Good afternoon. My name is Claudine. I'll be your conference coordinator today. I'd like to welcome everyone to the NVIDIA Financial Results Conference Call. All lines have been placed on mute. After the speaker's remarks, there will be a question-and-answer period. Participants can register for a question by pressing the one followed by the four on your telephone. This conference is being recorded Thursday, May 12th, 2016. I would now like to turn the call over to Arnab Chanda, Vice President of Investor Relations at NVIDIA. Please go ahead, sir.
Thank you. Good afternoon, everyone. Welcome to NVIDIA's conference call for the first quarter of 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 today's call is being webcast live on NVIDIA's investor relations website. It is also being recorded. You can hear a replay by telephone until the 19th of May, 2016. The webcast will be available for replay up until next quarter's conference call to discuss Q2 financial results. The content of today's call is NVIDIA's property. It cannot be reproduced or transcribed without our prior written consent. During the course of this call, we may make forward-looking statements based on current expectations.
These forward-looking statements are subject to a number of significant risks and uncertainties. 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, the 12th of May, 2016, 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. In March, we introduced our newest GPU architecture, Pascal. This extraordinary scalable design built on the 16-nanometer FinFET process provides massive performance and exceptional power efficiency. It will enable us to extend our leadership across our four specialized platforms, gaming, professional visualization, data center, and automotive. Year-on-year revenue growth continued to accelerate, increasing 13% to $1.3 billion. Our GPU business grew 15% to $1.08 billion from a year ago. Tegra processor business was up 10% to $160 million. Growth continued to be broad-based across all four platforms. Record performance in data center was driven by the adoption of deep learning across multiple industries. In Q1, our four platforms contributed nearly 87% of revenue, up from 81% a year earlier. They collectively increased 21% year-over-year. Let's start out with our gaming platform. Gaming revenue increased 17% year-on-year to $687 million.
Momentum carried forward from the holiday season, helped by the continued strength of Maxwell-based GTX processors. Last weekend at DreamHack Austin, we unveiled GeForce GTX 1080 and GTX 1070, our first Pascal GPUs for gamers. They represent a quantum leap for gaming and immersive VR experiences, delivering the biggest performance gains from the previous generation architect in a decade. Media reports and gamers have been unanimously enthusiastic. The Verge wrote, "What NVIDIA is doing with its new GTX 1000 series is bringing yesteryear's insane high-end into 2016's mainstream." We also extended our VR platform by adding spatial acoustics to our VRWorks software development kit, which helps provide an even greater sense of presence within VR. We introduced simultaneous multi-projection, enabling accurate, efficient projection of the real world to surround monitors, VR headsets, as well as future displays.
To showcase these technologies, we created our own amazing open source game called NVIDIA VR Funhouse, available on Steam. In addition, we announced Ansel, an in-game photography system which enables gamers to capture high resolution and VR scenes within their favorite games. Moving to professional visualization. Quadro grew year-on-year for the second consecutive quarter. Revenue rose 4% to $189 million. Growth came from higher-end products and mobile workstations. We launched the M6000 24GB and are seeing good success among multiple customers, including Toyota and Pixar. Roche is using the M6000 to speed its DNA sequencing pipeline by 8X, enabling more affordable genetic testing. We see exciting opportunities for our Quadro platform with virtual reality and NVIDIA Iray, a photorealistic rendering tool that enables designers effectively to walk around their creations and make real-time adjustments.
Moving to data center, revenue was a record $143 million, up 63% year-on-year and up 47% sequentially, reflecting enormous growth in deep learning. In just a few years, deep learning has moved from academia and is now being adopted across the hyperscale landscape. We expect growing deployment in the coming year among large enterprises. GPUs have become the accelerator of choice for hyperscale data centers due to their superior programmability, computational performance, and power efficiency. Our Tesla M4 is over 50% more power efficient than other programmable accelerators for applications such as real-time image classification for AlexNet, a deep learning framework. Hyperscale companies are the fastest adopters of deep learning, accelerating their growth in our Tesla business. Starting from infancy three years ago, hyperscale revenue is now similar to that from high-performance computing. NVIDIA GPUs today accelerate every major deep learning framework in the world.
