NVIDIA Corporation (NVDA) Earnings Call Transcript & Summary

We're good. Okay. Hi, everybody. So I'm Joe Moore from Morgan Stanley. We have NVIDIA virtually with us here. Want to first read our safe harbor. Please note that all important disclosures, including personal holdings disclosures and Morgan Stanley disclosures, appear at the Morgan Stanley public website or at the registration desk. So I think we have Colette online. And Colette, just to set the stage for you, we have a room full of people, pretty packed house. So I, in no way, feel self-conscious about sitting by myself on stage. Think a Republican convention. But anyway, glad you're able to join us today virtually.

And maybe we could just start out with -- the question is on everybody's mind, the coronavirus impact. You guys, as one of the more recent companies to report, actually gave numbers around that, $100 million impact. Can you talk about that impact? And what's causing disruption to your business? To that extent, obviously, still better than consensus expectations, but some disruption. And can you talk about what the variables might be going forward? And by the way, just -- can you hear me okay? Okay. It's...

I do. I can. I can hear you.

Okay. Perfect. Yes.

Okay. Right. So thank you, Joe, for holding this [indiscernible]. Let's start out with [indiscernible] overall [indiscernible], that's probably one of [indiscernible]

Okay. Great. And you had some news today that you postponed the investor meeting, and you're going virtual for your develop -- the GTC developers conference later this month. Just anything you can say about that in terms of just clarity and maybe any idea when that investor meeting will happen or if that'll be virtual as well.

So later this month, our schedule for GPU Technology Conference is virtually held. [indiscernible] So we are still holding the [indiscernible] but also [indiscernible] over the next few days. [indiscernible]

Okay. Great. In terms of -- maybe we could dig into some of the businesses. In terms of gaming, you guys have gotten back to year-over-year growth as of last quarter. There's obviously been a lot of moving parts in that ecosystem with cryptocurrency impacts a year ago and things like that, but it seems like you're on pretty firm footing now. Central to your value proposition has been ray tracing. Can you maybe talk about where we are with ray tracing adoption and the growth of your gaming business overall?

Okay. Thanks for the question. So with our overall gaming [indiscernible] they have all adopted [indiscernible] overall ray tracing [indiscernible] additionally [indiscernible]

That's great. And can you maybe just give us some historical context? I mean you guys have had these major investments in gaming before with Shader algorithms. There's always a bit of a chicken-and-egg thing where you introduce these breakthrough technologies, and it takes a while for software to support it. How does has played out relative to those prior cycles? And what we see with -- when we have new hardware, will there be a big jump forward in ray tracing? It just seems like this is a very important investment, and maybe it held you guys back a little bit from a growth perspective as that was starting to percolate. And you're getting to the point now where you should start to see the benefits.

Yes. So it was definitely a chicken-and-egg model. In order for the overall algorithm [indiscernible] you bring the overall hardware [indiscernible] both attracting the overall gaming developers but also the [indiscernible] underneath [indiscernible] overall ray tracing. We entered into a market where probably some gaming [indiscernible] half way through development [indiscernible]. But now what we are seeing is a true force where we focus [indiscernible]. When you think about the next holidays, [indiscernible]. We're excited that we have [indiscernible]

Great. And Colette and the AV people, I guess, we're getting reports that the webcast is very difficult to hear. So I don't know if there's anything that anybody can do about that, but just if there is anything you could do to help the audio quality on your end, that would be great because they don't want to just hear me. That would be particularly useless. But anyway, moving forward, the mobile gaming business is another business that you guys have been pretty positive about. You've seen nice growth there this year, kind of establishing these notebook form factors with discrete GPUs in them. Can you talk a little bit about that? Where are we in the adoption? And how much growth do you see looking forward from mobile?

Yes, thanks so much. So when you think about our notebook business for gaming, it is really an important part of our overall gaming business as a whole. Many years ago, looking at overall notebook, it was never [indiscernible] our notion [indiscernible] when you think about high performance, but also in a powerful [indiscernible] only overall notebook. So what we see [indiscernible] here is the ability for us to [indiscernible] our own Max-Q technology into high performance in a thin-and-light form factor. This allows the notebook to essentially double [indiscernible] to double-digit year-over-year growth for 8 consecutive quarters as we incorporate Max-Q technologies. So Max-Q technology has -- I'm in no hurry to give up that overall thin-and-light form factor. I still want the high performance [indiscernible] on my desktop. And we [indiscernible] that [indiscernible] within the notebook. So over the past [indiscernible], we had a record number of overall notebooks. We've had more than 125 models out in Q4 versus about 94 in the year ago. And the Max-Q model are 2x year-over-year over this time. We expect notebooks to continue to [indiscernible] going forward as well.

