NVIDIA Corporation (NVDA) Management Presents at Morgan Stanley 2020 Cloud Secular Winners Virtual Conference (Transcript)

NVIDIA Corporation (NASDAQ:NVDA) Morgan Stanley 2020 Cloud Secular Winners Virtual Conference Call June 1, 2020 12:00 PM ET

Company Participants

Colette Kress - CFO

Conference Call Participants

Joseph Moore - Morgan Stanley

Joseph Moore

Okay. Welcome back everyone. This is Joe Moore from Morgan Stanley and I'm very happy to have Colette Kress, the Chief Financial Officer of NVIDIA. Before we start, I just quickly need to read a disclosure. For important disclosures, please see the Morgan Stanley Research disclosure website at www.morganstanley.com/researchdisclosures.

Please note that this call is for Morgan Stanley institutional clients and financial advisors only. This call is not for members of the press. If you are a member of the press, please disconnect and reach out separately. Comments made on this call are not for attribution by members of the press. Please note also this call is not for individual wealth management clients. If you have questions, please reach out to your Morgan Stanley financial advisor for more information.

So, Colette, thank you very much for joining us today. We're really excited to have you.

Maybe we could just start out, COVID has presented some unique challenges in the environment for all of your customers and suppliers. You guys have navigated it pretty well. Can you just start with maybe how you guys have navigated this so well? And how you guys are managing disruptions in the supply chain and the health of your employees and things like that?

Question-and-Answer Session

A - Colette Kress

Sure. Thanks so much for the question. We just released our results for our Q1 results last week, and we had provided overall Q2 guidance. Certainly, COVID-19 was included in our Q1 results. When we announced our guidance for Q1 back in February, I think probably one of the things that in looking back was helpful for us is the amount of our business, amount of our supply chain which is in the Asia-Pac area that allowed us to be front and center of the early stages of COVID-19, and what the early signaling was done.

It's hard to know with specifics how much was impacted with overall COVID-19, but some of the pieces that we saw right off the bat was an overall close of the retail channel, which impacted our gaming business at the start in terms of retail sales as well as iCafe was now shut down in the overall Asia-Pac area, but what we quickly saw was a move to eTail.

People in the Asia-Pac area had begun working from home, the schools were shut down, and it was a time for them to also learn from home and teach from home during that period. As you know, being in a shutdown worldwide, it led to also determining what the overall entertainment that would happen over that period of time, and gaming really was a place that people turned toward to find entertainment in their overall homes.

So, we were able to quickly, an agile company, focus on moving towards eTail, focused on how to get the overall gaming to our end customers, and that worked pretty well.

But then it moved to a pandemic, that pandemic within the quarter moved to worldwide, and we were able to really leverage the learnings that we had in the Asia-Pac and China area to think about that influence worldwide.

Keep in mind though, things in every single country, every culture was different, and it really focused us on understanding our suppliers, understanding our supply chain process, and pretty much a connection on a weekly basis, daily basis in some cases to assure that we understood what they were facing.

The overall Asia-Pacific area restarted faster than some of the other parts of the world, but it was about getting overall people into the manufacturing and into actually building the overall products.

There still were some challenges in terms of logistics, there were challenges in terms of just moving supply from place-to-place, and our connections with our top suppliers and connections with those manufacturing aided us during this process.

What was unique was our demand, particularly for our overall compute, and our data center stayed intact from what we had thought at the very beginning of the quarter. And so really, we were just working through the overall supply and logistic issues through the quarter. We feel fortunate as a company in terms of the businesses that we have chosen, and we know that there are many out there that are still challenged by COVID-19 with their overall businesses.

We expect this to also continue through to Q2. We had highlighted that there are probably some continuations of COVID-19 into our Q2 overall guidance, particularly regarding the automotive business. The automotive business and our legacy infotainment systems will be challenged, and we will likely see automotive decline approximately 40% from where we had in Q1.

We also expect our professional visualization project -- products to continue to seek where the overall business comes back from COVID-19, but will likely have a decline sequentially between Q1 and Q2 as well.

