fixed in their cars. So the speed of the develop flow talks about how do you build a self-driving software stack that when you find an issue, you actually can root cause it truly root causes, there's so many things in a self-driving vehicle stack. What was the problem that caused that accident? How do you go solve that problem? Sometimes you need to go get new data, train a new network for that data and test it and then put it in the car, test it in shadow mode and then OT it in the car.
How long does it take you to go through that loop? And that capability and expertise is hard because no one has done it for AI code when safety and security is also critical and what I'm saying is, is look at, look at the challenge Tesla has had doing it. Even though they give us OTAs, I have the client, they give us OTAs. The car actually does drive better, but they don't solve all the problems. There's actually bugs that they still haven't root caused and fixed. And the fact that it's taken two and a half years is really not acceptable for a self-driving car because people's lives are at stake.
And so this flow, this development flow, building it in a way that you can actually do this efficiently requires an understanding of the infrastructure and tools needed for a new kind of software development. There's very few companies in the world who actually even know how to do that. And for NVIDIA it's something where unique, right, is that all these companies that are training their eye networks are training them on DGX systems.
And then we've built our own simulator based on Omni burst to make it such that when you simulate AV, it's actually truly accurate. You can actually train networks on simulated objects that we built. We can't do that on like an open GL simulator. You have to build a new kind of simulator. That's a part of the software development flow and you need to reinvent and recreate a development flow built for AB driving. And that's why I say it's so hard and having the expertise to do that is really not something possible for many OEMs and tier ones and so it's really good to partner there and sort of build-up that expertise and I think that's where we really try to differentiate our partners.
C.J. Muse
So you spoke a bit around working with tier twos, tier threes, and I guess was curious to hear your thoughts around what you think the landscape will look like when we're full autonomous. Is this a world where you're going to be kind of partnering with best of breed, kind of radar, LIDAR vision and coming together with full kind of hardware solution and force maybe tier ones or twos to become more specialized or how do you see kind of the future here?
Ali Kani
I think for us, the key is we really want to design our platform to be open and modular so that if an OEM and a tier one wants to develop their application in their own ways, they can. And so for example, there are cases like Zenseact and Zenseact is due to the builds, the self-driving application for Volvo. And so we just let them build the self-driving application on our platform and they pick what they want, what sensors, what radar they develop the self-driving application themselves. And we support them in that way. So our platform is open as pick your own sensors, pick your own AI algorithms, either in some cases, partner with a tier three or build it all yourself if you're the tier two, they can make all those decisions and we're very supportive of it.
And then, we just have other cases where some OEMs might want a little bit more support from us. And so this is why I say our platform is designed to be modular and open. The first example is we're open if you want to build most of it yourself, but it's modular such that if let's say in the case of Mercedes-Benz, you actually want us to give you the self-driving application. And so then we kind of sort of help inform them, well, then let's we want to invest on these sensors and we'll build that software. So we kind of are more involved in what happens. But the platform is designed to be modular, such that it really depends on the audience. So I don't think one way is the right way and the other way is the wrong way. I actually feel like it's just based on where does an OEM want to invest?
Some OEMs should maybe focus on electrification and Core OS today and worry about self-driving application development maybe later, or maybe never, right. And so we try to have a solution from them but if you actually can do it, I think we're perfectly fine saying, let the tier two and the tier three and tier one define what they want to do and just develop on our platform. And, I think that's great business and we want to make sure we can support those customers.
C.J. Muse
Helpful. A lot of excitement around your partnership with Mercedes-Benz, so curious if there's any sort of update that you can share in terms of recent activity here and timing of ramp would drive AGX.
Ali Kani
I think there's no update. I think it's just, we're working hard together. It's a great partnership. We are trying to build a software-defined vehicle that's fully upgradeable over the life of the car. We think it's going to have some of the richest parking and self-driving applications in the industry and, we're designing it such that we're going to make it better. Customer's going to buy the car, but what the car can do after it goes off that lot will improve over the life of the car. So I think it's a really exciting partnership and we're really excited to be working with them beyond.
C.J. Muse
And I guess more, yeah, more broadly speaking, what are the most critical designs that we should be watching perhaps from a magnitude of revenue contribution perspective over the next five years?
Ali Kani
Well let me sort of break it out in a few areas. I think there's, of course the Mercedes-Benz engagement is something you should look at. It's going to kick in, in the back half of that range, right. When a car goes to production in 2024, it does take years to ramp into the full, two and a half million vehicles that they have in their fleet, but it will ramp into an entire fleet. And then I think you have some OEM designs, traditional branded designs from Novo and SAIC that will also be ramping and will be significant.
Then I would say there's a couple of buckets that you should sort of put together, and there's all these NEVs and they ramp starting next year and there's a lot of cars that they're selling and there's multiple computers in their car, right? Neo announced four Orients per vehicle. And, many of the others are multiple Orients per vehicle. And so, once all of that scales up, it's a sizeable and attractive business that we'll be scaling.
And then finally, it's all those self-driving vehicles. If you look at robo taxis and you look at self-driving trucks, almost all of them are developing on our platform, and that is going to kick in and the volume when you put all of the ecosystem together and the ASP's of course, for robo taxi or not, they're not eying a single Orion. They're buying like multiple discrete GPU's, our highest Stan Ampere GPU's. So there's in some cases, more than $10,000 of content per robo taxi for NVIDIA and when you take that and you multiply it by a large volume base, I think that's another big sort of inflection point that you'll see in the years to come. So I think all of those will ramp in, in this timeframe that you're seeing and that's why, the pipeline actually is growing and healthy,
C.J. Muse
Really helpful. I know, Jensen has a focus on driving recurring revenues and just curious, I obviously -- there's a hardware and software aspect to your focus, but is there a priority as you think about pricing to kind of driving that razor blade model or are you more kind of indifferent in terms of pricing hardware versus software?