Great. And another category that you guys have invested a lot in and talked a lot about in these various developers' forums is content creation where you're focusing on consumers who are becoming content creators around YouTube and Twitch and things like that. Probably a harder business to quantify since a lot of the products may look like traditional gaming products. But how do you feel about those investments? And do you think that, that's a growth category we should consider?

[indiscernible] very important project [indiscernible] in terms of who we refer to as the creatives. These are independent content creators and a large and growing market that represents probably 10 [indiscernible] so essentially [indiscernible] with not a lot of options of high-performance notebooks and/or [indiscernible]. We've targeted this with a [indiscernible] for Studio [indiscernible] which is now supported by more than 60 different systems through desktops, notebooks and workstations in the market and with different algorithm. The engineering roll out a new model at CES launching our 13 RTX Studio's [indiscernible] or partners and [ AC ] was the very first RTX-based all-in-one [indiscernible] as well. So we're optimizing for over 55 different creative and design apps [indiscernible] significant amount of time both in encouraging everything for the product catalog to [indiscernible] as well as [ ACC Adobe Cloud. ] So we feel this is really an untapped market. We're in the early stages, but so far, the awareness [indiscernible]

Great. So maybe if we could just talk a little bit about the seasonality of the business. You guys have talked about, I think, on the call, normally, a double-digit decline in January, and you said maybe it's a little worse than that. It's like it's a pretty good outcome given the challenges that you're dealing with in one big geography in China. Can you just talk about how you see that seasonality playing out this quarter and beyond?

Sure. So the gaming market is essentially a seasonal market, which tends to be a lot of [ mature ] purchasing in overall holiday season. So essentially what we [indiscernible] end customer markets are [indiscernible] But we also added another [indiscernible] in the last [indiscernible] or so, and that's associated with the overall growth that we're [ seeing ] in our notebook business and our overall consoles that we serve. In order for [indiscernible] overall -- reach the overall market, we find that our Q2s and Q3s tend to be the peaks of those as they are building, with the overall OEMs for them to hit the overall market for the overall holiday. So therefore, what we've seen is our Q1s are -- and our Q4s are essentially a downtick in terms of our overall seasonality. So what we saw in between Q4 and Q1 as we roll to the end of the overall holiday season, plus a little bit more pronounced due to the overall coronavirus, we have seasonally down in Q1. But other than these overall dynamics, when you look at things from a full year perspective, we still see the growth across all of these markets.

Okay. Great. That's helpful. And then in terms of competition on the gaming side, you guys tend to talk about new products when you're shipping. Your competitor talks a little bit about them before they're out. But -- so maybe just kind of help us -- I mean we know they have products in the second half. We don't know about yours. But generally, your confidence in the competitive outlook and your commitment to retain that high-end leadership.

Yes. So thanks for the question. We generally keep our competition top of mind, and we do respect a lot of work is going on both internally to focus on the competitive nature of many of our competitors out there. Right now, we continue to be the market leader in both features and performance. 1.5 years ago, we heard and understood that many of our competitors were focused on 7 nanometer and bringing that to market. And although we are still on a node at 12 nanometer, our features, our performance and our overall performance to cost is still tremendously better than the overall competition. We're pleased with our overall performance through the holiday season, even though the competitor came forth at overall 7 nanometer through here. We'll continue to focus not only on just the overall design of the overall hardware. We do know that there's a significant amount of work that we can do with our overall ecosystem, with our overall gaming mix ecosystems as they build out games, particularly for ray tracing, to both improve the performance with overall software and improve the performance of the overall games. So we have done this in the past. We will continue to do that. So when we think about our engineering and we think about our innovation, our innovation doesn't just stop at the overall chip. It works all the way through the overall ecosystem, and we'll continue to bring things such as ray tracing, the AI algorithm using DLSS and other features to the market as quickly as possible.

Great. So I want to shift to your HPC cloud business. And maybe -- I know there's a bunch of near-term drivers that I want to talk about. But maybe if we could step back and look at what's happened in the last few years. You grew this business tenfold in a fairly short period of time. And then you saw this period of digestion last year that was probably more severe than I would have expected it to be, and now you're back on pretty strong footing, growing 33% quarter-on-quarter this quarter and seeing good growth next quarter. Can you put that into a context of -- that digestion, is that something that is just part of HPC cloud that you think we'll have to deal with? Or were there sort of NVIDIA-specific drivers where you feel more confident that it'll be more steady going forward?

