NVIDIA Corporation (NVDA) Earnings Call Transcript & Summary

Okay. I think we're good to go. Thank you all for coming. Big crowd, nothing less expected from NVIDIA. I'm Toshiya Hari, I cover the semiconductor and semi cap equipment sector at Goldman Sachs. Very pleased, very honored to have Colette Kress, SVP and CFO from NVIDIA with us today. I will kick off with some questions, but certainly happy to take questions from you all. [Operator Instructions] Colette, first of all, thank you for taking the time.

Thank you for having us. Appreciate it.

I definitely want to spend more time on longer-term strategic questions, but quite a few topical things going on in the world today. So I was hoping to get an update from you on the near term. Obviously, you're going through an inventory correction in the gaming business, you've got a weaker macro backdrop, which is for everybody. And then more recently, announcements around export restrictions on high-end data center GPU. So you can spend 40 minutes on that. Hopefully, you don't, but if you can kick us off with some opening remarks, that would be helpful.

Sure. Let me just start off with, we had finished our Q2 results. We did preannounce them and then provided an outlook for our Q3. Our Q3 outlook of $5.9 billion, correct, has in there a -- can you hear? Can you hear me? All right. I'll try and speak louder, is this better? Okay. We'll try that out. We provided guidance for our Q3 quarter that we provided $5.9 billion. Within that would be a situation that we will do an inventory correction, in our gaming, but also our pro visualization business, both our desktops and notebooks. That means that we will be under selling into the market as we work through that inventory that is in the channel. That will be a sequential decrease from Q2 to Q3 for those businesses. But we had expected a growth in our automotive business, similar to what we had in Q2, moving to Q3, as we're continuing to ramp our Orin architecture on many of our passenger cars and then future robo-taxis, as well as AI cars as well. Our Data Center business solid. We expect also to have solid results. However, we have a new risk that we articulated just shortly after we gave the guidance regarding the U.S. government had put an additional restriction in terms of what we can sell of our A100 into China. That risk is about $400 million. We are working right now with our China customers to provide them alternative products versus the A100. And we expect that to be our best path to assist them. If they also want a license, we can do that. But right now, we're working with them on alternative products. So at this time, though, we still have that risk. We're going to see how the quarter develops. We've still been processing working with the U.S. government on that restriction, and we think we now have a path for them. So those are the things in the most nearest term that we're working on in Q3.

So Colette, is it fair to say that what's incremental since you gave guidance of the export restriction, everything else in the business is more or less tracking, I don't want to put words in your mouth but...

That's correct. I don't have any additional update at this time to the rest of the business. The 2 days after giving guidance for the exports issue, but we hopefully -- we'll be able to work that through in the next couple of quarters as well.

Okay. I definitely want to come back to that, but I had a couple of questions on the gaming side, if that's okay. Given what you guided for the October quarter, assuming you come in in-line with your guidance for gaming, implicit guidance anyways, you will have declined around 60% in revenue in gaming peak to trough, which is a sharper decline than what you saw back in 2019. So the hope is you've sort of de-risked your gaming business. But on your call, you talked about sell-through for the July and October quarters collectively being about $5 billion. I'm curious, how did you solve for that $5 billion estimate? Has your ability in tracking sell-through or real demand improved over the past couple of years? And most importantly, based on what you know today, at what point do you expect to be selling in, in line with sell-through in gaming?

Yes. So let me, for the full audience, give them an understanding of what we provided when we talked about a normalized demand on our call. We talked about a normalized demand looking at our sell-through or estimated sell-through that we see for a couple of quarters. We used our Q2 estimate that we had at the end of that quarter, it was an actual, and we made a thought in terms of what we're seeing for Q3 and put those together to come up with $5 billion of end demand. End demand, meaning what is our end customers buying from the channel. Those numbers a little bit in Q2 with a small rise as we move sequentially into Q3. A muted overall growth, a muted as we get ready for that back-to-school but also for the holiday. That, I felt was an important understanding that, yes, demand for GPUs is still strong. There was just too much inventory in the channel. We're going to be watching that carefully this quarter and further, but that was our overall goal in providing that demand and helping folks understand that the GPUs from a pricing side have reached about the normal levels that we had hoped. We took action immediately to also reprice price programs to move this inventory through. We believe this is probably going to take us a couple of quarters to get through. Maybe at the end of Q4, we'll be back at that normalization. We'll start that positioning going forward where sell-in would be stronger than the sell-through.