We power IBM Watson and Facebook's Big Sur server for AI, and we are in AI platforms at hyperscale giants such as Microsoft, Amazon, Alibaba, and Baidu for both training and real-time inference. Twitter has recently said they use NVIDIA GPUs to help users discover the right content among the millions of images and videos shared every day. During the quarter, we hosted our seventh annual GPU Technology Conference. The event drew record attendance, with 5,500 scientists, engineers, designers, and others across a wide range of fields, and featured 600 sessions and 200 exhibitors. At GTC, we unveiled the Tesla P100, the world's most advanced GPU accelerator based on the Pascal architecture. The P100 utilizes a combination of technologies including NVLink, a high-speed interconnect allowing application performance to scale on multiple GPUs, high memory bandwidth, and multiple hardware features designed to natively accelerate AI applications.
The Next Platform, an enterprise IT site, called it a beast in all of the good sense of that word. Among the first customers for our Pascal accelerator is the Swiss National Supercomputing Centre, which will use it to double the speed of Europe's fastest supercomputer. At GTC, we also announced the DGX-1, the world's first deep learning supercomputer. Loaded with 8 P100s in a single box interconnected with NVLink, it provides the deep learning performance equivalent to 250 traditional servers. DGX-1 comes loaded with a suite of software designed to aid AI and application developers. Universities, hyperscale vendors, and large enterprises developing AI-based applications are showing strong interest in the system. Among the first to get DGX-1 will be the Massachusetts General Hospital.
It launched an initiative that applies AI techniques to improve the detection, diagnosis, treatment, and management of diseases, drawing on its database of some 10 billion medical images. In our GRID graphics virtualization business, we are seeing interest across a variety of industries, ranging from manufacturing, energy, education, government, and financial services. Finally, in automotive, revenue continued to grow, reaching $113 million, up 47% year-over-year and up 22% sequentially, reflecting the growing popularity of premium infotainment features in mainstream cars. NVIDIA is working closely with partners to develop self-driving cars using our end-to-end platform, which starts with Tesla in the data center and extends through the deployment with DRIVE PX 2. Since we unveiled DRIVE PX 2 earlier this year, worldwide interest has continued to grow among car makers, tier-one suppliers, and others.
We are now collaborating with more than 80 companies using the open architecture of DRIVE PX to develop their own software and driving experiences. At GTC, we demonstrated the world's first self-driving car trained using deep learning and showed its ability to navigate on roads without lane markings, even in bad weather. Additionally, we announced that DRIVE PX 2 will serve as the brain behind the new Roborace initiative in the Formula E racing circuit. The circuit will include 10 teams racing identical cars, all using DRIVE PX 2. Beyond our four platforms, our OEM IP business was $173 million, down 21% year-on-year, reflecting weak PC demand. Turning to the rest of the income statement. We had record GAAP and non-GAAP gross margins for the first quarter, at 57.5% and 58.6%, respectively.
Driving these margins was the strength of our Maxwell GPUs, the success of our platform approach, and strong demand for deep learning. GAAP operating expenses for the first quarter were $506 million and declined from $539 million in Q4 on lower restructuring charges. Non-GAAP operating expenses were $443 million, flat sequentially and up 4% from a year earlier, reflecting increased hiring for our growth initiatives and development-related expenses associated with Pascal. GAAP operating income for the first quarter was $245 million, up 39% from a year earlier. Non-GAAP operating income was $322 million, also up 39%. Non-GAAP operating margins improved more than 470 basis points from a year ago to 24.7%. For the first quarter, GAAP net income was $196 million. Non-GAAP net income was $263 million, up 41%, fueled by the strong revenue growth and improved gross and operating margins.
