Earnings Call: Q4 2017

Feb 9, 2017

Good afternoon. My name is Victoria, and I'm your conference operator for today. Welcome to NVIDIA's Financial Results Conference Call. All lines have been placed on mute. After the speakers' remarks, there will be a question and answer period. Thank you. I'll now turn the call over to Arnab Chanda, Vice President of Investor Relations to begin your conference. Thank you. Good afternoon, everyone, and welcome to NVIDIA's conference call for the Q4 fiscal 2017. With me on the call today from NVIDIA are Jensen 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. It's also being recorded. You can hear a replay by telephone until the 16th February 2017. The webcast will be available for replay up until next quarter's conference call to discuss Q1 financial results. The content of today's call is NVIDIA's property. It cannot be replaced 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 ks and 10 Q, and the reports that we may file on Form 8 ks with the Securities and Exchange Commission. All our statements are made as of today, 9th February, 2017, 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, Arnab. We had a stellar Q4 and fiscal 2017 with records in all of our financial metrics: revenue, gross margin, operating margins and EPS. Growth was driven primarily by data center tripling with a rapid adoption of AI worldwide. Quarterly revenue reached $2,170,000,000 up 55% from a year earlier and up 8% sequentially and above our outlook of $2,100,000,000 Fiscal 2017 revenue was just over $6,900,000,000 dollars up 38% and nearly $2,000,000,000 more than fiscal 2016. Growth for the quarter and fiscal year was broad based record revenue in each of our 4 platforms: gaming, professional visualization, data center and automotive. Our full year performance demonstrates the success of our GPU platform based business model. From a reporting segment perspective, Q4 GPU revenue grew 57 percent to $1,850,000,000 from a year earlier. Tegra processor revenue was up 64% to $257,000,000 Let's start with our gaming platform. Q4 gaming record revenue was a record $1,350,000,000 rising 66% year on year and up 8% from Q3. Gamers continue to upgrade to our new Pascal based GPUs. Adding to our gaming lineup, we launched GTX 10 50 class GPUs for notebooks, bringing esports and VR capabilities to mobile at great value. The GTX 1050 1050ti were featured in more than 30 new models launched at last month's Consumer Electronics Show. To enhance the gaming experience, we announced G SYNC HDR, a technology that enables displays which are brighter and more vibrant than any other gaming monitor. Our partners have launched more than 60 G SYNC capable monitors and laptops, enabling smooth play without screen terror artifacts. Esports 2 continues to attract new gamers. Major tournaments with multimillion dollar purses are drawing enormous audiences. This last quarter, DOTA 2 held its 1st major tournament of the season in Boston. Tickets sold out in minutes. The prize pool reached 3,000,000 and millions of gamers watched online. Moving to professional visualization. Quadro revenue grew 11% from a year ago to a record $225,000,000 driven by demand for high end real time rendering and mobile workstations. We recently launched a family of Pascal based GPUs designed for mobile workstations, which leading OEMs are embracing. Earlier this week, we introduced Quadro GP100, which creates a new supercomputing workstation. This new type of workstation enables engineers, designers and artists to take advantage of new technologies of photorealism, fluid simulation and deep learning. Next, data center. Revenue more than tripled from a year ago and was up 23% sequentially to $296,000,000 Growth was driven by AI, cloud service providers deploying GPU instances, high performance computing, grid graphics virtualization and our DGX AI supercomputing appliance. AI is transforming industries worldwide. The first adopters were hyperscale companies like Microsoft, Facebook and Google, which use deep learning to provide billions to customers with AI services that utilizes image recognition and voice processing. The next area of growth will occur as enterprise in such fields as healthcare, retail, transportation and finance embrace deep learning on GPUs. At November's SC 16 Supercomputing Conference, Microsoft announced that its GPU accelerated Microsoft Cognitive Toolkit is available both in Azure Cloud and on premises with our DGX-one AI supercomputer. In a series of related announcements at SC-sixteen, we described our plans to join the Cancer Moonshot project in conjunction with the National Cancer Institute, the U. S. Department of Energy and several national labs. To help build predictive models and guide treatment under this project, we are collaborating on a new AI framework called CANDL, the Cancer Distributed Learning Environment. And to support this work, we unveiled our new own supercomputer, the NVIDIA DGX Saturn V, which joins together 124 DGX-1 systems. It's currently the world's 28th fastest supercomputer and the number one system in energy efficiency. Our grid graphics virtualization business doubled year on year, driven by strong growth in the education, automotive and energy sectors. We are excited to be hosting our 8th Annual GPU Technology Conference here in Silicon Valley from May 8 to May 11. This will be the year's most important event for AI and accelerating computing. And we expect it to be our largest DTC yet attended by thousands of application developers, scientists and academics, as well as entrepreneurs and corporate executives. Finally, in automotive, revenue grew to a record $128,000,000 up 38% year over year. At Jensen's CES opening keynote, we demonstrated our leadership position in self driving vehicles. With a growing list of industry players adopting our AI car platform, we also showcased AI CoPilot, a technology that will recognize a driver in their preferences, monitor their