NVIDIA Corp (NVDA) — Q1 2020 Earnings Call Transcript

Thank you. Good afternoon, everyone. And welcome to NVIDIA's conference call for the first quarter of fiscal 2020. With me on the call today from NVIDIA are Jen-Hsun Huang, President and Chief Executive Officer, and Colette Kress, Executive Vice President and Chief Financial Officer. I'd like to remind you that our call is being webcast live on NVIDIA's Investor Relations website. The webcast will be available for replay until the conference call to discuss our financial results for the second quarter of fiscal 2020. The content of today's call is NVIDIA's property and cannot be reproduced or transcribed without our prior written consent. During this call, we may make forward-looking statements based on current expectations. These are subject to a number of significant risks and uncertainties, and our actual results may differ materially. For a discussion of factors that could affect our future financial results and business, please refer to the disclosure in today's earnings release, our most recent Forms 10-K and 10-Q, and the reports that we may file on Form 8-K with the Securities and Exchange Commission. All our statements are made as of today, May 16, 2019, based on information currently available to us. Except as required by law, we assume no obligation to update any such statements. During this call, we will discuss non-GAAP financial measures. You can find a reconciliation of these non-GAAP financial measures to GAAP financial measures in our CFO commentary, which is posted on our website. With that, let me turn the call over to Colette.