Okay. So October and January still muted sell in, perhaps below sell-through? And then April, maybe you're kind of back at the sell-through levels...

Yes. We hope so. We'll take a look at it. We don't have any guidance at this time for Q4 and further, but we're watching it carefully.

Okay. Got it. And then on the pricing dynamics within gaming, we got a bunch of questions from investors about your ability to grow ASPs and pricing gen-to-gen, which has been the normal course of business for you over the past, I don't know, how many years. As we think about this rumored intro of your new product line over the next, call it, 12 months, how should we think about, again, your ability to grow pricing? I guess the bear thesis that we hear is your ASPs were -- or your mix rather was inflated due to crypto over the past 2 years. That was one time and perhaps next cycle, you're in a disadvantageous position, if you will, but how should we think about that dynamic?

So let's get some background of history from what we've seen. You're correct that over history, we have had the tendency not only for new gamers coming in or those repurchasing to upgrade both an increase in our units that we sell, but also in terms of our average ASPs. New gamers come in, don't always buy the entry level. They may buy in the middle of the stock. Those returning will also purchase higher-end performance for the types of games that they're doing. We have seen over a period of maybe 5 years when we've been growing revenue more than 20%, you can almost see an equal split between the units and the ASPs. Now over the last part, through the pandemic, since the onset of the pandemic, we have grown revenue about 70% in our gaming business. In this case, ASPs were a little bit more influencing versus the units, but absolutely also seen unit growth over that time. Now we haven't announced any details of a new architecture. We have announced that we are getting ready for our architecture announcement. So stay tuned for more of that but we'll probably see things start to return to normal or have maybe a little bit less of ASPs in the short term, as we transition just based on that peak that we had seen during the pandemic. Over the long term, we still feel that there is that opportunity for ASPs to be a meaningful amount of our growth in gaming as well.

Got it. And Colette, on your last point, as you think about your through-cycle long-term growth rate in gaming, I think you provided very helpful metrics at Analyst Days in the past, but how are you thinking about that volume versus pricing kind of dynamic again for the gaming business specifically?

We always see the gaming industry looking for both new opportunities for the use of graphics, but also in terms of reaching more and more gamers. So units are still going to be an important part. As we think about ASPs, we'll continue to come out with a mix of a stock, and we'll probably see some early on, also purchasing that stock for the great performance that they'll get. We just don't have any details right now in terms of how that will be in that current quarter.

Got it. The other question, I guess the last question on gaming that I often get is on crypto and the exposure there, not just for you but as a broader industry. And I realize you don't have much visibility in determining or engaging how big or how small that part of the business is. But if you had to provide your sort of best guess or best estimate on how significant that market was in 2021 and what that's doing to the channel today, what would be your response there?

Yes. It's still very, very difficult. Coming up with an estimate is just not feasible to do in any manner. What we are seeing right now is a change in terms of the profitability of mining, that has gone down over the last several months. And then for quite some time, there's been discussion of the proof of stake coming. And that proof of stake is expected to be here maybe this month, maybe later this month, but now that is well known that a proof of stake is in market. Those 2 things together, less profitability in mining, as well as future proof of stake decreases the demand or use of GPUs. So I believe that the market is likely already taken that into account in terms of the purchasing that we see. It has certainly probably impacted the pricing that you see come out in retail and e-tail of the cards as well that have come back down to manufacturer's suggested prices. So that's what we see today. We get asked often possibly a resell. Will people resell their cars. We watch it carefully, but our gamers tend to buy new cars. We'll see a little bit from time to time on resell and that's normal. But it's not like a car that you can determine how many miles that GPU has been through, and so people tend to focus on brand new types of GPUs.