alertness, understand natural spoken language and provide alerts to dangerous situations. One of the highlights of CES was the demonstration of our own autonomous car dubbed BB8. More than 500 passengers took rides in the back seat without a driver behind the wheel. We announced a number of new partnerships at the show. Among them were collaborations with Bosch, the world's largest automotive supplier and ZF, Europe's leading supplier for the truck industry, both center on developing AI car computers with DRIVE PX2 technology. We also announced that we're working on cloud to car mapping collaborations with HERE, focused on the U. S. And Europe and Zenren focused on Japan. These complement partnerships announced in Q3 with Europe's TomTom and China's Baidu. Our mapping partnerships now span all geographies. Jensen was joined on the CES stage by Audi of America's President, Scott Keogh. They announced the extension of our decade long partnership to deliver cars with level 4 autonomy starting in 2020, powered by DRIVE PX technology. Audi will deliver level 3 autonomy in its A8 luxury sedan later this year through its Z fast system powered by NVIDIA. We also shared news at CES of our partnership with Mercedes Benz to collaborate on a car that will be available by year's end. During the quarter, Tesla began delivering a new autopilot system powered by the NVIDIA DRIVE PX2 platform in every new Model S and Model X to be followed by the Model 3. Tesla's cars will be capable of fully autonomous operation via future software updates. In addition, Volvo started turning over the keys to initial customers of its Drive Me program. Its XC90 SUVs equipped with Drive PX2 are capable of fully autonomous operation on designated roads in Volvo's hometown of Gothenburg, Sweden. With Invidius powering the market's only self driving cars and partnerships with leading automakers, Tier 1 suppliers and mapping companies, we feel very confident in our position as the transportation industry moves to autonomous vehicles. Next, our OEM and IP business was $176,000,000 down 11% year on year. Now turning to the rest of the income statement for Q4. Gross margins were at record levels, with GAAP gross margins at 60% and non GAAP at 60.2%. These reflect the success of our platform approach, as well as strong demand for GeForce gaming GPUs and deep learning. GAAP operating expenses were 570,000,000 dollars Non GAAP operating expenses were $498,000,000 up 12% from a year earlier, reflecting headcount related costs for our AI growth initiatives as well as investments in sales and marketing. We are investing into huge market opportunities, AI, self driving cars, cloud computing and gaming. Thus, we expect our operating expense growth rate to be in the high teens over the next several quarters. GAAP operating income was $733,000,000 and non GAAP operating income was $809,000,000 both more than doubled from a year earlier. Our GAAP tax rate was 10% and our non GAAP was 13%. These rates were lower than expected primarily due to a decrease in the amount of earnings subject to U. S. Tax. GAAP EPS was $0.99 Non GAAP EPS was $1.13 In fiscal year 2017, we returned $1,000,000,000 to shareholders through dividends and share repurchases in line with our intentions. For fiscal year 2018, we intend to return $1,250,000,000 to shareholders through dividends and share repurchases. Now turning to the outlook for the Q1 of fiscal 2018. We expect revenue to be $1,900,000,000 plus or minus 2%. At the midpoint, this represents 46% growth over the prior year. We expect data center to grow sequentially. Our GAAP and non GAAP gross margins are expected to be 59 point 5% 59.7%, respectively, plus or minus 50 basis points. This guidance assumes that our licensing agreement with Intel ends at March and does not renew. GAAP operating expenses are expected to be approximately $603,000,000 Non GAAP operating expenses are expected to be approximately 520,000,000 dollars GAAP OI and E is expected to be an expense of approximately $20,000,000 including additional charges from the early conversions of convertible notes. Non GAAP OA and E is expected to be an expense of approximately $4,000,000 GAAP and non GAAP tax rates for the Q1 of fiscal 2018 are both expected to be 17% plus or minus 1%, excluding any discrete items. With that, I'm going to turn it back for the operator, so we can open up for questions. Please limit your questions to just one. Operator, let's start with the questions. Certainly. Your first question comes from the line of C. J. Muse with Evercore. Can you hear me? Yes, my apologies. I stuck on a plane here. Great results. I guess, I was hoping to get a little more color on the data center side. Now that we've completed a full fiscal year 2017. Would love to get some clarity on the different moving parts and contributions there. And then I guess looking into 2018, how you see the growth unfolding thereafter? Thank you. Yes. C. J, first of all, thanks a lot. Well, the single biggest mover would have to be data center. I mean, when you look back on last year and when you look forward, there's a lot of reasons why data center business overall grew 3x, grew by a factor of 3. And so I would expect that to happen to continue. There are several elements of our data center business. There's the high performance computing part. There's the AI part. There's grid, which is graphics virtualization. There's cloud computing, which is providing our GPU platform up in the cloud for startups and enterprises and all kinds of external customers to be able to access in the cloud as well as a brand new AI supercomputing appliance that we created last year for anybody who would like to engage deep learning and AI, but don't have the skills or don't have the resources or don't have the desire to build their own high performance computing cluster. And so we integrated all of that with all of the complicated software stacks into an appliance that we maintain over to cloud. We