Thanks, Simona. Q1 revenue was $2.2 billion, in line with our outlook, down 31% year-on-year and up 1% sequentially. Starting with our gaming business, revenue of $1.05 billion was down 39% year-on-year and up 11% sequentially, consistent with our expectations. We are pleased with the initial ramp of Turing and the reduction of inventory in the channel. During the quarter, we filled out our Turing lineup with the launch of mid-range GeForce products that enable us to delight gamers with the best performance at every price point, starting at $149. New product launches this quarter included the GeForce GTX 1660 Ti, 1660 and 1650, which bring Turing to the high-volume PC gaming side for both desktops and laptops. These GPUs deliver up to a 50% performance improvement over their Pascal-based predecessors, leveraging new shader innovations such as concurrent floating point and integer operations, a unified cache, and adaptive shading—all with incredibly power-efficient architecture. We expect continued growth in the gaming laptops this year. GeForce gaming laptops are one of the shining spots of the consumer PC market. This year, OEMs have built a record of nearly 100 GeForce gaming laptops. GeForce laptops start at $799 and go all the way up to an amazing GeForce RTX 2080 4K laptops that are more powerful than even next-generation consoles. The content ecosystem for ray-traced games is gaining significant momentum. At the March Game Developers Conference, ray tracing sessions were packed. Support for ray tracing was announced by the industry's most important game engines, including Microsoft DSR, Epic's Unreal Engine, and Unity. Ray tracing will be the standard for next-generation games. In March, at our GPU Technology Conference, we also announced more details on our cloud gaming strategy through our GeForce NOW service and the newly announced GeForce NOW alliance. GeForce NOW is essentially a GeForce gaming PC in the cloud for the one billion PCs that are not game-ready, expanding our reach well beyond today's 200 million GeForce gamers. It's an open platform that allows gamers to play the games they own instantly in the cloud on any PC or Mac, anywhere they like. The service currently has 300,000 monthly active users with one million more on the waitlist. To scale out to millions of gamers worldwide, we announced the GeForce NOW alliance, expanding GFN through partnerships with global telecom providers; SoftBank in Japan and LG UPlus in South Korea will be among the first to launch GFN later this year. NVIDIA will develop the software, manage the service, and share the subscription revenue with alliance partners. GFN runs on NVIDIA's edge computing servers as telcos race to offer new services for their 5G networks. GFN is an ideal new 5G application. Moving to the data center, revenue was $634 million, down 10% year-on-year and down 7% sequentially, reflecting the pause in hyperscale spending. While demand from some hyperscale customers bounced back nicely, others paused or cut back. Despite the uneven demand backdrop, the quarter had significant positives consistent with the growth drivers we outlined on our previous earnings call. First, inference revenue was up sharply, both year-on-year and sequentially, with broad-based adoption across a number of hyperscale and consumer internet companies. As announced at GTC, Amazon and Alibaba joined other hyperscalers such as Google, Baidu, and Tencent in adopting the T4 in their data centers. A growing list of consumer internet companies is also adopting our GPUs for inference, including LinkedIn, Expedia, Microsoft, PayPal, Pinterest, Snap, and Twitter. The contribution of inference to our data center revenue is now well into the double-digit percent. Second, we expanded our reach in enterprise, teaming up with major OEMs to introduce the T4 enterprise and edge computing servers. These are optimized to run NVIDIA's CUDA-X AI acceleration libraries for AI and data analytics. Within the easy-to-deploy software stack from NVIDIA and our ecosystem partners, this wave of NVIDIA edge AI computing systems enables companies in the world's largest industries—transportation, manufacturing, industrial, retail, healthcare, and agriculture—to bring intelligence to the edge where the customers operate. And third, we made significant progress in data center rendering and graphics. We unveiled a new RTX server configuration packing 40 GPUs into an 8U space and up to 32 servers in a pod, providing unparalleled density, efficiency, and scalability. With a complete stack, this server design is optimized for three data center graphic workflows: rendering, remote workstation, and cloud gaming. The rendering opportunity is starting to take shape with early RTX server deployment at leading studios, including Disney, Pixar, and several others. In the quarter, we announced our pending acquisition of Mellanox for $125 per share in cash, representing a total enterprise value of approximately $6.9 billion, which we believe will strengthen our strategic position in the data center. Once complete, the acquisition will unite two of the world's leading companies in high-performance computing. Together, NVIDIA's computing platform and Mellanox's interconnects power over 250 of the world's top 500 supercomputers and have customers among every major cloud service provider and computer maker. Data centers in the future will be architected as giant compute engines with tens of thousands of compute nodes, designed holistically with their interconnects for optimal performance. With Mellanox, NVIDIA will optimize data center scale workloads across the entire computing, networking, and storage stack to achieve higher performance, greater utilization, and lower operating costs for customers. Together, we can create better AI computing systems from the cloud to the enterprise to the edge. As stated at the time of the announcement, we look forward to closing the acquisition by the end of this calendar year. Moving to pro visualization, revenue reached $266 million, up 6% from the prior year and down 9% sequentially. Year-on-year growth was driven by both desktop and mobile workstations, while the sequential decline was largely seasonal. Areas of strength included the public sector, oil and gas, and manufacturing. Emerging applications such as AI, AR, and VR contributed an estimated 38% of pro visualization revenue. The real-time ray tracing capabilities of RTX are a game changer for the visual effects industry, and we are seeing tremendous momentum in the ecosystem. At UTC, we announced that the world's top 3D application providers have adopted NVIDIA RTX in their product releases set for later this year, including Adobe, Autodesk, Chaos Group, Dassault, and Pixar. With this rich software ecosystem, NVIDIA RTX is transforming the 3D market. For example, Pixar is using NVIDIA RTX ray tracing on its upcoming films, while digital is using it for upcoming Disney projects, and Siemens and other ray trace studios will be able to generate rendered images up to 4 times faster in their product design workflows. We are excited to see the tremendous value that NVIDIA RTX is bringing to the millions of creators and designers served by our ecosystem partners. Finally, turning to automotive, Q1 revenue was $166 million, up 14% from a year ago and up 2% sequentially. Year-on-year growth was driven by growing adoption of next-generation AI cockpit solutions and autonomous vehicle development deals. At GTC, we had major customer and product announcements. Toyota selected NVIDIA's end-to-end platform to develop, train, and validate self-driving vehicles. This broad partnership includes advancements in AI computing, infrastructure using NVIDIA GPUs, simulation using NVIDIA Drive Constellation platform, and in-car AV computers based on the DRIVE AGX Xavier or Pegasus. We also announced the public availability of Drive Constellation, which enables millions of miles to be driven in virtual worlds across a broad range of scenarios, with greater efficiency, cost-effectiveness, and safety than what's possible to achieve in the real world. Constellation will be reported in our data center market platform. And we introduced NVIDIA Safety Force Field, a computational defensive driving framework that shields autonomous vehicles from collisions, mathematically verified and validated in simulation. Safety Force Field will prevent a vehicle from creating or contributing to an unsafe starting situation. We continue to believe that every vehicle will have autonomous capability one day, whether with a driver or driverless. To help make that vision a reality, NVIDIA created an end-to-end platform for autonomous vehicles, from AI computing infrastructure to simulation to in-car computing, and Toyota is our first major win that validates the strategy. We see this as a $30 billion addressable market by 2025. Moving to the rest of the P&L and balance sheet, Q1 GAAP gross margins was 58.4% and non-GAAP was 59%, down year-on-year due to lower gaming margins and mix up sequentially from Q4 which had a $128 million charge from DRAM boards and other components. GAAP operating expenses were $938 million and non-GAAP operating expenses were $753 million, up 21% and 16% year-on-year respectively. We remain on track for high single-digit OpEx growth in fiscal 2020 while continuing to invest in the key platforms driving our long-term growth, namely graphics, AI, and self-driving cars. GAAP EPS was $0.64 and non-GAAP EPS was $0.88. We did not make any stock repurchases in the quarter following the announcement of the pending Mellanox acquisition. We remain committed to returning $3 billion to shareholders through the end of fiscal 2020 in the form of dividends and repurchases. So far, we have returned $800 million through share repurchases and quarterly cash dividends. With that, let me turn to the outlook for the second quarter of fiscal 2020. While we anticipate substantial quarter-over-quarter growth for Q2, the outlook is somewhat lower than our expectations earlier in the quarter when our outlook for fiscal 2020 revenue was flat to down slightly from fiscal 2019. The data center spending pause around the world will likely persist in the second quarter, and visibility remains low. In gaming, the CPU shortage, while improving, will affect the initial round of our laptop business. For Q2, we expect revenue to be $2.55 billion, plus or minus 2%. We expect a stronger second half than the first half, and we are returning to our practice of providing revenue outlook one quarter at a time. Q2 GAAP and non-GAAP gross margins are expected to be 59.2% and 59.5%, respectively, plus or minus 50 basis points. GAAP and non-GAAP operating expenses are expected to be approximately $985 million and $765 million, respectively. GAAP and non-GAAP OIME or both expected to be income of approximately $27 million. GAAP and non-GAAP tax rates are both expected to be 10%, plus or minus 1%, excluding discrete items. Capital expenditures are expected to be approximately $120 million to $140 million. Further financial details are included in the CFO commentary and other information available on our IR website. In closing, let me highlight upcoming events for the financial community. We'll be presenting at the Bank of America Global Technology Conference on June 5; at the RBC Future of Mobility Conference on June 6; and at the NASDAQ Investor Conference on June 13. Our next earnings call to discuss financial results for the second quarter of fiscal 2020 will take place on August 15. We will now open the call for questions. Operator, will you please poll.