Okay. Got it. Shifting over to Data Center, just want to hit the export restriction point at the very top. So the $400 million is the part of your business that is impacted by the recently announced restrictions. And to your point, you're working with customers in search for potential offsets. I guess how seamless or not seamless is that process? When customers can say, "Hey, you can no longer buy the A100, let's buy this entry-level data center GPU, how common should we expect that to be, or is that going to be fairly limited in scope?

Yes. So happening in the middle of our quarter was probably the hardest piece to absorb. And then now we spend the time in terms of communication and helping understand what does this mean, what can they do in piece. Once we work through there, and we talk about alternative products, it can be either something with a slightly different performance level, but also you could think about a group of products together to lead the customer with a set of performance that they desired with the A100. That work is in progress, full steam ahead. In terms of helping them, now that I think we understand exactly what the U.S. government has asked us to do, we will follow all of those pieces. And we do hope within this quarter, we can mitigate some of that risk. But at this time, it's just very unclear how much within the quarter we're going to be able to do.

Got it. And I guess, as a quick follow-up, the other question that I get on this specific topic is, what's the risk of this broadening out to just data center or GPUs period as opposed to just A100 the H100. And I guess this is sort of a speculative perhaps unfair question, but how should we think about that risk coming through?

Well, I think the world of export controls has been quite active probably now for a decade. Not only specific types of chips or any type of machinery, there's often licenses required. There's restrictions in terms of certain companies that they can't ship to. There are certain countries that people cannot ship to. So I think that is a continuing pace. It just now surface something very specific for us here at NVIDIA. There's more discussion probably out there in the news today that more may be coming into the semiconductor market. We are aligned and working with the U.S. government. It was our job to help them understand how our processes work in the Asia Pacific area so that we continue not only following what they had planned for China and what they wanted to see, but also making sure that we could continue our work for all of the other regions around the world and our customers around the world. And that has been a very helpful process for us with the U.S. government.

Got it. And then on the broader environment and data center, on your earnings call, you talked about robust demand trends and visibility in terms of U.S. hyperscaler demand. I think on the China side, even prior to their export restrictions, you talked about some softness. In enterprise, I think you talked about some deals taking longer than expected or a little bit more of a mixed environment. I guess, what are your updated thoughts there? Any deviation in how you're thinking about those respective buckets?

Yes. So we've -- we talked about at our call, correct, that we felt that the U.S. market for hyperscales was -- did that get better? Okay, that the demand for our U.S. hyperscales was quite strong and a little muted in terms of our China hyperscales, or China hyperscales even before the restrictions, significant economic issues in the country, possibly some of it related to the COVID lockdowns and what they needed to do, but other challenges just with the economy and that was known. But even additionally, our work right now with our customers, enterprises and particularly in terms of our hyperscales, there are still some challenges in the supply chain with commodity types of chips. We refer to this as our work that we're doing to help kit them. And along with selling GPUs, they may be looking for very important pieces of our networking products or very important parts in terms of cabling, things that seem so simple for their data centers, but are holding up the ability for them to establish and roll out new data centers. You'll see this because it's actually very difficult to get a GPU in the cloud. They're sold out in many different places. As these logistics, distributions and just the overall volume of what they need is still lacking for them to do. We've been working carefully with them. And I guess, our work right now on selling networking and GPUs is working quite well. We called out our kitting process to really help those hyperscales build out these data centers as quickly as possible. But that is something that we are seeing worldwide.

Okay. And I guess in U.S. hyperscale specifically, the big, big concern that many of your investors have is the macro is going to ultimately impact their cadence from a spending perspective and things of that nature. As you go back and forth with your customers, are you sensing pushes and pulls given the environment? Or you guys provide something that's so strategic and so critical to their process, I would expect your business to be a lot more resilient than sort of down the middle company. But what are you hearing from your customers, what's kind of the outlook there?