Ali Kani
Yes. I think we just try to charge based on the value we think we're bringing to the table. So we just think of that like that. And, I think there's recurring revenues in multiple ways. Of course, if you build an application and when a customer upgrades a feature or a function after they buy the car, then there's recurring revenues from that. But even when we don't build an application, when you're building a production vehicle that has to be safe and secure, we need to maintain that release over the life of the car and so there's fees there.
Every OEM that goes to production with safe code, we almost need to branch that code for them and give them a stable branch and so there's recurring revenues there as well for all our customers. And so, I think that recurring revenue stream based on software and support, will pick up as you know both the software application customers, as well as just traditional chip and OS customers go to production because we'll need to continue to maintain and support those vehicles for the life of those cars.
C.J. Muse
Pretty helpful. I guess maybe we'd love to hear your perspective on the competitive landscape, obviously, Intel Mobileye. You now have Qualcomm with being here. How do you kind of see how you kind of see things and how do you see things evolving, over the next five, 10 years?
Ali Kani
I think the market has a good number of competitors who are capable, who have reasonably good strategies. And so I feel like it's good for the industry. There's, good options, good choice. And, we just try to differentiate with our strategy, right? Like we're unique in that first, our platform is open. You can develop on our platform if you're not an OEM, you can just be at tier two and tier three, and we take all those learnings and we improve the software and the hardware and the next generation because of those engagements.
And the second is, is that we really do have solutions on the infrastructure side and like I said, that back end the flow, we think is actually a key differentiator for anyone building autonomous driving. So helping customers with a training solution or a simulation solution, that's kind of unique to NVIDIA. And so having that end to end platform, especially with differentiation and infrastructure really helps us differentiate.
And then we have cases where customers want to build the self-driving application of some more traditional way. You mentioned linear [ph] and Mobileye. Their approach is a little bit more end cap, fixed function CV based AV. Our approach is a little bit more deep learning software defined computer. We designed our platform to be programmable and be very heavy on deep learning and it's different than billing an un-captured. It's very fixed function.
I've never seen a Mobileye product be OTA in its life. They've been in production for so many years. They'll try to OTA car one day, but I'm just saying it wasn't built to be software programmable and software defined. It's more of a DSP type architecture. So we're trying to approach it in a different way. And so, I think where we're investing and where we're differentiated are the places that OEMs are going and it's the place as they seek differentiation. So I really like our positioning and I think the results show with all the wins that we've been announcing.
You haven't seen too many new wins from other people, right? Like it's pretty much been a bunch of NVIDIA wins. The one area where I think you'll see more competition is in level four, level five, I think Mobileye will have some level four, level five ones. But our strategy is a little bit different. There is we want the Zeuxis and the cruises and the Argos and those guys to sort of be the leaders in that industry and they clearly are.
And I think what Mobileye does is they go to people who can't build it, and they just kind of build them on the platform. And, our strategy is a little bit more Hyperion there. We'll give you the hardware platform. Then you build the application, you build the AV. I think our strategy is the right one for robo taxi and this cycle and then longer term, we may have an AV solution that can help people in that space, right. It just takes time where we'll focus on L2 and L3 today but, L4 and L5 can come in the future.
C.J. Muse
So we've got roughly two minutes left. So I will leave that time to you in terms of perhaps summarizing or if I failed to ask anything that you think is important and relevant and lets you kind of conclude our half hour with investors. Can you hear me?
UnidentifiedCompany Representative
Yeah. I don't think Ali can hear you. I couldn't hear anything.
C.J. Muse
Can you hear me now? I can hear you. I can hear you. Okay. So we've got a minute left. I was just going to turn it over to you to either share something that we didn't touch on that you think is important or perhaps just give a quick summary and ending thoughts?
Ali Kani
No, I think you've asked really good questions and I think you covered most of the things. I think the one thing I'll say is that the automotive industry right now, the amount of areas that we see fundamental transformation are significant. It's not just in self-driving cars. It's also in electrification. And I think one thing I just wanted to say is, I think one area we haven't yet seen significant investment, which I think is going to be exciting is inside the cockpit.
Today a cockpit experience is about cluster and infotainment. Those are legacy applications and you can meet those needs with multiple solutions, but long-term the experience in a car, it's going to be kind of like your concierge, but AV is like your chauffeur driving the car for you, but you actually will have a concierge in a car that's extremely intelligent and personalized to what you like to do, and you can interact with it, you can talk to it, you can enjoy shows, you can do video conferencing, the amount of AI that's needed to enable a true concierge experience is significant and we're not yet seeing in the car. And I think that that's an area where we'll see more innovation longer term. And that also is an area where NVIDIA is unique in being able to provide some of those experiences.
It's not self-driving software. It's like AI software, it's video conferencing software, speech, ASR, all those kinds of technologies and putting them together with driver monitoring and occupant monitoring and fusing it with the AV experience outside the car, such that if maybe your child is about to get out of the car and there's a car coming from behind you to not let that happen, or if you're about to leave a car and you might've left your purse or your laptop for the concierge to tell you that your wallet or your purse is in the back, and do you want to leave it in the car? So much innovation is possible by fusion fusing AAV with AI inside the car. And I think that's going to be an area for differentiation term that we're excited to work on today. We don't have success to mention yet, but I think it will be an area where we're going to have some exciting opportunities long-term
Can I get Jensen's oven inside my car? Well, Ali, thank you so much for today. It's always great to spend time with you. So, thank you and wish you a great health and we'll close there.
Ali Kani
Thank you, guys. Appreciate it. |