call that DGX 1. And so these pieces, AI, high performance computing, cloud computing, grid and DGX, all in contribution contributed to our growth in data center quite substantially. And so my sense is that as we look forward to next year, we're going to continue to see that major trend. Of course, gaming was a very large and important factor and my expectation is that gaming is going to continue to do that. And then longer term, our position in self driving cars, I think, is becoming more and more clear to people over time. And I expect that self driving cars will be available on the road starting this year with early movers and no later than 2020 for Level 4 by the majors and you might even see some of them pull into 2019. And so those are some of the things that we're looking forward to. Your next question is from Vivek Arya with Bank of America. Thanks. I actually had one question for Jensen and one sort of clarification for Colette. So, Jensen, where are we in the gaming cycle? It's been very strong the last few years. What proportion of your base do you think has upgraded to PASCAL? And where does that usually peak before you launch your next generation products? And then for Colette, just inventory dollars and days ticked up, if you could give us some comment on that. And then just on OpEx productivity, you did a very good job last year, but this time you're saying OpEx would go up mid teens. Do you still think there is operating leverage in the model? Thank you. Well, let's say, we typically assume that we have an installed base of a couple of 100,000,000 GeForce gamers. And we've upgraded about 2 quarters of them as in 2 operating quarter out of 4 years. Takes about 3 to 4 years to upgrade the entire installed base. And we started ramping PASCAL, as you know, a few quarters ago. And our data would suggest that the upgrade cycle is going well and we have plenty to go. Thanks, Vivek. On your question on inventory, as you know, in many of our businesses, we are still carrying significant architectures and a broad list of different products for those architectures across. We feel comfortable with our level of inventory as we look forward into fiscal year 2018 and our sales going forward. Your second question was regarding OpEx and comparing it to where we finished in 2017 and moving into fiscal year 2018. We do have some great opportunities, large businesses for us to go capture the overall TAMs 4, And we are going to be continuing to invest in the data center, specifically in AI, self driving cars, as well as gaming. And so rather than a focus on what the specific operating margin is, we're going to focus primarily just on growing the overall TAM and capturing that TAM on the top line. Your next question comes from the line of Mark Lipacis from Jefferies. Thanks for taking my question. Question back on the data center, the growth was impressive. And I'm wondering, you mentioned that the hyperscale players really have embraced the products first. And I'm wondering if you could share with us to the extent that you think that they're embracing it for their own use or to the extent that they're deploying it for services such as machine learning as a service and enterprises are really kind of tapping into this also through the hyperscale guys. And I'm wondering if you could help you mentioned that the enterprise is where you expect to see embracing the technology next in healthcare, retail, transport, finance. And I'm wondering if you could share with us how you feel about that visibility, where you're getting that visibility from? Thank you. Well, on hyperscale, you're absolutely right that there's internal use for deep learning and then there's the hosting GPU in the cloud for external high performance computing use, which includes deep learning. Inside the hyperscalers, the early adopters are moving obviously very, very fast and but everybody has to follow. Everybody has to follow. Deep learning has proven to be too effective. And you guys everybody knows now that every hyperscaler in the world is investing very heavily in deep learning. And so my expectation is that over the next coming years, deep learning and AI would become the essential tool by which they do their computing. Now when they host it in the cloud, people on the cloud use it for a variety of applications. And one of the reasons why the NVIDIA GPU is such a great platform is because of its broad utility. We've been working on GPU computing now for coming up on 12 years. And industry after industry, our GPU computing architecture has been embraced for high performance computing, for data processing, for learning and such. And so when somebody hosted up in the cloud, for example, Amazon putting our GPUs up in the cloud, that instance has the ability to do molecular dynamics to deep learning training, to deep learning inferencing. Companies could use it for offloading their computation to start ups being able to build their company and build their application and then host it for 100 of millions of people to use. And so I think the hyperscalers are going to continue to adopt GPU both for internal consumption and cloud hosting for some time to come. And we're just in the beginning of that cycle. And that's one of the reasons why have quite a fair amount of enthusiasm around the growth here. You mentioned enterprise. And enterprise has all woken to the power of AI. And everybody understands that they have a treasure trove of data that they would like to find a way to discover insight from. In the case of real applications that we're engaging now, You could just imagine that in the transportation industry, in car companies creating self driving cars, one car company after another needs to take all of their road data and start to train their neural networks for their future self driving cars. And so they use our DGX or Tesla GPUs to train the networks, which is then used to run their cars running on DRIVE PX. So that's one application