Yes. So when we look back at last year and particularly in the first half of 2019 calendar, we and many others in the industry were affected by that overall digestion period. What we faced was a period of time that we were able to concentrate working with the individual engineering teams at many of our customers as well as our own engineering team working on building out, expanding overall workloads to take advantage of the overall AI growth as well as overall new workloads that we see coming around the corner, which led us to the overall summertime. The overall summertime was where we heard of many hyperscales coming forth on the work that they were doing on natural language processing and natural language understanding. The BERT model also surfaced during this time. The important aspects of these new techniques is really a focus on even larger and larger data sets, more complexity in those data sets in sometimes over 100x to some of the data sets and compute that they were doing before. This has now increased both a new wave of workloads following this understanding. We're seeing conversational AI be a big part of the overall purchasing that we began in the second half of 2019. We're also seeing recommendators as well as deep recommendators be a very important part of hyperscales as well the consumer Internet companies. We expect, as we go forward, for this to continue as we're still in the early stages of conversational AI and what that could mean, not only to the overall hyperscales, but also to the overall enterprise. When we think about our guidance for Q1, for example, we think our year-on-year growth in data center business will accelerate further in Q1. And looking over the next few years, we see AI to probably be one of the largest opportunities in front of us. And we're in a pretty good position to lead that.

It's really interesting that -- I mean when we talked about this 18 months ago, a lot of the conversation was around computer vision and autonomous driving. And I know those are still important applications. And now we're transitioning to more language-based, and I want to dig into that a little bit more. But did we see a pause in those vision applications? Was that part of the digestion that we saw? Or do you think it was more just kind of the sort of more -- I mean everyone in the industry saw some form of digestion. But was there some application-specific aspect where we're having to hand off from vision to language as kind of a key growth driver?

No. It's not necessarily a handoff in terms of a stop of overall computer vision. There's many cases where even with the deep recommendators, you can still have overall computer vision incorporated in there. And if you think about even both from an online purchasing, online shopping, how do I search and find for the things that I want, the computer vision of images and/or live overall film is extremely important. But what you are seeing is a expansion of using the BERT model from just the basic techniques that they were using in terms of how they design search. So search commands previously really just took what we'll refer to as the nouns in terms of how individuals spoke or how they actually wrote and really ignoring maybe some of the additives or some of the propositions around that. So the overall BERT model really looks at the techniques of -- the word choice built before and after those nouns really, really changes the meaning of how people are searching and looking for overall answers. You can see this already when you see the results of overall search commands, whether they be spoken live or whether the search that comes out in terms of written form. They are really gearing that the answer to your question is probably in the first one or first 2 clicks that you may do. This is a real transformation in terms of how search commands do, but it does require a significant amount more data and computational work in order to do that, both on the training side, but also in terms of on the inferencing side later and the use of overall inferencing and inferencing using GPUs to do so.

How does -- I mean one of the things that got us pretty excited looking at conversational AI was just the detail of these models. And really, if you look at what they're doing with BERT, transformers, there are tens of billions of parameters versus tens of -- maybe tens of thousands for some of these vision-based designs. How does that translate to your business? Does that -- obviously, your silicon is moving forward at a rate that also is much faster than Moore's Law. But does that imply a larger number of GPUs to solve one individual problem? Or is it more that this is just a more pervasive technology set?

It's a little bit of both. Both on the first side of the larger overall data set, yes, you are stringing together more GPUs together, more systems together in order to compute that. You actually are even going back through even some forms of retraining models as you choose to train models jointly together, also improving the overall demand for overall GPUs. But these use cases are also more broadly adopted and can be adopted not only in terms of in the hyperscales or with the consumer Internet companies. There is great use cases in terms of what we are seeing in terms of with enterprises. In cases in terms of natural language processing, you can see enterprises that focus a significant amount of their energy in terms of -- on call centers, that they can massively improve that workload and the overall customer experience if using GPUs to improve the overall language understanding, both from the customer as well as from the corporation itself.

Okay. And you've been talking quite a bit more as well about these recommendation engines. And I guess that's less intuitive to me because they've been around for a while, but obviously, deep learning approaches are changing that. Can you talk about what's changing such that, that's a bigger aspect of your business? And what were the complexity of those models looking forward?