Overall, gaming is still an area where I think demand is important as well as our overall data center as we continue to ramp our new products such as A100 and the lineup for overall AI.

Joseph Moore

Great. Well, that's a helpful overview and it's an impressive job of maneuvering in what's obviously been a really tough environment for everyone. So, I wanted to walk through the segments a little bit more in detail, starting with cloud, since that's sort of the theme of our conference as well as the fact that it's now over half of the company or on a trajectory to be over half of the company at least.

So -- and maybe we could start just looking back a little bit and putting this strength that you've seen into context. You grew 80% year-on-year this quarter, which is not something I saw coming.

But again, last year, you saw this kind of digestion. So, I want to talk about some of the newer drivers and the things that are driving the strength, but I think what's the postmortem on the digestion that you saw a year and a half ago, and sort of the prognosis going forward for -- is this going to continue to be a lumpy business or those kinds of things, if you could just kind of maybe put that into perspective for us, that'd be helpful

Colette Kress

Sure. So, when we look back over even the last three plus years, with the introduction of the V100, and deploying that into so many overall data centers for internal use as well as for the cloud, both a lot of lessons learned but also a lot of overall expansion in terms of the customer sets and what put into place.

When we look in terms of our initial rollout of V100, sure, we started with the overall hyperscales, we started with the large customers as they found the importance of AI for their overall business models, for their overall monetization, and really saw it as an important piece of the future of whether they think about their applications and even some of the basics of how they think about search, and search on the internet. So, we began a significant amount of work working with them as they built out using overall accelerated computing.

Moore's Law was quickly coming to an end, it was very well discussed, and there was also a need to now allow AI to be used in the overall cloud environment. So, those three things around there really started an overall surge of interest, a surge of overall demand in terms of not only for internal engineering teams but also surfacing that up to the cloud.

Why they were important is that was the way that the overall enterprises, the higher education, the research could get their first hands on using overall accelerated computing with that.

Now, if you recall during that time, we also had some challenges with other providers having a shortness of overall supply. CPU shortage was being signaled in the overall market. So, when we look past that overall digestion, which can be called in when you're building out overall data center builds, the influence of also CPU shortages made it even more cloudy for people to understand where the overall demand would be.

And through that time, purchasing ahead of what they needed as the data center, so they would not be short of any overall products is likely what happened over that period of time. So, as we've turned now to a new architecture that we have announced for A100 and our overall work over the last three years, we have built out both a stronger list of customers, adeptness of customers not only from a procurement and certain parts of overall engineering, but through and through understanding their workloads, their builds, their projects, so that we can actually help influence the importance of lead times and how long it will take in order for them to build out their overall data centers and/or qualify, and then what we can do to overall help that.

That overall significance breadth and depth that we now have with our customers, we hope will help us as we go forward, but there is still likely that from time-to-time a single customer may be digesting, may be finishing out their build before they start a new one.

Our hopes is that more overall large projects stack up on top of each other that keeps the overall outlook smooth, and we believe that our process of working with them from a forecasting and that lead-time will help improve so that we have a better understanding of the forecast going forward.

Joseph Moore

Okay, that's very helpful. Thank you. So, maybe if we could talk about some of those new applications. First, conversational AI, it's something that you guys started talking about and then interestingly we started hearing it directly from the cloud customers, sort of talking about the priority around these transformers for language translation.

And when we look into the models, the order -- the amount of complexity, the number of parameters is literally millions of times what the Computer Vision's models look like. So, clearly a big jump in complexity, pretty big compute workload, can you talk about where we are in that ramp, and is that going to be something that's narrow across a couple of cloud customers or how does that broaden out from an application side so that it's more interesting to a broader base of enterprises?

Colette Kress

Sure. So, the concept of conversational AI and the very initial models that have been built over last summer or late spring at about this time last year really started to surface. The BERT models surfaced. There have been many derivatives of the BERT, other ones that have been started by many of the overall hyperscales.