example. Another application example which is quite significant is going to be the future of processing all of the HD maps in the world. You guys might have seen that we announced at GTC this API SDK called Mapworks. Mapworks takes video information, video information that is recorded from a car and reconstructs the 3-dimensional terrain information from that live video. And so it has to do computer vision, 3 d reconstruction, has to determine and detect where the lanes are, the signs are, the lights are, and even some interesting three d features, maybe buildings and curves and such. And it would do that automatically and we need to process that for the world, for the planet. And you could just imagine how much video is being recorded today and how much data is being generated and how much inferencing, computer vision and 3 d reconstruction that has to be done. And our GPUs are really quite perfect for it. And so Mapworks runs on top of our GPUs and we're working with just about every mapping company in the world today to adopt Mapworks and to do HD processing for their maps. So that's another example. Medical imaging companies all over the world have recognized the importance of deep learning and their ability to detect cancer and retinopathy and the list of examples goes on and on. And so all the different modalities have now recognized the importance of deep learning and you're going to start to see one medical imaging company after another. The list of examples just keep on going. I mean, the fact of the matter is, at this point, deep learning and AI has really become how future software development is going to be done for a large number of industries. And that's the enthusiasm that we're seeing around the world. Your next question comes from the line of Atif Malik with Citigroup. Hi, thanks for taking my question and congratulations to the team on great results and guide. My first question is for Jensson. Jensson, on the adoption of VR for gaming, if I look at the price points of headsets and the PC, a little bit high for a wider adoption. Could the use of GPU in the cloud like you guys are introducing with GeForce now be a way for the price points on VR to come down? And then I have a follow-up for Colette. The 1st year of VR has sold several 100,000 units and many 100 of thousands of units. And our VRWorks SDK, which allows us to process graphics in very low latency, dealing with all of the computer vision processing and whether it's lens warping and such, it has been has delivered really excellent results. The early VR is really targeted at early adopters. And I think the focus of ensuring an excellent experience that surprises people, that delight people by Oculus and by Valve and by Epic and by Vive, by ourselves, by the industry has really been a good focus. And I think that we've delivered on the promise of a great experience. The thing that we have to do now is that we have to make the headsets easier to use with fewer cables. We have to make it lighter. We have to make it cheaper. And so those are all things that the industry is working on. And as the applications continue to come online, you're going to see that they're going to meet themselves and find success. I think the experience is very, very clear that VR is exciting. However, remember that we are also in the VR we also brought VR to computer aided design and to professional applications. In this particular area, the cost is just simply not an issue. And in fact, many of the applications previously were power walls or caves, VR caves that cost 100 of 1,000 of dollars. And now you can put that same experience, if not even better, on the desk of designers and creators. And so I think that you're going to find that creative use and professional use of VR is going to grow quite rapidly. And just recently, we announced a brand new Quadro 5000 P5000 with VR, the world's 1st VR notebook that went to market with HP and Dell, and they're doing terrifically. And so I think, I would think about VR in the context of both professional applications as well as consumer applications. But I think the 1st year was absolutely a great success. Your next question comes from the line of Ramesh Shah with Nomura. Yes. Thank you. And first of all, congratulations on a strong fiscal 2017. If I may, gentlemen, the revenue beat this quarter wasn't as big as we've seen in the last several periods and most of it came from data center. I totally understand that when your gaming business expands as much as it has, it becomes harder to beat expectations by the same margin. But I was wondering if you could just spend some time talking about gaming demand and how you think it was during the holiday season? Well, the global PC gaming market is still vibrant and growing. And the number of esports gamers around the world is growing. You guys know that Overwatch is a home run. Activision Blizzard's Overwatch is raging all over Asia and Esports fans all over the world are picking it up. And it's graphically very intensive. Without a 10, 50 class and above. It's simply a nonstarter and really enjoy it. You need at least a 1060. And so this last quarter, we launched a 1050 and a 1050 Ti all over the world and we're seeing terrific success out of that. And my expectation going into next year is that Overwatch is going to continue to spread all over the world. It's really basically just started. It started in the West and it's now moving into the East where, the largest esports markets are. And so Overwatch is going to be a huge success. League of Legends is going to continue to be a huge success. And my expectation is that eSports along with AAA titles that are coming out this year, is going to keep PC gaming continue to grow. And so I thought I quite frankly thought Q4 was pretty terrific. And we had a record quarter, we had a record year. And I don't remember the last time that a large business the size of ours and surely the size of a data center business grew