Yes. Thanks for taking the question. Colette, I was wondering if you could give a little bit more color or discussion around what exactly you've seen in the data center segment. And whether or not, or what you're looking for in terms of signs that we can kind of return to growth or maybe this pause is behind it. I guess what I'm really asking is kind of what's changed over the last, let's call it three months relative to your prior commentary from a visibility perspective and just demand perspective within that segment.

Sure, thanks for the question. When we discussed our data center business three months ago, we mentioned that our visibility entering the new calendar year was low. We faced challenges in finalizing some deals at the end of that quarter. However, as we moved into Q1, we felt confident about how we wrapped things up. We observed a mix of progress, continuing our capital expenditures and building out what is necessary for the data centers, while some others are still on pause. Looking ahead to Q2, we see a continuation of our current visibility, which remains not ideal but still solid regarding the benefits we believe our platform provides. Our priorities align with the needs of hyperscale and enterprise clients as they incorporate AI into various workloads. We will need to see how things develop moving forward, but for now, our visibility is about the same as it was three months ago.

Okay. And then, as a quick follow-up on the gaming side last quarter you talked about that being down—I think it was termed as being down slightly for the full year. Is that still the expectation or how has that changed?

So, at this time, we don't plan on giving a full year overall guidance. I think our look in terms of gaming—all of the still drivers that we thought about earlier in the quarter and we talked about at our Investor Day and we have continued to talk about are still definitely in line. While the drivers of our gaming business and Turing RTX for the future are still on track. But, we're not providing guidance at this time for the full year.

Good afternoon. Thanks for taking my question. On the last earnings call, you had mentioned China gaining demand is a headwind. At the Analyst Day in mid-March, I think Jen-Hsun had mentioned that the team was already starting to see better demand trends out of China, maybe given the relaxed stance on gaming bans. Do you anticipate continued China gaming demand on a go-forward basis and maybe talk about some of the dynamics driving that demand profile in the China geography?