Yes, you hit on some really good points there, that says when you focus on the enterprise and the hyperscales, the hyperscales, essentially their end customers are those enterprises. We're seeing solid, solid demand continuing from the enterprises, even in this macro conditions. We have a portfolio, a platform that is key to many of the both lighthouse accounts and further down the stocks in many industries that differentiate themselves and is a competitive advantage if they can think about how leveraging their data, building solutions to help monetize better their platforms. These are going to be the things that you do in these very difficult times. You have to focus on your future and what's going to be necessary and we're right in that right position to do so. We have an upcoming architecture as well. Folks are extremely interested in what that also brings to them in helping them build out.

And Colette, I guess on that last point, the H100, which you'll be ramping over the next several quarters, as you sort of compare and contrast this upcoming ramp with what you saw during the A100 and the V100 ramps, anything -- apart from it being a much better chip, any difference in how we should think about the cadence of customer purchases and things of that sort.

Well, a lot has changed since the early years of V100 and even A100. V100 was a surprise bringing to market. Customers took that opportunity to take it, qualify it and try and bring it up into market and it was a race to see who could qualify faster and bring it to market. Things have changed. Things have moved to our bring-up. And our bringing a product is not just about our engineers, it's about our customers' engineers. Do they think the product is ready? That's how much work we are doing in terms of scaling for the next architecture. They are integral in terms of us bringing this product to market. It's important to see also the changes that we initiated even with A100. A100, it became a system. It wasn't just a GPU. We assisted in terms of bringing a full-scale board to put the GPUs on a full system prequalified allows them not to full use. This is important for them to both understand what is the performance level that they can expect, what are the projects that they have coming to work, H100 will certainly meet those needs. As you know, we've talked about it being very key for transformers, very key in terms of the large language models that we see in the future will be the extension of natural language processing. So I think we feel great about the architecture coming to market, our work with all the different customers, but a lot has changed from the early days of something like a V100.

Got it. Okay. And the Grace CPU is a product that a lot of folks are excited about. It's a new part of the TAM that you'll be addressing going forward. I guess, how should we think about the percentage of the TAM that you'll be addressing? I think you've characterized it as being a relatively niche part of the market, initially at least. How should we see that progressing going forward? And any sort of traction, customer pull that you're seeing today with this super CPU?

So our Grace CPU is special, our Grace CPU is special in terms of the workloads and what it is geared for. It was really geared towards high-performance computing types of applications, supercomputing and of course, the field of overall AI. Now that's an important piece to understand. Yes, it has a TAM. Yes, it's a sizable TAM, and it moves from going from zero to us from a revenue. So that's a very strong growth rate going from zero. But what it is not, it's not a general purpose CPU. We targeted to really think about that field of high-performance computing and AI. The full data center look of what we could do with CPU and the time of the data that was inside the data center and how long it took through and what we could do for the CPU. So we're excited to bring to market. It's going to be here next year, probably in the first half of next year and bring to market. So a different town, yes, but also a great opportunity for us entering into a new market.

Yes, got it. And then I wanted to touch on the software franchise that you're building from a pretty low base, but you've been pretty loud about that opportunity, both in terms of data center as well as the automotive side. But focusing on data center, what kind of progress have you made over the past several quarters? And where do you see that opportunity going, going forward?

We'll probably talk a little bit more about this at our GTC that Jensen has coming up in the next couple of weeks. But let me kind of talk about how we see software. Software is integral in terms of our progress already in data center and our products that we sell. Essentially, a chip without software is just a chip, and we've established a full platform stack in the work that we are doing. We bring software every single time that we move to a new architecture, the software still works, and it's there to work with the developers that are out there. Now software is different depending on the type of customer. When you think about the hyperscales, the hyperscales usually design their own software. We assist though so many of the enterprises of just -- that are getting started. The enterprises are not equipped with software developers that would develop that whole stack before it gets to the application. They're going to work on the application. Our job is to stitch together all of those software components up to that application and working with our full platform. Now we have a couple of opportunities to monetize software on its own. It is incorporated today in what you see in terms of our pricing, but now the monetization could be different. NVIDIA AI enterprise very geared towards the enterprises. Think about it as the AI operating system that sits on top of the platform. The beginning starts that both that you can maneuver your entire platform, your data center using NVIDIA AI, but also comes with the support that you have for any of the SDKs that we've established for all of the industry workloads that we see out there as well. So that's one piece of monetization. And we can monetize by cores, by GPUs with NVIDIA AI enterprise. Secondly, more application or platform application level is Omniverse. Omniverse is our ability to help provide the stitching together of a platform to build for the Metaverse. The Metaverse focused on anything from digital twins to an overall simulation of a factory floor or a manufacturing floor and how to redesign it for both efficiencies as well as AI. So these are very important 2 key platforms in the data center, aside from our software that correct, we can also put in the car in terms of automotive.