by a factor of 3. And so, I think we're in a great position going into next year. Your next question comes from the line of Rajeev Gill with Needham and Company. Yes, thanks. Jensen, can you talk a little bit about the evolution of artificial intelligence and kind of make a distinction between artificial intelligence versus machine learning versus deep learning. There are different kind of categorizations and implementations of those different sub segments. So I wanted to get a sense from you how NVIDIA's end to end computing platform kind of dominates machine learning relative to say the competition. Then I have a question on the gross margins if I could. Yes. First of all, thanks. Thanks for the question. The way to think about that is deep learning is a breakthrough technique in the category of machine learning. And machine learning is an essential tool to enable AI, to achieve AI. If a computer can't learn and if you can't learn continuously and adapt with the environment, there's no way to ever achieve artificial intelligence. Learning, as you know, is a foundational part of intelligence. And deep learning is a breakthrough technique where the software can write software by itself by learning from a large quantity of data. Prior to deep learning, other techniques like expert systems and role based systems and hand engineered features where engineers would write algorithms to figure out how to detect a cat. And then they would figure out how to write another algorithm to detect a car. You could imagine how difficult that is and how imperfect that is. It basically kind of works, but it doesn't work good enough, well enough to be useful. And then deep learning came along. The reason why deep learning took a long time to come along is because its singular handicap is that it requires an enormous amount of data to train the network and it requires an enormous amount of computation. And that's why a lot of people credit the work that we've done with our programmable GPUs and our GPU computing platform and the early collaboration with deep learning AI researchers as the big bang, if you will, that catalyst that made modern AI possible. We made it possible to crunch through an enormous amount of data to train these very deep neural networks. Now the reason why deep learning has just swept the world, It started with convolution neural networks, but reinforcement networks and time sequence networks and all kinds of interesting adversarial networks. And the list of types of networks, I mean, there are 100 networks being created in a week. And papers are coming out of come out of everywhere. The reason why is because deep learning has proven to be quite robust. It is incredibly useful and this tool has, at the moment found no boundaries of problems that it's figured out how to solve. And I think that the traditional methods of machine learning are still going to be useful if the absolute precision of the prediction or classification is not necessarily super important. For example, if you wanted to understand the sentiment of consumers on a particular new product that you sent. Whether the sentiment is exactly right, so long as you understand the basic trend and you largely understand the sentiment, I think people would consider that information to be useful. However, if you're using machine learning for cancer detection, obviously, we need to have a level of precision that is quite high. And so whether it's in healthcare or financial services or high performance computing and in some areas where, for example, ad supported Internet search, small differences in accuracy could make a very large difference in financial results for the advertiser and for the people hosting the service. And so in all these cases, deep learning has found a great utility. And that's one of the reasons why we're seeing so much growth. And obviously, for self driving cars, being kind of right is not a good idea and we'd like to be exactly right. Your next question comes from the line of Matt Ramsay with Canaccord. Thank you very much. Jensen, I was you guys obviously have won some business, with your automotive supercomputer at Tesla in recent periods. And I was curious if you could comment on some of the application porting and moving of features from the previous architecture onto your architecture. And I guess how that's gone and what you guys have learned through that process and how it might be applied to some of your future partnerships? Thank you. Sure. First of all, you know that we are a full stack platform. The way we think about all of our platforms is from the application all the way back to the fundamental architecture in the semiconductor device. And so in the case of Drive PX, we created the architecture optimized for neural net, for sensor fusion, for high speed processing. The semiconductor design in the case of DRIVE PX2 called Parker, Tegra Parker, The system software for high speed sensor fusion and moving data all the way around the car, the better you do that, the lower cost the system will be. The neural networks on top of that, that sits on top of our deep learning SDK called cuDNN and TensorRT, basically frameworks of AI. And then on top of that, the actual algorithms for figuring out how to use that information from perception localization to action planning. And so and then on top of that, we have an API and an SDK that is integrated into Mapmakers and we integrate into every single map HD map service in the world from here to Tantom to Zenrin in Japan to Baidu in China. So this entire stack is a ton of software. But your question specifically has to do with the perception layer. And that perception layer, quite frankly, is just a small part of the self driving car problem. And the reason for that is because in the final analysis, you want to detect lanes. You've got video coming in, you want to detect lanes. You have video coming in, you want to detect the car in front of you. And all we have to do, it's