And with our direct working with them, we really understood their desire to build the overall complexity of models to support overall conversational AI. Conversational AI, the underpinnings of it, which is the natural language understanding, which has many parts to it, it's not just a single overall function of work. There is a three-step, four-step process that is necessary for much of the natural language understanding that is happening.

In terms of understanding the overall speech, the tongue, the overall dialect, as well as understanding the words that were used in terms of what is the question, what are you trying to solve for, and then lastly being able to respond in a solid, low amount of time, in order for it to be within the reason of what you were expecting a response back.

Now, what you are seeing is we're still in the very early stages, but what they needed in order to complete that was a size and an ability of compute in order to help them through that process. So, our coming out with overall software, but more importantly, A100 allows them to expand their work to take on larger and larger models and more complex models to do their work.

What we're seeing is the desire to not just bring in overall speech, or to bring in just written overall data, but taking in any type of form factor and ability for the models to be at the largest sizes to address some of this very difficult work.

And so, we're in the early stages, it's not necessarily something as just for the overall hyperscales, because you will see a lot of it also be part of overall cloud computing, and that cloud computing is really about enterprises using the overall cloud and you can think of many other instances that they will need the overall conversational AI, whether that be overall call centers, whether that be ways that they converse as an overall company as a whole.

So, focusing first in terms of some of the work that the hyperscales have done that they really see this as a new way of collaboration that will likely happen in the overall enterprises as well.

Joseph Moore

Great. Thank you. Any other area that you guys have talked about quite a bit recently, recommendation engines, or various terminology to refer to those. The complexity of those models is sort of maybe less intuitive to me, but obviously really important to the economics of your cloud customers. Can you talk a little bit about that workload?

Colette Kress

Sure. The recommendator engines we see all across the overall internet, not just in terms of the hyperscales, but also the Tier 2 or the consumer internet companies that may be on top of overall cloud instances, in terms of how they run their business.

As you know, just at the basics that the news that comes to you every single day is generally a high end, overall recommendator engine, what type of news interests you? What types of things have you read? How do we continue to provide things that meet your needs and overall interest? But this can go all the way in terms of overall procurement as well.

When you think about what you procure as a business and/or personally, this is another area where recommendator engines are very important. You can think about how complex those models can become in terms of targeting marketing to you. And also the speed-back in terms of those recommendations in terms of whether or not you are driving or whether or not you're just on your phone.

So, we are with our overall processors definitely able to process the significant amount of recommendator engines, but also remember we have a set of application work that we are doing to better serve the overall recommendator engines in terms of software that supports those functions as well.

Joseph Moore

Okay, great. And then the other thing that I found pretty exciting about your business is AI inference. The NVIDIA chips have historically dominated the market for training, for building large databases, but inference, the access of those databases has happened more on traditional CPUs. But that business has become pretty sizable for you guys. And clearly you continue to make progress and ultimately, it's a market that that should offer similar sizes training. So, can you talk about your progress there?

Colette Kress

Yes. So, inference, as we had outlined a couple of years ago, as we felt it was a very large market, a large market that we felt was appropriate for overall GPU computing. The overall inferencing we're talking about though is not necessarily the inferencing of the past. You're correct. It's a very CPU-dominated overall piece. A lot of that has been single function or very binary type of overall inference.

The CPU, again may be perfectly fine to do, but when you think about using GPUs for overall inferencing going forward, the complexity of the inferencing and the models and the multiple stages puts the GPU in a great place, both for the overall programmability, but also the overall speed that can be accomplished using overall GPUs.

Using conversational AI pass when you think of the training process for conversational AI and moving to the inference is a great overall use case of GPUs for overall inferencing.

We started this business with unique form factors for inferencing that could be slotted directly into existing OEM servers in terms of a PCIe slot, our T4, for example, is just well-engineered in a candy bar size as low as very low wattage in order for them to just out that overall inferencing to their overall compute.

And we have reached a point where inferencing is a sizable percentage of our overall data center business. It is solidly in the double-digits percent of our overall data center.

Now, what's interesting it's in a doubling as year-over-year in this last quarter, but also when you think about A100, A100 has been engineered and built so that it can accomplish both training and inferencing as we move forward.