Sure. China looks fine. I think China has stabilized. The gaming market in China is really vibrant and continues to be vibrant. Tencent's releasing new games. I think you might have heard that Epic's store is now open in Asia, and games are available from the West. So, there are all kinds of positive signs in China. There are some 300 million PC gamers in China, and people are expecting it to grow. We're expecting the total number of gamers to continue to grow from the 1 billion plus PC gamers around the world to something more than that. And so, things look fine.

Thanks for that. And then, as a follow-up, a big part of the demand profile in the second half of the year for the gaming business is always the lineup of AAA rated games. Obviously, you guys have a very close partnership with all of the game developers. How does the pipeline of new games look, kind of they get launched in the October-November timeframe—either a total number of blockbuster games and also games supporting real-time ray tracing as well as some of your DLSS capabilities?

Yes. Well, it's seasonal. In the second half of the year, we expect to see some great games. We won't preannounce anybody else's games for them. But this is a great PC cycle because it is the end of the console cycle. And PCs are where the action's at these days. With Battle Royale and Esports and so much social interaction going on, the PC gaming ecosystem is just really vibrant. Our strategy with RTX was to take a lead and move the world to ray tracing. And at this point, I think it's fairly safe to say that that the leadership position that we've taken has turned into a movement that has turned next-generation gaming ray tracing into a standard. Almost every single game platform will have to have ray tracing, and some of them have already announced it. The partnerships that we've developed are fantastic. Microsoft DXR is supporting ray tracing, Unity is supporting ray tracing, Epic is supporting ray tracing, leading publishers like EA have adopted RTX and are supporting ray tracing, and movie studios, including Pixar, have announced that they are using RTX and will use it to accelerate their rendering of films. And so, I think at this point it's fair to say that ray tracing is the next generation and it's going to be adopted all over the world.

Thank you. I guess the first question is for Colette. So, what went into the decision to pull for your guidance versus just cutting it? Is it really just fear around how long it could take for data center to come back? Thank you.

Yes. I'll start off here and kind of go back to where our thoughts were in Q1 and why we provided full-year guidance when we were in Q1. When we looked at Q1 and what we were guiding, we understood that it was certainly an extraordinary quarter, something that we didn't feel was a true representative of our business. And we wanted to get a better view of our trajectory of our business in terms of going forward. We are still experiencing uncertainty as a result of the pause with the overall hyperscale data centers. We do believe that's going to extend into Q2. However, we do know and expect that our Q2—assuming our H2 will likely be sizably larger than our overall H1. And the core dynamics of our business at every level is exactly what we expected. That said, we're going to return to just quarterly guidance at this time.

Okay. Thanks. And then, just as a follow-up, can you give us some even qualitative if not quantitative sense of the $320 million incremental revenue for July. How that breaks out? Is the thinking sort of that data center is going to be flat to maybe up a little bit and pretty much the remainder of the growth comes from gaming? Thanks.

Yes. So, when you think about our growth between Q1 and Q2, yes, we do expect our gaming to increase. We do expect our Nintendo Switch to start again in sizable amount. Once we move into Q2, and we do at this time expect probably our data center business to grow.

Thanks for taking the question. Jen-Hsun, I had a follow-up on the data center business. I was hoping you could provide some color in terms of what you're seeing not only from your hyperscale customers, which you've talked extensively on, but more on the enterprise and the HP side of your business, and specifically on the hyperscale side. You guys talk about this pause that you're seeing from your customer base. In your conversations with your customers, did they give you a reason as to why they're pausing? Is it too much inventory of GPUs and CPUs and so forth? Or is it optimization giving them extra capacity? Is it caution on their own business going forward? Or is it a combination of all the above? Any color on that would be helpful too. Thank you.