Got it. I want to pause here and see if -- okay, there's one in the front row. If we can get a mic, please, in the very front.

Yulia Hofstede from Martin Currie. Colette, could you please quantify the importance of Chinese hyperscalers in reference to either your previously communicated TAM or maybe future growth beyond the third quarter?

The Chinese hyperscale was the question regarding how do we size it both now and in terms of the future. They are different, for example, the U.S. hyperscales. The U.S. hyperscales are much larger and take up a larger percentage, for sure, in terms of just with -- the type of AI that they accomplish. If you think about those hyperscales in China, they are multifaceted. Not all of them have those same types of things, that are not sisters to exactly what is here in the U.S. They have always been smaller of that component. We've sold to them for years, indicating what percent of it, much smaller than what we have here in the U.S., and that will probably continue.

So beyond -- less than 25%?

No, they're not at that level, much smaller.

And there was -- yes, right there, third row.

Chris Hanaway from Allspring Global Investments. Can you talk about just not in terms of your company specifically, but just the puts and takes of having the federal government just so getting so deeply involved in the chip manufacturing and where that gets done now, just at a high level the puts and takes of that?

Well, I'm not sure I could articulate from the U.S. government side, but from our communications, our work with them is nothing different from any chip company. There has been export controls that we need to follow. And it's a serious thing. U.S. export controls is broad and wide. And in fact, it's more than just the U.S. or a single country it is worldwide, and we're very careful with it. Careful to both understand what the plans, what the intentions are, but also very rigorous in terms of following through on it. I think going through this last process, as you had seen us publish the 8-K and our second 8-K after clarification with the U.S. government, helping them understand how our just internal operations work. Asia for many of the semiconductor companies is a very important part, meaning we are a global semiconductor company. We use folks -- companies in the Asia area. We also have a very big part of it here in the U.S. So helping them understand how that works, just to make sure that we could both follow what they wanted but also we could continue the operations. And I think we helped explain to them, and we feel very good about that relationship. So that's what I can say about the U.S.

Colette, I guess on that last point, just on your foundry strategy broadly, in the past and to this day, you work closely with suppliers like TSMC and Samsung. There's a very large domestic supplier who is on its way to ramp capacity. When you discuss your foundry strategy with Jensen and the broader team and the Board, how topical is localization? And how often does it come up? Is it a big problem than most of your productions in Asia? How do you think about it? What's the debate internally?

I think we started this process years ago when we went to a dual-fab strategy. I think we're probably one of the largest semiconductor companies with a dual-fab strategy. That helps us both with finding the best provider for the solutions that we may be building. That doesn't mean that it's always the same, node to node. Each of them have different processes, different skills, and we've been working with them. We couldn't be more happy about that strategy of having 2 of them, and we'll continue over there. The plan to have more here domestically, absolutely is a great strategy as well. Having something closer to home for us, I think, could be beneficial. But we do know that that's a long-term standing goal to get there. It's going to take some years to get there, but we wouldn't be more excited to be working even with those same fabs or a new fab here closer to the U.S. So it's front and center for us to always think about the fabs. It's where some of our initial processes stand, but it's not the only thing. As you know, we have a lot of system builders. We have a lot of contract manufacturers. We have a ton of suppliers, and many of us just think it's us and the manufacturer, but it's a host of a very complex supply chain that we have.