So, overall customer does not have to choose that says which workload am I trying going to do? You have to build the ability to choose A100 to do both. It has the ability for it to be partitioned, partitioned into multiple virtual instances that each individual instance could be used for overall inferencing, overall performance improvement in for -- versus using V100, but also just the flexibility of a single platform to accomplish both the training and the inferencing.

Joseph Moore

Great. And that's a good segue. I did want to ask about A100 and the Ampere family. These things don't come along that frequently for you guys. It's your first new architecture in three years. And it appears to be both from what you've said and what we've heard from your customers, significant advancement and state-of-the-art. Can you talk a little bit about Ampere, the performance benefits that it brings and really what that will do for your business in terms of growing the number of workloads that you can address and driving your business going forward as well as dealing with competition?

Colette Kress

Great, so yes, our A100 and our Ampere architecture, right before earnings, a week before, was an opportunity for us to launch for the world to see what we've been working on for the last three years. Originally probably targeted to be communicated at GTC, San Jose back in March, but we just had to move that out a little bit and to try and find the right form to communicate.

So, we launched it just a couple weeks ago. It is in full production and was a part of our overall Q1 results. The A100 architecture is our largest leap in production -- performance versus any other architecture in the past.

The A100 alone is 20x stronger in performance versus the overall V100. It is on 7-nanometers. It is the greatest performance on 7-nanometer. It is also the largest chip on overall 7-nanometer. We're using this opportunity to launch A100 in a unique manner that we think will help both deployment and ability to bring it up in so many of our different customers.

We are launching it with eight GPUs together on an overall baseboard. That overall configuration easily slips into their existing overall infrastructure and will, in the future, also be with overall OEM centers as well.

What is unique about it as we discussed is it provides the ability to execute both training as well as inferencing. This is an important piece as much time is spent by customers, whether they be hyperscales, whether they be enterprises in terms of choosing their AI platform or their accelerated platform. And this is an important piece that allows them mobility to do both and create over seven instances per GPU and being there's eight GPUs; you have a multiplication factor of how many overall instances you can have not.

Only important in terms of the A100 architecture is the overall performance straight out of the overall hardware, but keep in mind, you can look at this as a very important software play that we do as well.

We are on CUDA 11 now that has come out. CUDA 11 is the underlying development language platform that we have on every one of our GPUs, as well as therefore, our overall architecture with Ampere. It is important piece as it is the underlying driver of many of the enterprise-specific overall application that we will address.

We also provide software that is specific to many -- the key industries, vertical industries and key places that we believe accelerated will bridge to. Some of the things that we also launched with the Ampere architecture is new software enhancements that can help both for conversational AI, can help for overall recommendators, can help for overall -- anything in terms of the data analytics that is out there and the work with overall Spark.

So, our work -- this is the third generation tensor cores that we have with the overall Ampere architecture that is an important piece that allows both for high performance computing, but also the dialing up in terms of precision that you may use the overall GPUs for the positions of your different workloads. So, we're excited about not only the overall Ampere architecture, but all of the software that we have also brought to market on top of overall CUDA and CUDA-X.

Joseph Moore

Great. And then maybe that's a good segue to the competitive question. And in particular there's a question on the webcast that I get a lot which is how do you think about custom silicon from your hyperscale customers? Obviously, Google, most prominently, but the others have talked about that as well. Can you just talk about the importance of the new product in that competition? And do you expect that to be a negative for your growth at some point down the road that your customers are trying to develop customs to look in?

Colette Kress

There are some cases and have been for a while. We're looking at custom ability and custom silicone largely due to large workloads that they may see in front of them that may have a overall custom silicon to support it. But what we've found is we are the platform that is probably the most agnostic to any platform that is out there that we need to work with, any CPU, any other types of components in the data center, as well as any other type of software that is out there.

So, our ability to work seamlessly across different frameworks for AI, as well as different types of cloud infrastructures is extremely important and why people think about using overall NVIDIA's GPUs to do that.