Hyperscalers are digesting the capacity they have. At this point, I think it's fairly clear that in the second half of last year, they took on a little bit too much capacity. And so, everybody has paused to give themselves a chance to digest. However, our business on inference is doing great. And we're working with CSPs all over the world to accelerate their inference models. Now, the reason why recently the inference activity has gotten just off the charts is because of breakthroughs in what we call conversational AI. In fact, today I think I just saw it, but have known about this work for some time. Harry Shum's Group, Microsoft AI Research Group, announced their multitask DNN general language understanding model today and it broke benchmark records all over the place. Basically, what this means is the three fundamental components of conversational AI, which are speech recognition, natural language understanding—this multitask DNN is a breakthrough in and it's based on a piece of work that Google did recently called BERT—and text to speech. All of the major pieces of conversational AI are now put together. Of course, it's going to continue to evolve, but these models are gigantic to train. In the case of Microsoft's network, it was trained on vault to GPUs, and these systems require large amounts of memory, as the models are enormous and take an enormous amount of time to train these systems. So, we're seeing a breakthrough in conversational AI, and across the board, Internet companies would like to make their AI much more conversational so that you can access through phones and smart speakers and engage AI practically everywhere. The work that we're doing in industries makes a ton of sense. We're seeing AI adoption in industries from transportation to healthcare to retail to logistics, industrials, and agriculture. The reason for that is that they have a vast amount of data that they're collecting. I heard a statistic the other day that some 90% of today's data was created just two years ago and it's being created by the industrial systems all over the world. If you want to put that data to work, you can create the models using our systems, our GPUs for training, and you can extend that all the way out to the edge. This last quarter, we started to talk about our enterprise server based on T4. This inference engine that has been really successful for us at the CSPs is now going out into edge and enterprise servers. These edge systems are going to do AI basically instantaneously. It's too much data to move to the cloud. You might have data sovereignty concerns—you want to have very low latency—maybe it needs to have multi-sensor fusion capabilities so that it understands the context better. For example, what it sees and what it hears has to be harmonious. So, you need that kind of AI, those kinds of sensor computing at the edge. We're seeing a ton of excitement around this area. Some people call it the intelligent edge, some people call it edge computing, and now, with 5G networks coming, we're seeing a lot of interest around the edge computing servers that we're making. Those are the activities we are seeing.

Thank you. As a quick follow-up on the gaming side, Colette, can you characterize product mix within gaming? You cited mix as one of the key reasons why gross margins were down year-on-year, albeit off a high base going into Q2 and the back half. Would you expect SKU mix within gaming to improve or stay the same? I ask because it's important for gross margins, obviously. Thank you.

Yes. When you look at our sequential gross margin increase, that will be influenced by our larger revenue and better mix, which you're correct is our largest driver of our gross margin. However, we will be beginning the Nintendo Switch back up, and that does have lower gross margins than the company average, influencing our Q2 gross margin guidance that we provided. As we look forward towards the rest of the year, we think mix and the higher revenue again will influence and likely raise our overall gross margins for the full year.

Great. Thank you. We've talked quite a bit about GeForce NOW in the prepared remarks and at the Analyst Day. It seems like cloud gaming is going to be a big topic. Is that going to be your preferred way to go to market with cloud gaming, and do you expect to sell GPUs to traditional cloud vendors in a non-GeForce NOW fashion?

Yes. Our strategy for cloud gaming is to extend our PC position for GeForce gamers into the cloud. Our strategy for building out our network is partnerships with telcos around the world. We'll build out some of it, and on top of the service, we have our entire PC gaming stack, and when we host the service, we will move toward a subscription model. With our telcos around the world, who would like to provision the service at their edge servers, many of them would like to do so in conjunction with their 5G telco services to offer cloud gaming as a differentiator. In all these different countries where PC exposure has been relatively low, we have an opportunity to extend our platform out to that billion PC gamers. And so, that's our basic strategy. We also offer our edge server platform to all of the cloud service providers. Google has NVIDIA GPU graphics in the cloud, Amazon has NVIDIA GPU graphics in the cloud, and Microsoft has NVIDIA GPU graphics in the cloud. These GPUs will be fantastic also for cloud gaming, workstation graphics, and ray tracing. The platform is capable of running all of the things that NVIDIA runs, and we try to put it in every data center in every cloud from every region that's possible.

Thanks for taking my question. I actually had a clarification for Colette and a question for Jen-Hsun. Colette, are you now satisfied that the PC gaming business is operating at normal levels? When you look at Q2 guidance, are all the issues regarding inventory and ratios behind us? Or do you think that the second half of the year is more than normalized run rate for your PC gaming business? And then, Jen-Hsun, on the data center—NVIDIA has dominated the training market; inference sounds a lot more fragmented and competitive. There's a lot of talk about software being written more on the framework level. How should we get the confidence that your lead in training will help you maintain a good lead in inference also? Thank you.