not trivial, but it's also not monumental. We have to detect and sense the lanes in the cars and we train our networks to do so. And as you know very well now, the deep neural net has the ability to detect objects far better than any human engineered computer vision algorithms prior to deep learning. And that's one of the reasons why Tesla and others have jumped on top of the deep learning approach and abandoned traditional hand featured computer vision approaches. And so anyways, the answer to your question is that by working on self driving cars, end to end, we realized that this is much more than computer vision that the self driving car platform is a stack of software and algorithms that's quite quite complex. And now we've had a lot of experience doing so. And then recently at CES, we announced partnerships with Audi, which we announced that we will have Level 4 self driving cars on the road by 2020. We announced a partnership with Daimler. We announced a partnership with ZF and Bosch, 2 of the world's top Tier 1 suppliers. We also announced partnerships with all of the mapping companies. And so if you put all that stuff together, we have the processor, we have the Tier 1 partnerships for the integration of the systems, we have all the software on top of it, the deep learning networks, the car partnerships, of course, and integrated into maps around the world. And all that entire stack when you put them all together should allow us to have self driving cars on the road. Your next question comes from the line of Joe Moore with Morgan Stanley. Great. Thank you for taking the question. I wonder if you could talk a little bit about the inference market. Where are you in terms of hyperscale adoption for specialized inference type solutions? And how big do you think that market can ultimately be? Thank you. Yes. The inference market is going to be very large. And it's going to it's as you know very well, in the future, almost every computing device will have inferencing on it. A thermostat will have inferencing on it. A bicycle lock will have inferencing on it. Cameras will have inferencing on it. And self driving cars would have an imporrage amount of inferencing on it. Robots, vacuum cleaners, you name it, smart microphones, smart speakers, all the way into the data center. And so I believe that long term, there'll be a trillion devices that has inferencing connected to edge computing devices near them connected to cloud computing devices cloud computing servers. Okay. So that's basically architecture. And so the largest inferencing platform will likely be ARM devices. I think that that goes without saying. ARM will likely be running inferencing networks, 1 bit, XNOR, 8 bit and even some floating point. It just depends on what level of accuracy do you want to achieve, what level of perception do you want to achieve and how fast do you want to perceive it. And so the inferencing market is going to be quite large. We're going to focus in markets, where the inferencing precision, the inferencing, the perception scenario and the performance by which you have to do it is mission critical. And of course, self driving cars is a perfect example of that. Robots manufacturing robots will be another example of that. In the future, you're going to see in our GTC, if you have a chance to see that, we're working with AI City partners all over the world for end to end video analytics and that requires very high throughput, a lot of computation. And so the examples go on are several areas where inferencing is quite vital. I mentioned one earlier, just mapping the earth, mapping the earth at the street level, mapping the earth in HD and three-dimensional level for self driving cars. That process is going to require just a pilot GPUs running continuously as we continuously update the information that needs to be mapped. There's inferencing, which is called offline inferencing, where you have to retrain a network after you deployed it. And you would likely retrain and recategorize, reclassify the data using the same servers that you use for training. And so even the training servers will be used for inferencing. And then lastly, all of the nodes in cloud will be inferencing nodes in the future. I've said before that I believe that every single node in the cloud data center will have inferencing capability and accelerated inferencing capability in the future. And I continue to believe that and these are all opportunities for us. Your next question comes from the line of Charles Huang from Goldman Sachs. Hello, can you hear me? Sure. Hi, this is Toshiya from Goldman. Thanks for taking the question and congrats on the results. I had a question on gross margins. I think you're guiding Q1 gross margins only mildly below levels you saw in fiscal Q4 despite the royalty stream from Intel rolling over. And I'm guessing, improvement in mix and data center and parts of gaming are driving this. But A, is that kind of the right way to think about puts and takes going into Q1? And B, if that is indeed the case, should we expect gross margins to edge higher in future quarters and future years as data center becomes a bigger percentage of your business? Yes. This is Colette. Let me see if I can help answer that. So you're correct in terms of how to look at that in Q1. The delta from Q4 to Q1 is we only have a partial part of recognition from the Intel and that stops in the middle of March. So as we move forward as well going into Q2, we will also have the absence of what we had in Q1 moving to Q2. I'm not here to give guidance on Q2 because we just give give guidance out 1 quarter. But keep that in mind, there's a partial amount of Intel still left in Q1 and then it depletes in Q2. If you think about our overall model, our overall business model, it has moved to higher end value added platforms, and that's what we're selling. So our goal is absolutely to continue to concentrate on providing those higher value platforms that gives