You always have the ability with the frameworks to continue to write on top of those frameworks. However, if the frameworks still need revision, still need more work to be done, you always have the ability to jump into the overall CUDA development platform to add and expand to that.

We believe our performance as well as meeting the needs of our customers is always been upheld quite strongly. If you think back to the overall fall timeframe when MLPerf came out, both for overall training, as well as for inferencing, this is really a time for us to demonstrate from a benchmarking standpoint, the hardware and software combination that we have puts us in best of breed.

And often when you see that custom overall ASICs that may be out there, it becomes very challenging for them to overall compete with the overall universal capabilities of the overall GPU and based on the speed of AI and AI development.

Something may work today, but as you've seen the 3,000x improvement in terms of model size, as well as complexity of AI, having that flexibility with our platform really enables us. So, if our overall hyperscales or other partners look for a need that they see a workload that may need a specific ASIC, that may work for a time period, but we just don't think it really applies to the broad market that is out there that really needs the flexibility as things are moving so fast.

Joseph Moore

Great. And for what it's worth, I mean, we've heard that directly from some of the developers of custom silicon is that there are applications where it works and they're still very high demand for programmable GPUs from NVIDIA.

Another question from the webcast is on the geography of the cloud business. Can you talk a little bit about your positioning in China versus the U.S.? And is it possible that to the extent that the U.S. goes through a digestion phase that China would continue to be strong for you?

Colette Kress

Yes, so each of the regions does operate differently and we do serve all of the major overall cloud providers around the world. The cloud providers that are here in the U.S. are built with larger overall CapEx budgets and are just bigger in nature than the ones that are in China. China has hyperscales, but also has a second tier that is quite broad. And that's not just in China, but it is in the overall Asia-Pac area.

And -- so we continue to serve that. I think that helps us in terms of the differentiation of our overall customers in the overall cloud. And we continue to see those projects overall stack up. All go through some form of digestion, but the more ability to expand our overlook customer sets, projects at each one of them can help if any single one customer goes through digestion.

Joseph Moore

Okay. So, to maybe just sum all this up on the -- your organic cloud business, you sort of said, it seems like it's pretty healthy growth in July and you said that you probably need more time to assess what happens beyond July which seems totally reasonable, but it seems like a business that offers a lot more breadth of application and breadth of customer than what you saw a couple years ago. Is that all fair?

Colette Kress

It is fair, when you think about A100; we're in the early stages of a multiyear ramp. The demand is strong for A100. We do have that visibility into Q2, which what we used for providing our overall guidance for Q2. As we look at Q3 and Q4, just given the overall environment out there, we just need a little bit more time to really size in terms of what we see for Q3 and Q4. And we'll provide that when we when we get to that time.

Joseph Moore

Great, makes sense. Then maybe talk a little about Mellanox, you've talked about the rationale for doing the deal, their business has obviously been really strong. So, now that the deal is closed, can you give us an update on how this fits into the NVIDIA portfolio and maybe I know you acquired Cumulus right after, seems to be that you're committed to networking and maybe a bigger way than I had realized before that second transaction. It seems like this is an area where NVIDIA plans to continue to grow.

Colette Kress

Yes. So, our Mellanox acquisition, we're really pleased that the acquisition did close about a month ago and our Q2 results will have a full quarter of overall Mellanox business.

Mellanox was doing quite well in this last year. I think you saw even after the overall announcement of the acquisition and our continued partnership with them, they still stand alone, did quite, quite well over this last year. So, we're really pleased to bring them on part of our overall business and will be incorporated into our data center business.

Now the question is why. Why is that a great addition to our family of products that we have in our overall data center business? As you've seen us focused in terms of accelerated computing, accelerated computing, one of the key areas is overall AI, but it's also our base of overall high performance computing and overall supercomputing.

Mellanox brings their expertise absolutely into those same areas of customers that we do. What we've learned is not only compute and the time that is spent overall computing data information is essential in terms of that acceleration. But also the other components are very, very key to influence overall acceleration of the overall data center as a whole.

When you think about the modernization of data centers as we go forward, really thinking about the other layers that software-defined and acceleration can influence acceleration as a whole, is where we're overall focused on.