Yes. So what you had seen in the past in terms of recommendation engines were really these engines use content filtering to match users with items. But there was limitations to this approach. Currently, right now, what you see in terms of recommendation engines is they are really understanding from all of the different sources that you look at to get a better understanding of who you are and what you would like to see in there. These new engines really make deep learning and make the overall results more relevant to them. For example, Alibaba and others are seeing click-through rates as up to 10% increase in terms of using overall GPUs for this overall case. So it's an important area that has a lot of room for improvement, but the use of GPUs and merging together a lot of the things that you both read, both look at and overall purchase to help recommend better things to you.

Okay. And then I'll have 2 more questions on HPC cloud, and then I'll see if there's questions from the audience. The question I get probably the most frequently for you guys in HPC cloud, in addition to the growth drivers, is competition. And I think every time you come here, we talk about custom silicon from cloud partners. We talk about a dozen high-profile start-ups, which have sometimes changed from your view, which ones they are. We have products from several of your competitors. And it seems like you guys are still doing pretty well in these workloads, and you have a product that's almost 3 years old in this space. But can you talk about the competition? And do we -- how should we frame that competition in the context of -- some of your big customers are making big investments to try to complement what they're doing with NVIDIA. And some of your big competitors are making those investments as well. So can you just talk about that competitive dynamic?

Sure. So the competition out there is likely still in their overall early stages. The early stages were maybe not necessarily ready for the big market that we see in front of us for overall AI. Our success today has been establishing a platform that can universally be adopted in many different types of workloads from training to overall inferencing, high-performance computing, supercomputing as well as the significant amount of new enterprise workloads that we now support. But the key thing is not just the overall hardware platform but the overall software that we have enabled through this. Many years ago, more than 10 years ago, the development of overall CUDA on top of each and every single one of these GPUs and the same CUDA or software, though it is at version 10 plus, allows us to hit a set of developers that they can continue to write on top of these overall GPUs for the expanding AI workloads that we see. One of the challenging things that people have with new overall custom ASICs is they are not programmable. Or if you are looking for some form of FPGA or other forms, you are talking about something that's configurable and becomes extremely challenging to overall design for the overall workloads at a software level. Our ability to be a broad-based overall platform, that we now have more than 1.7 million developers using our overall platforms, allows it to continue to grow and reach an advanced use of overall workloads. People come to us to seek advice on the best solution. They come to us looking for advice on the best architecture. And they help us in terms of designing new features in terms of in CUDA and/or in terms of in our underlying chips. Our platform is very well-known and takes a leadership that allows us to really deal with the overall competition that's looking for a single overall workload to adapt possibly their hardware that they develop. So we feel confident in terms of where we stand. Yes, we understand that there's a lot of ideas out there for overall competition. But so far, we haven't seen much of it affect our current position in the workload.

Great. And then the nearer-term visibility and confidence that you guys have, I mean, a quarter ago, you were able to say that the January quarter will show significant growth in HPC cloud. When you reported this quarter, you're saying the other businesses are down seasonally double digits. So there's a very clear kind of double-digit type of growth in HPC cloud. I feel like in the past, this is a business where you maybe haven't had that kind of visibility to talk quarter to quarter. Is it just the breadth of the application side? Or what is it that kind of gives you the confidence to be able to give us a little bit more directional guidance on HPC cloud from quarter-to-quarter?

Yes. So good question. We've been really focused on our business and getting our products to market with each and every of our customers as effectively and efficiently as possible. This allows us to be working with them at an engineering level and an architectural level and helping them understand that the more high level the platform that we are building, the longer the overall lead times are. So from end-to-end, our work from an engineering standpoint to helping them work in terms of the qualification of these products with inside their infrastructure is very key. That allows us to provide that visibility that we're seeing today. We'll continue to look for improvements on that efficiency of getting to market, but it is really starting to take off and improve as we see today.

Great. So we have about 8 minutes left. We'll see -- I have a lot more questions if we need to, but see if there's questions from the audience. Turn on the mic here.

Vishal Patel, 1832 Asset Management. Colette, can you just give us a sense of how we should think about NVIDIA's exposure to crypto markets? And what have you learned about crypto markets over the past couple of years and your exposure to those markets?

Sure. We have had exposure to crypto markets in the past. However, at this time, I don't believe we are seeing any exposure to the crypto market as we see today. A pretty hard time to both understand historically the impact of that, a very volatile and vibrant market that was very hard to manage, both as it was rising as well as when it was leaving. But right now, we are not seeing any impact from crypto today.

Questions from the audience? Well, raise your hand if -- wait, there's a follow-up here.