us the opportunity for gross margin as we make those investments in terms of an OpEx. We'll see what that kind of mix looks like as we go into Q2. But just to leave you with an understanding, Intel, is probably what we can do here, okay? Your next question comes from the line of Stephen Chen from UBS. Hi, thanks for taking my questions. First one is on the data center segment. Just given the expected sequential growth in that business during the April quarter, can you talk about what products are helping to drive that? Is it possibly the DGX-one supercomputer box or is it more GPUs for training purposes at the hyperscale cloud data center? It would have to be Tesla processors used in the cloud. There are several SKUs of Tesla processors. There's the Tesla processors used for high performance computing and it has, FP64, FP32, ECC, it's designed, and it has CUDA of course and has been optimized for molecular dynamics, astrophysics, quantum chemistry, fluid dynamics, the list goes on and on. The vast majority of the world's high performance supercomputing applications, imaging applications, 3 d reconstruction applications has been ported onto our GPUs over the course of the last decade and some. And that's a very large part of our Tesla business. Then of course, we introduced on top of the architecture our deep learning stack. Our deep learning stack starts with cuDNN, the numeric kernels, a lot of algorithms inside them to be optimized for numerical processing of all kinds of different precisions. It's integrated into frameworks of different kinds. There are so many different frameworks from TensorRT to Caffe to Torch to Theano to MXNet to CNTK. The work that we did with Microsoft, which is really excellent, scaling it up from 1 GPU to many GPUs across multiple racks. And that's our deep learning stack and that's also very important. And then the third is GRID. GRID is a completely different stack. It's the world's 1st graphics virtualization stack fully integrated into Citrix, integrated into VMware, every single workstation and PC application has been verified, tested and has the ability to be streamed from a data center. And then last year, starting I think we announced it in we started shipping it in August, our DGX-one, the world's 1st AI supercomputer appliance, which integrates a whole bunch more software of all different types and has the ability to we introduced our first NVIDIA docker. It containerizes applications. Makes it possible for you to have a whole bunch of users use 1 DGX. They could all be running different frameworks because most environments are heterogeneous. And so that's the GX-one. And it's got an exciting pipeline ahead of it. And it's really designed for companies and work groups who don't want to build their own supercomputer like the hyperscalers and aren't quite ready to move into the cloud because they have too much data to move to the cloud. And so everybody basically can easily buy a DGX-one and it's fully integrated, fully supported and get to work on deep learning right away. And so each one of these are all part of our data center business. And but the largest because it's been around the longest is our Tesla business, but they're all growing every single one of them. Your next question comes from the line of Steve Smiggy with Raymond James. Great. Thanks a lot for the time. Just a quick question in the auto market. At CES, you had some solutions you were demonstrating. It showed a pretty significant decline in terms of the size of what was being offered. You really shrunk it down a lot, yet still having great performance. If you think out to sort of the Level 4 solution that you talked about for 2020, how small can you ultimately make that? It seems like you could be sort of relative to the size of the car pretty small. So just curious if you could comment on that and what impact having the system in the car makes on it? We currently have DrivePX today is a 1 chip solution for Level 3. And it can have and with 2 chips, 2 processors, you could achieve Level 4. And with many processors, you could achieve Level 5 today. And some people are using many processors to develop their Level 5 and some people are using a couple of processors to develop their Level 4. Our next generation, so that's all based on the Pascal generation. It's all based on the Pascal generation. Our next generation, the processor is called Xavier. We announced that recently. Xavier basically takes 4 processors and shrink it into 1. And so we'll be able to achieve level 4 with 1 processor. That's the easiest way to think about it. So we'll achieve level 3 with 1 processor today. Next year, we'll achieve level 4 with 1 processor. And with several processors, you could achieve level 5. But I think that the number of processors is really interesting because we need to do the processing of sensor fusion and we got to do perception, we have to do localization, we have to do driving. There's a lot of functional safety aspects to it, failover functionality. There are all kinds of black box recorders, all kinds of different functionality that goes into the processor. And I think it's really quite interesting. But in the final analysis, what's really, really hard and this is one of the reasons why our positioning in the autonomous driving market is becoming more and more clear is that in the final analysis, it's really a software problem. And it's an end to end software problem. It goes all the way from processing in the perception processing in the car to AI processing, to helping you drive connected to HD clouds for HD map processing all over the world. And so this end to end stack of software is really quite a large undertaking. I just don't know where anybody else is currently doing that with the exception of 1 or 2 companies. And so I think that that's really where the great complexity is. We have the ability to see and optimize across the entire range. Now the other thing that we announced at CES