Their leadership in terms of the -- in the interconnects, their leadership in high performance computing, and overall supercomputing puts them as probably one of the best partnerships that we could imagine bringing on board at this time for overall data center.

You'll see us both working together with the same sets of customers, but also thinking about products, as we now understand the full footprint of the data centers with these customers on how we can build better things to influence acceleration going forward.

So, right now, just that merge of who we sell to, getting that right product to market is where we're focused. But you'll see more and more of a partnership both at the software level, as well as definitely the hardware level of new products that we can bring into market.

Now Cumulus, we announced that we are in process of overall purchasing. It's not closed yet, but this is another example of software-defined. Software-defined in terms of the overall networking, that can be a key component in terms of us spreading our wings of all of the different processes within the data center.

Joseph Moore

Great. Thank you. So, we could spend this whole time on just cloud and I didn't get to ask you about DGX or edge-based computing, which I know are important initiatives. But I do want to get some gaming questions in.

Your gaming numbers have been good, better than I would have expected, particularly if you've told me the challenges that you've alluded to with traditional retail, with internet cafes, the supply chain there, is there a case to be made that there's pent-up demand from those things? And then when internet cafes opened back up, that there's more incremental demand or has there been a shift where in the regions where those cafes are prominent that people have shifted more to gain from home?

Colette Kress

Yes, thanks for the question regarding the success that we've seen in overall gaming through this period. Remaining agile during this period has really benefited us. As we thought about all of the different ways that we sell gaming to make sure we could address the different regions around the world, the different countries of the world as they worked through their overall COVID-19 issues.

The iCafes right now have probably returned to business, but not necessarily at 100%. As you could imagine, social distancing and overall density in the overall iCafes as they function is still important. So, they are all carefully bringing them back up on board and we're watching to see how we can help them in terms of get back into business because it is such an important part of their overall economy as well when they think about that.

The retail, it’s a same thing, how quickly can the overall retail come up to speed, it's still unknown and it's not clearly up to anywhere near 100% at this time, there's a long way to go. But reaching the overall gamers, whether it be a high end overall laptop, whether it be a overall desktop card that they can buy in eTail, and self-build their overall desktops as well at home and/or ability for them to share doing a high end overall compute with what they need to both work and/or learn in the overall home has really benefited us.

We also saw a demand surge as it related to overall steam, the amount of hours and time that they were playing online. This took the opportunity for folks to jump on GFN, which you know, is our cloud service that we recently just rolled out in February as an overall subscription offering as well.

We've seen a surge in that and each region being very differently as they use a streaming service to get their -- sometimes their first taste at overall gaming, as they haven't owned an own device to do that.

So, is there pent-up demand? Is there more? Just too early to overall say that. The demand is solid based on what we've seen, which doesn't surprise us. We've always found that during some of the hardest economic times, overall gaming and gaming on a PC platform is actually a great way to entertain and actually can be quite economical because there's so many different offerings at different price points to support that.

We didn't talk about it, but also our console business, our console business with Nintendo is doing quite well. And we're here to support Nintendo in the demand that they are seeing for the great consoles switch that they have right now.

Joseph Moore

Great. Talking about ray tracing, I mean, it's hard to pioneer new stuff, there's always sort of the chicken and egg issues of the complexity of writing software for something new and is kind of groundbreaking as real-time ray tracing. Can you talk about where we are on that? It seems like you've seen an inflection where the demand is -- has really picked up there.

And I know you can't talk about next generation silicon products, but I know historically, when you've had these kinds of breakthroughs, like shader algorithms, the next generation that implements it that much better once it's already become something that's in the ecosystem has been pretty good. So, can you just put some perspective around the importance of ray tracing in your lineup? That'd be great.

Colette Kress

Yes, so ray tracing is an important piece that will be the next generation of graphics as we go forward, not only for overall consumer graphics and therefore, gaming graphics, but also for the enterprise. So, it's making a material impact on both of those major markets.