Right. And on gross margins and operating margins going forward, despite the muted revenue outlook in the near term, your gross margins are extremely healthy. You had a big write-down in the July quarter, but that was pretty onetime in nature. As you think about your profitability going forward, what are the puts and takes at the gross margin level? And I think you talked about slowing down your hiring and managing OpEx a little bit more -- a little bit more scrutiny there. So if you can kind of walk us through those dynamics, that would be helpful.

So let me first start with a question regarding gross margin. And from a put and take standpoint, our biggest driver of our gross margin is always going to be mix. Our Data Center, including so many of the systems, continue to be drivers of gross margin upward. And after we finished Q2, stabilizing the channel, we'll probably see a little bit of muted results in terms of gross margin. But long term, the driver of data center mix and software should continue to fuel our gross margins going forward. Now on operating expenses, we took this opportunity just as many large companies did to both slow down hiring that we could concentrate on the employees that we have. The employees also faced inflationary issues, just like many other companies around the world. We focused on setting them right and increasing their salaries so that they could work through this. But now we are also working on how we can be efficient in our costs. Our cost going into Q4 will probably choose to be approximately flat and continue until we start to see the macro environment improve. That just takes work on making sure that our products coming to market are front and center, and that's where we're focusing our energy and then finding efficiencies behind the scenes as best as we can.

Okay. Got it. I'm hopping around here a little bit. I wanted to hit on automotive. It's a very robust design win funnel. I think the July quarter was the first quarter where we really saw the inflection to the upside, which was great. Again, can you talk about the visibility you have in that funnel hitting your P&L? And I think historically, your profitability in automotive from a gross margin standpoint was a little bit below the corporate average. How should we see that evolving over the next couple of years?

Let me give you some background of our automotive business. We have started to grow in Q1, then moving to Q2, and we do expect going into Q3. What's behind that? Earlier this year, our Orin architecture launched. We have nearly 40 design wins on Orin, ranging from passenger cars, robo-taxis and future AI vehicles. This is a case of anything from a Level 2, Level 2+ to Level 5 with Orin. And you'll see that continue to scale over time. I think we're in what we refer to as an inflection point for our automotive business. Now some of the early parts of the Orin platform before we get to the AI and software will still be below our company average. But once we have the software, these will be quite solid, great gross margins as we move forward. Our work there is broad. We have a pipeline that helps us with that visibility. What that means is your automotive are working in terms of their manufacturing lines, how many each quarter they want to build. They'll sometimes even give us a longer view that helps us with visibility also tracking in terms of that revenue.

Okay. Got it. Any questions from the audience? I guess in the last couple of minutes, I wanted to give you the opportunity to kind of speak to any aspects of your business, any macro dynamics that we should be aware of. Obviously, NVIDIA is a very well-covered stock. But any other topics that we should be hitting here.

Okay, sure. Let's see. NVIDIA, our position right now in the market, demonstrating the availability of AI for all couldn't be stronger. And we're approaching now a really great product cycle, a product cycle that takes our graphics architecture, takes our work in the Data Center, not only bringing compute to market at -- with new products, but also in our networking business. So these product cycles are important to us. Important to us for us to continue to fuel the AI market, the expansion of graphics as well. And that's going to be front and center. During these macroeconomic times, inventory corrections, not the most fun thing to do, but we also know that we'll pass in time. And hopefully, by the end of Q4, we'll be in a great position as well from a both sell-in and sell-through perspective going forward. But we have great new products coming. We couldn't be in a stronger leadership position. And that will take us far, I think, going forward.

And maybe lastly on capital allocation. Any thoughts there?

Yes. So capital allocation. We would always love to look at more and more M&A opportunities, but we are quite selective in finding the right ones. We felt that Mellanox was a great addition, both culturally and what it brought to us as products. We'll continue to look, but it's not easy to find those. In the meantime, we are going to focus in terms of stock repurchases and rebuying. We've done more than $5 billion in the first half, and we've got $12 billion of opportunity that would take us all the way to the end of next year's calendar year, over repurchasing as we speak.

Awesome. With that note, I'd like to close the session. Thank you so much for coming.

Thank you, thank you.

Thank you all.