is worth mentioning is that we believe in the future, level 4 means that you will have autopilot capability, hands free autopilot capability in many scenarios. However, it's unlikely to ensure and to guarantee that in every scenario that you can achieve level 4. It's just not practical for some time. However, during those circumstances, we believe that the car should still have an AI, that the car should be monitoring what's happening outside and it should be monitoring the driver. And when it's not driving for you, it's looking out for you. And we call that the AI co pilot, whereas AI autopilot achieves level 4 driving, AI copilot looks out for you in the event that it doesn't have the confidence to drive on your behalf. And so I believe that that's a really big breakthrough and we're just seeing incredible excitement about it around the industry because I think it just makes a lot of sense. And the combination of the two systems allows us to achieve build better car. Your next question comes from the line of Craig Ellis with B. Riley and Company. Thanks for sneaking me in and congratulations on the very good execution. Jensen, I wanted to come back to the gaming platform. You've now got the business running at a $5,000,000,000 annualized run rate. So congratulations on the growth there. I think investors look at that as a business that's been built on the strength of a vibrant enthusiast market. But at CES, you announced the GeForce NOW offering, which really allows you to tap into the more casual potential gamer. So the question is, is what will GeForce NOW do incrementally for the opportunity that you have with your gaming platform? Yes, I appreciate that. I think first of all, the gaming the PC gaming market is growing because of a dynamic that nobody ever expected, a dynamic that nobody ever expected 20 years ago. And that's basically how video games went from being a game to becoming a sport. And not only is it a sport, it's a social sport. And in order to play some of these modern eSports games, it's a 5 on 5. And so you kind of need 4 other friends. And so as a result, in order to enjoy, to be part of this phenomenon that's sweeping the world, that it's rather sticky. And that's one of the reasons why Activision Blizzard is doing so well. That's one of the reasons why Tencent is doing so well. These two companies have benefited from tremendously from the e sport dynamic and we're seeing it all over the world. And although it's free to play for some people, of course, you need to have a reasonably good computer to run it. And that's one of the reasons why you need GeForce in your PC so that you can enjoy these sports. When it's also a sport, nobody likes to lose. And surely nobody likes to blame their equipment when they do lose. And so having GeForce allows gives you confidence and gives you an edge And for a lot of gamers, it's just a world standard. And so I think that number 1, e sport is one of the reasons why gaming continues to grow. And I think at this point, it's fair to say that even though it's now the 2nd most watched spectator sport on the planet behind Super Bowl, it is also the 2nd highest paid winning sport behind football. It will soon be the largest sport in the world. And I can't imagine too many young people long term not coming into the sport somehow and as the sport continues to expand in genres. And so that's one of the core reasons. Now aside for you asked a question about GeForce NOW, which I really appreciate. The simple way to think about that is this. There are many computers in the world that simply don't have the ability to enjoy video games, whether it's extremely thin and light notebooks, in our Apple Macs, Chromebooks, the integrated graphics that don't have very good capabilities, I think that it's the reasonable thing to do is to put the technology in the cloud. And it took us some 5 years to make this possible to put the technology in the cloud and stream the video game experience with very low latency to the computer like Netflix does. And so we're basically turning the PC into a virtualized gaming experience and putting that in the cloud. And so I don't know exactly how big it's going to be yet, but our aspiration is that we would reach the parts of the markets where they're casual or they just want to have another way, another device where they can gain or somebody would like to come into the gaming world and isn't quite ready to invest the time in building a computer or buying into a GeForce PC yet. So, I'm anxious to learn from it. And when I learn more about GeForce NOW, I'll be more than happy to share it. Unfortunately, that is all the time we have for questions. Do you have any closing remarks? I want to thank all of you guys for following us. We had a record year, a record quarter. And most importantly, we're at the beginning of the AI computing revolution. This is a new form of computing, new way of computing where parallel data processing is vital to success and GPU computing that we've been nurturing for the last decade and some is really the perfect computing approach. We're seeing tremendous growth and exciting growth in the data center market. Data center now represents had grew 3x year over year and it's on its way to become a very significant business for us. Gaming is a significant business for us and longer term self driving cars is going to be a really exciting growth opportunity. The thing that has really changed our company, what really defines how our company goes to market today is really the platform approach that instead of just building a chip that is industry standard, we created software stacks on top of it to serve vertical markets that we believe will be exciting long term that we can serve. And we find ourselves incredibly well positioned now in gaming, in AI and in self driving cars. I want to thank all of you guys for following NVIDIA and have a great year. This concludes today's conference call. You may now disconnect.