What we're seeing is all of the great publishers around the world have jumped on overall ray tracing. It was a little bit of a chicken and the egg that says what comes first, the overall software or the overall hardware. We introduced it, both starting back with Microsoft and the overall development on top of the overall APIs as they became available.

So, for us to come into market as the leader and first-to-market has really, really breached a tremendous following of all of the different publishers, but also those that are building overall game to the future. We have right now quite a star lineup of terms of games that are on. And using ray tracing, we have 33 RTX games that have been announced and are actually shipping.

Probably most importantly is the bestselling game out there, Minecraft, is really a transformational shift for them. If you're familiar with overall Minecraft, and its and its overall architecture, and the use of overall ray tracing to bring realistic type of pictures, it has been a really transformation for that overall game. But keep in mind, there's a tremendous long list of other superior types of games out there.

Warrior 5 -- MechWarrior 5 is out there, Call of Duty, Control, you're going to see Wolfenstein, Battlefield 5, we can go on and on. Cyberpunk will probably be one that comes out in the future as well. So, these as you can see are important games and I'm sure every publisher as it thinks about games for the overall holiday season, or beginning to write games knows that they have to make that decision on how to incorporate overall ray tracing because in the future, it will probably be a very important for high end types of graphics games.

Now, being the leader into market and having the overall hardware and the performance of sports, it is only one piece of it. Keep in mind, we have added significant amount of software to enhance the overall ability for ray tracing, such as DLSS. DLSS is an important piece of using AI in overall gaming.

What that does is it takes the features of high end ray tracing and breaks that down to where you can infer in between two pixels, what that overall completion of the overall frame should be. It improves the overall performance on ray tracing, but gives you that realistic and great high end overall graphics as well.

We'll continue to take this not only to gaming, but to the enterprise, so many applications in film industry or things that are in terms of CAD drawings or otherwise, the use of realistic materials. We are going to get closer and closer that we will look at a catalog, we will look at a set of pictures to try and determine, is that real or is it rendered? But there's a great wave right now. We're excited in just such a short amount of period of time, the uptake to support high end ray tracing. So we look -- shouldn't look that this is ended. We have more to come in terms of the software enhancements that we have as well as just the overall architecture that supports ray tracing.

Joseph Moore

Great. Sounds good. I can't wait to see what developers do going forward. So, maybe we can talk about the financials a little bit. First, starting with the buyback, you suspended the buyback while you're working on the regulatory approvals for Mellanox; you still have quite a bit of authorization. When might we see you renew that? And what's the criteria that you're thinking about there?

Colette Kress

Sure. So, we used the time in terms of closure to Mellanox to try and begin the first path of rebuilding our overall cash. We had enough cash on hand, as you know, to complete the overall Mellanox acquisition with cash. But we wanted to make sure we had a solid amount of reserve on hand to support the investments into our business, or in business terms of a capital perspective, investments, capacity, and/or any future overall M&A.

So, we are building back that overall cash portfolio. We stay overall nimble and agile as we think about overall buyback. You are correct that we do have a authorization with our Board to repurchase up to $7.4 billion of stock repurchases before the end of 2022. So, we will stay nimble to find the opportunity once we continue to build back our net cash position, and then get started in terms of stock purchases.

Joseph Moore

Great. Thank you. And then gross margins continue to be moving upward 66% is a level that I didn't get to get to this quickly. Is that is that largely mixed driven? And are you seeing change in kind of like-for-like margins as you move forward? Or will the main driver be the move to HPC cloud?

Colette Kress

Yes. So, on our gross margin, our Q1 results were also a record level in terms of gross margin and as we move to Q2 and our guidance for 66%, it would also be also a record quarter. Mix definitely is the largest driver of our overall gross margins.

When we refer to that as mix, it's really talking about platforms that are encompassing a significant amount of software. The software, as you know, is being built with our overall R&D, engineering teams, and therefore, the overall cost of writing software isn't our overall OpEx.

It's therefore influenced our ability to maintain great gross margins for most of our overall businesses. We'll continue to find opportunities to enhance our overall gross margins going forward. I do believe that our past growth in gross margins has been quite, quite strong over the last five or 10 years, and as we go forward, we'll make some movements as well, but maybe not as the speed of what we've seen over the last 10 years.