Thanks for the question. So, let's start with your first part of the question regarding how we reached overall normalized gaming levels. When we look at our overall inventory in the channel, we believe that this is relatively behind us and moving forward that it will not be an issue. Going forward, we will probably reach normalized levels for gaming somewhere between Q2 and Q3, similar to our discussion that we had back at Analyst Day and at the beginning of the quarter.

NVIDIA strategy is accelerated computing. It is very different than an accelerator strategy. For example, if you were building a smart microphone, you need an accelerator for speech recognition ASR. Our company is focused on accelerated computing. The reason for that is because the world's body of software is really gigantic, and it continues to evolve. AI is nowhere near done; we're probably at the first couple of innings of AI. The amount of software and the size of the models are going to have to continue to evolve. Our accelerated computing platform is designed to enable the computer industry to bring forward into the future all the software that exists today, whether it's TensorFlow, Caffe, PyTorch, or classical machine learning algorithms like XGBoost, which is right now the most popular framework in machine learning overall. There are so many different types of classical algorithms, not to mention all of the handwritten engineered algorithms by programmers. Those algorithms and those hand-engineered algorithms would like to be mixed in with all of the deep learning or otherwise classical machine learning algorithms. This whole body of software doesn't run on a single function accelerator. If you would like the body of software to run on something, it would have to be sufficiently general purpose. The balance that we made was we invented this thing called a Tensor Core that allows us to accelerate deep learning to the speed of light. Meanwhile, it has the flexibility of CUDA, so that we can bring forward everything in classical machine learning, as people have started to see with RAPIDS, and it's being announced being integrated into machine learning pipelines in the cloud and elsewhere. Our company is focused on accelerated computing, and speaking of inference, that's one of the reasons why we're so successful in inference right now. We're seeing really great pickup. The type of models that people want to run on one application, let's just use one very exciting application, conversational AI. You would have to do speech recognition, you would have to do natural language understanding to understand the speech. You might have to convert, you have to translate to another language. Then, you have to do something related to maybe making a recommendation or making a search, and then you have to convert that recommendation and search and the intent into speech. While some of it could be 8-bit integer, some of it really wants to be 16-bit floating-point, and some of it, because of the development state of it, may want to be in 32-bit floating-point. The mix-precision nature and the computational algorithm nature, the flexibility nature of our approach make it possible for cloud providers and people who are developing AI applications to not have to worry about exactly what model it runs or not; we run every single model. If it doesn't currently run well, we'll help you make it run. The flexibility of our architecture and the incredible performance in deep learning is a great balance and allows customers to deploy it easily. So, our strategy is very different than an accelerator. I think the only accelerators that I really see successful at the moment are the ones that go into smart speakers. There are a whole bunch being talked about, but I think the real challenge is how to make it run real workloads. We're going to keep cranking along in our current strategy and keep raising the bar as we have in the past.

Hi, guys. Thanks for taking my question. This is a question for Colette. Colette, you said inference and rendering within data center were both up very strongly, but I guess that has to imply that the training flash acceleration pieces is quite weak, even weaker than the overall. Given that these should be adding to efficiency, I'm just surprised it's down that much. Is this truly just digestion? I mean is it share? I mean like your competitor is now shipping some parts here? I guess how do we get confidence that just we haven't seen a ceiling on this? I mean do you think, given the trajectory, you can exit the year above the prior peaks? I guess maybe just any color you could give us on any of those trends would be super helpful.

So, as we discussed, Stacy, we are targeting many of the hyperscale definitely purchasing in terms of the inferencing into the installment that continues. Also in terms of the training; the training instances that they will need for the cloud or for internal use, absolutely. We have some that have pods and going through all those periods, so that we do believe because this will come back. We do believe as we look out into the future that they will need that overall deep learning for much of their research as well as many of their workloads. So, no concern on that. But right now, we do see a pause. I will turn it over to Jen-Hsun to see if he has additional comments.