Just a follow-up question. You're generating a lot of cash. How are you thinking about capital allocation? And you have a lot of opportunity sets, whether it's in auto or data center. And how do you think about capital allocation, return on investments for all the cash flow that you're generating? And then whether it's dividends, buybacks or reinvesting in the business?

Sure. So you're right, we have a little bit of cash. We have over $10 billion, and we're going to use a really good portion of that to purchase overall Mellanox when we reach regulatory approval. After that, we will sit back. We will look at our overall investments. And we focus on making sure we have enough cash for internal overall capital that we may need. We will focus on M&A if we have a need for overall M&A. And then we'll look in terms of any excess cash that we have in terms of distributing to overall shareholders through stock repurchases primarily, but also through our dividend.

Other questions?

Can you talk about your biggest [ competitor ] in the self-driving car and your competition relative to Mobileye of Intel?

Sure. So from a self-driving car and our focus in terms of -- on autonomous vehicles, we are focused on an end-to-end platform that helps serve the overall automotive industry. That starts all the way back in terms of the work that they do, in terms of the data center, in terms of collecting data, testing data, writing overall algorithms, but also working on what they may need for the overall production inside of the car. We have the true platform today in production that allows them, if they wanted to take AV to market today, we have a full hardware platform that they can incorporate inside of the car. Different than our overall competitors at this time, who are looking in the next couple of years to overall bring that to market. It's essential to have a sampling today. The auto manufacturers need to be sampling and also spending this time testing, incorporating the safety and writing the full set of software. We sometimes help them with that overall software development with our overall software. We are also even helping them with the testing as we always have test vehicles as well. So our overall focus strategically is truly on autonomous vehicles, end-to-end, whether those vehicles be passenger cars, whether they be robotaxis as well as an important transformation of the overall trucking industry. The overall automotive industry, though, right now realizes that this will probably take a little bit longer than they originally expected, and we'll probably see large amounts of production hit in probably a 2-year time frame from now.

And can I just follow up on that real quick. You've talked a little bit about Level 2+ as part of your strategy where -- as differentiated from sort of full autonomous. When you think about revenue 2 years out, is it at that Level 2+ kind of capability? Is it full autonomous for robotaxi applications? Just where do you expect that revenue to come up first?

Yes. The revenue in the future will come from both. So it will, in that 2-year time frame, come from both the passenger vehicles and the robotaxis. The robotaxis, given it'll be full autonomous, will probably have a higher price point. And then in the Level 2+, we hope to get to where we'll have overall platforms that can take them further to Level 3 and above based on the overall hardware that they may consume inside those platforms. But we do believe we'll be able to interestingly go to market with both at that time.

Great. We have one more question.

Colette, given the coronavirus issues, could you comment on if you are having any issues in securing manufacturing capacity at your foundries?

Yes. Our foundry work right now is at TSMC and Samsung. And so those are long-standing capacity needs that we have procured. And at this time, no, we don't have any concerns.

Great. We may have one more question at this time. Maybe I could just ask about inference then. You guys have built a pretty sizable business now around inference. I think you talked about a fourfold growth this last quarter, and yet you still seem to be at a relatively early stage of penetrating the machine learning inference market, which has historically been mostly on CPUs. Can you give us a little context around that?

Yes. We're actually extremely pleased with our success in our inferencing workload. We use our G4 product to reach our overall inferencing workloads that might be out there. A couple of years ago, more than a couple of years ago, we indicated that we were going to be entering this market, which you speak about being primarily an overall CPU market. We're coming with an overall GPU that has the ability for programmability at the inferencing and really addresses the new overall inferencing that we see going forward. The new inferencing is a much more compute-intensive type of inferencing. If you even think about natural language processing, that has possibly 3 to 5 steps from speaking first to actual -- all the way around to responding in speech as well. The important piece of that is the performance needs requires a 300-millisecond-or-less response from a spoken overall search command. Many, if not any other processor, is unable to serve that overall market, and the GPU is just extremely well positioned for that. So we're seeing advanced overall inferencing. And we have established -- where more than solidly into our double digits as a percentage of our data center revenue is focused on inferencing. Even while we are also seeing record levels in terms of our V100 and overall training workloads as well, we are still well into the double digits of inferencing. Inferencing has grown 4x where it was just a year ago. So yes, we're still in the early stages, and we're still at a very, very low overall market share of the overall inferencing that is available, but we continue to focus on applications that can be accelerated, applications that can leverage using our overall inferencing platform to solve those workloads.

Great. That brings us up to the end of our time. Colette, thanks so much for joining us under trying circumstances. And thanks, everyone. Enjoy lunch.

Thank you.