But what you should see in terms of something like overall data center, we're in the early ramps of overall A100. When you're in the early ramp, you're also in the early days in terms of the manufacturing process of that, we'll probably see yields improve a bit. We'll look in terms of value engineering types of projects across our platforms, including the overall A100 architecture.

From season-to-season or quarter-to-quarter, we will have different mixers across our products overall, but also a different mix within some of our overall market platforms such as gaming. When we move to higher end platforms in such as gaming, we can achieve overall higher gross margins as well.

We haven't mastered perfectly yet that we have all of our gaming platforms at the high end, we do see people coming into market for the first time buying our higher end platforms. But we still have volume SKUs that bring first-time gamers or even those that are coming back to get the overall new architecture.

Ray tracing has allowed us to provide higher end platforms and therefore, achieve higher gross margins as well. So, a lot of different mix factors that, yes, a focus of ours to continue going forward to improve our gross margins.

Joseph Moore

Great. And maybe we could finish with a question from the webcast, which kind of ties everything together is your automotive business. First, can you touch on the declines this quarter? Is that a macro issue or is there any share issues in the infotainment space? And then probably more importantly, when do you start to see the autonomous programs and the big investment that you've made there start to come to fruition from the revenue side?

Colette Kress

Yes, this last quarter and the quarters we go into Q2 have been some of the most challenging quarters for the automotive industry. The overall crisis around COVID-19 again, was an industry that was quite impacted; quite difficult for them during this time.

We are here to support them as they bring back up their manufacturing lines and as they work through the overall shapes of their cars, but before we even entered COVID-19, there was a lot of discussion with auto manufacturers, as they knew that the future related to AV and EV were very important part of their strategy as auto manufacturers going forward.

Looking for a solid way to both develop on these platforms, but synergize across possibly a single platform that could take them from ADAS all the way up to robo-taxi became an important topic for them.

Supporting multiple platforms in in-market or during development is tricky for them, and very hard to find the resources to support multiple platforms. We've been working with them on our end-to-end platform that can support ADAS all the way up to robo-taxis to help influence a more strategic and refined investment that they need to make to develop these overall platforms.

So, as we work with them, we're excited to have an end-to-end platform today that can already go into production, but also working with them for the long-term as they think about their -- as their fleet and bringing us.

So, stay tuned as they work through COVID-19 bringing things back. But again, we probably are seeing something more in the two years out the term before we get to solid amount of production of autonomous vehicles in market.

Joseph Moore

Great. Well, Colette, thank you very much for spending the time with us today and sharing all this information. This has been great.

A - Colette Kress

Okay. Thanks so much.

Joseph Moore

Thanks everyone for listening. Bye.

Comments

[grxbstrd]

tl;dr

GPU in the data center: "we are the platform that is probably the most agnostic to any platform that is out there . . . any CPU, any other types of components in the data center, as well as any other type of software"

re cloud home grown asics: "you've seen the 3,000x improvement in terms of model size, as well as complexity of AI, having that flexibility with our platform really enables us. So, if our overall hyperscales or other partners look for a need that they see a workload that may need a specific ASIC, that may work for a time period, but we just don't think it really applies to the broad market that is out there that really needs the flexibility as things are moving so fast."
IOW, let them experiment, we keep raising the bar and enable workloads that will break their asics. [and soon enought, they'll just give up]

"when you think about A100; we're in the early stages of a multiyear ramp. The demand is strong for A100."

"Mellanox was doing quite well in this last year. . . we're really pleased to bring them on part of our overall business and will be incorporated into our data center business.

"[Gaming] demand is solid based on what we've seen . . . our console business with Nintendo is doing quite well."

"all of the great publishers around the world have jumped on ray tracing"

"Q2 - our guidance [is] for 66%. . . we can achieve overall higher gross margins"

"we probably are seeing something more in the two years out term before we get to solid amount of production of autonomous vehicles in market"