Let's see. I think that when it comes down to training, if your infrastructure team tells you not to buy anything, the thing that suffers is time to market and some amount of experimentation that allows you to better pause and wait a little longer. I should mention that for computer vision type algorithms and recommendation type algorithms, that posture may not be impossible. However, the type of work that everybody is now jumping on top of, which is natural language understanding and conversational AI, and the breakthrough that Microsoft just announced. If you want to keep up with that, you're going to have to buy much larger machines. I'm looking forward to that, and I expect that that's going to happen. But in the latter part of last year, Q4 and Q1 of this year, we did see a pause from the hyperscalers. I don't expect it to last. The acquisition will enable data centers around the world, whether it's U.S. or China, to advance much more quickly. Now we're going to invest in building infrastructure technology, and as a combined company, we'll be able to do that much better. This is good for customers, and it's great for customers in China. The two matters, whether it's the acquisition and competition in the market, or whether it relates to trade, are different. In our case, we bring so much value to the marketplace in China. I'm confident that the market will see that.

Yes. Good afternoon and thank you for taking my question. I guess a question on the non-cloud part of your data center business. If you think about the trends you're seeing in enterprise virtualization and HPC, and all the work you're doing around RAPIDS, rendering, etc. Can you kind of talk through the visibility you have today for that part of your business? I think that's roughly 50% of the mix, so is that a piece that you feel confident can grow in 2019, and any color around that would be appreciated.

We expect to grow in 2019. A lot of our T4 inference work is related to what people call edge computing, and it has to be done at the edge because the amount of data that otherwise would be transferred to the cloud is just too much. It has to be done at the edge because of data sovereignty issues and data privacy issues. And it has to be done at the edge because the latency requirement is really high; it has to respond basically like a reflex to make a prediction or make a suggestion or stop a piece of machinery instantaneously. A lot of that work that we're doing in T4 inference is partly in the cloud, but a lot of it is also at the edge. T4 servers for enterprise were announced, I guess about halfway through the quarter, and the OEMs are super excited about that because the number of companies in the world who want to do predictive data analytics is quite large. The size of the data is growing significantly, and with Moore's law ending, it's really hard to power through terabytes of data at a time. We've been working on building the software stack from the new memory architectures and storage architectures all the way to the computational middleware. The activity in GitHub is fantastic, with all kinds of companies jumping in to make contributions because they would like to take that open-source software and run it in their own data center on our GPUs. I expect the enterprise side of our business, both for enterprise big data analytics and edge computing, to be a really good growth driver for us this year. Our Automotive Strategy has several components. There's the engineering component where our engineers and their engineers have to co-develop autonomous vehicles. Then there are three other components: AI computing infrastructure, which we call DGX, and our OEM servers that include our GPUs that are used for developing the AIs. The cars are collecting a couple of terabytes per day per test car, and all of that data has to be powered through and crunched through in the data center. We have an infrastructure called DGX that people could use, and we're seeing a lot of success there. We recently announced a new infrastructure called Constellation that lets you essentially drive thousands and thousands of test cars in your data center. They're all going through random or directed scenarios that allow you to either test untestable scenarios or regress against previous scenarios that we call Constellation. Lastly, after working on a car for several years, we install the computer inside the car, which we call DRIVE. These are the four components of opportunities that we have in the automotive industry. We're doing great in China, where a whole bunch of electric vehicles are being created and robot taxi developments around the world are largely using NVIDIA technology. We recently announced a partnership with Toyota. There's a whole bunch of stuff that we're working on, and I'm anxious to announce them to you. This is an area that is the tip of the iceberg of a larger space we call robotics and computing at the edge. If you think about the basic computational pipeline of a self-driving car, it's no different from smart retail or the future of computational medical instruments. Robotics, industrial inspection, delivery drones all essentially use the same technique. This foundational work we do is for the larger space that people call intelligent edge computing.

Thank you. Good afternoon. First question is on notebooks. Just to clarify what's been different from your expectations this year? Is it simply that the OEMs didn't launch the new models you'd expected given the shortage, or is it more just about unit volume? Following up on that, what's your level of confidence that coming back will be a driver as you go into the second half of the year?

In Q2, we had to deal with some CPU shortage issues at the OEMs. It's improving, but the initial ramp will be affected. The CPU shortage situation has been described fairly broadly, and that's affected our initial ramp. We don't expect it to affect our ramp going forward. The new category of gaming notebooks that we created called Max-Q has made it possible for really amazing gaming performance to fit into a thin and light form factor. These new generations of notebooks with our Max-Q design and Turing GPU—super energy-efficient in combination—make it possible for OEMs to create notebooks that are both affordable, all the way down to $799, and thin and really delightful, up to something incredible with an RTX 2080 and a 4K display. These are thin notebooks that are really beautiful that people would love to use. The invention of the Max-Q design method and all the software that went into it that we announced last year—last year, we had some 40 notebooks or so, maybe a little bit less than that, and this year we have around 100 notebooks that are being designed at different price segments by different OEMs across different regions. I think this year is going to be quite successful for notebooks. It's also the fastest-growing segment of consumer PCs. It's very largely under-penetrated because, until Max-Q came along, it wasn't really possible to design a notebook that is both great in performance and experience and something that a gamer would like to own. Finally, we've been able to solve that difficult puzzle and create powerful gaming machines that are inside notebooks that are really wonderful to own and carry around. This is going to be a really fast-growing segment, and all the OEMs know it, which is why they put so much energy into creating all these different types of designs and styles and sizes and shapes. We have 100 Turing GPU notebooks ramping right now. In Q2, we'll have to see when we start dealing with issues of strengthening demand, based on our expectations for the second half. It should be powerful in the automotive market, and we should expect significant ramp there. For level 2+ systems, we anticipate more robust growth going into 2020.

Thank you very much. Good afternoon. I have two questions—one for Jen-Hsun and one for Colette. I guess, Jen-Hsun, you've said in many forums that moving down to the new process node of 7nm across the business was not really sufficient to have a platform approach, and I agree with that. Could you talk a little bit about your product plans, at least in general terms, around 7nm franchising in the gaming business and in your training accelerator program? I wonder if that might be waiting for some of those products or the anticipation of those might be the cause of a little bit of a pause here. Secondly, Colette, maybe you could talk us through your expectations. I understand there's a lack of visibility in certain parts of the business on revenue that maybe you could talk about OpEx trends through the rest of the year, where you might have a little more visibility. Thank you.

The entire reason for Q4 and Q1 is attributed to oversupply in the channel as a result of cryptocurrency; it has nothing to do with Turing, in fact. Turing is off to a faster start than Pascal and it continues to be on a faster pace than Pascal. The pause in gaming is now behind us. We're on a growth trajectory with gaming. RTX has taken the lead on ray tracing and will become the standard for next generation gaming. Support from basically every major platform and software provider on the planet is really strong. Our notebook growth is going to be great because of the Max-Q design that we invented, and the last couple of quarters have also intersected with the seasonal slowdown; however, we’re getting back into normal build cycles. Outreach across various designs will begin to impact us between Q2 and Q3.

We remain on track to our thoughts on leaving the fiscal year with year-over-year growth and overall OpEx on a non-GAAP basis in a high single digit. We'll see probably an increase sequentially quarter-to-quarter, but our year-over-year growth is expected to decline as we will not be growing at the speed that we did in this past year. However, we do believe we're on track to meet that goal.

Thanks, everyone. We're glad to be returning to growth. We are focused on driving three growth strategies. First, RTX ray tracing. It's now clear that ray tracing is the future of gaming and digital design, and NVIDIA RTX is leading the way with support from Microsoft DXR, Epic, Unity, Adobe, and Autodesk. Game publishers like EA and movie studios like Pixar—industry support has been fantastic. Second, accelerated computing and AI computing—the pause in hyperscale spending will pass. Accelerated computing and AI are the greatest forces in computing today, and NVIDIA is leading these movements, whether cloud or enterprise or AI at the edge for 5G. NVIDIA's one scalable architecture from cloud to edge is a focal point platform for the industry to build AI upon. Third, robotics, some call it embedded AI, some edge AI, or autonomous machines. The same computing architecture is used for self-driving cars, pick and place robotics arms, delivery drones, and smart retail stores. Every machine that moves or machines that watch other things that move, whether with a driver or driverless, will have robotics and AI capabilities. Our strategy is to create an end-to-end platform that spans NVIDIA's DGX AI computing infrastructure to NVIDIA Constellation simulation to NVIDIA AGX embedded AI computing. We're also super excited about the pending acquisition of Mellanox. Together, we can advance cloud and edge architectures for HPC and AI computing. See you next quarter.