During the first quarter, we'd entered into a $500 million accelerated share repurchase agreement and paid $62 million in quarterly cash dividends. Since the restart of our capital return program in the fourth quarter of fiscal 2013, we have returned over $3.5 billion to shareholders. This represents over 100% of our cumulative free cash flow for fiscal years 2013 through this Q1. For fiscal 2017, we intend to return approximately $1 billion to shareholders through share repurchases and quarterly cash dividends. Turning to the outlook for the second quarter of fiscal 2017. We expect revenue to be $1.35 billion, plus or minus 2%. Our GAAP and non-GAAP gross margins are expected to be 57.7% and 58.0%, respectively, plus or minus 50 basis points. GAAP operating expenses are expected to be approximately $500 million. Non-GAAP operating expenses are expected to be approximately $445 million.
GAAP and non-GAAP tax rates for the second quarter of fiscal 2017 are both expected to be 20%, plus or minus 1%. Further financial details are included in the CFO commentary and other information available on our IR website. We will now open the call for questions. Operator, could you please poll for questions? Thank you.
Thank you. Ladies and gentlemen, if you'd like to register a question, please press the one followed by the four on your telephone. You'll hear a three-tone prompt to acknowledge your request. Should you wish to withdraw your registration, please press the one followed by three. If you're using a speakerphone, please lift your handset before entering your request. Our first question comes from the line of Vivek Arya with Bank of America. Please go ahead.
Thank you for taking my question, good job on the results and the guidance. As my first one, Jensen, how do you assess the competitive landscape in PC gaming? AMD recently claimed to be taking a lot of share, they're launching Polaris soon. Just if you could walk us through what does NVIDIA do better than AMD, so that helps you maintain your competitive edge in this market, what impact will Pascal have on that?
Vivek, thank you. Our PC gaming platform, GeForce, is strong and getting stronger than ever. I think the reason for that is several folds. First of all, our GPU architecture is just superior, and we dedicate an enormous amount of effort to advancing our GPU architecture. I think the engineering of NVIDIA is exquisite, and our craftsmanship is really unrivaled anywhere. The scale of our company in building GPUs is the highest and the largest of any company in the world. This is what we do. This is the one job that we do. It's not surprising to me that NVIDIA's GPU technology is further ahead than any time in its history. The second thing, however, it's just so much more than just chips anymore, as you know. Over the last 10 years, we've started to evolve our company into much more of a platform company.
It's about developing all the algorithms that sit on top of our GPUs. A GPU is a general purpose processor. It's a general purpose processor that's dedicated to a particular field of computing, such as computer graphics here, physics simulation, et cetera. The thing that's really important is all of the algorithms that sit on top of it, and we have a really, really fantastic team of computational mathematicians that captures our algorithms and our know-how into GameWorks, into the physics engine, and recently, the really amazing work that we're doing in VR that we've embodied into VRWorks. Lastly, it's about making sure that the experience always just works. We have a huge investment in working with game developers all over the world from the moment that the game is being conceived of all the way to the point that it's launched.
We optimize the games on our platform. We make sure that our drivers work perfectly. Even before a gamer downloads or buys a particular game, we've already updated their software so that it works perfectly when they install the game. We call that GFE, GeForce Experience.
Vivek, it's really about a top to bottom approach, and I haven't even started talking about all of the marketing work that we do in engaging the developers and engaging the gamers all over the world. This is really a network platform, and all of our platform partners that take our platform to market. It's a pretty extensive network and it's a pretty extensive platform, and it's so much more than chips anymore.
Got it. Thank you, Jensen. As my follow-up, it seems like data center products were the big upside surprise in Q1, grew over 60% from last year. Could you give us some more color on what drove that upside? Was it the initial Pascal launch? Is that impact still to come? Just broadly, what trends are you seeing there in HPC versus cloud versus some of these new AI projects that you are involved with?
Yeah, thanks. You know that I've been rather enthusiastic about high-performance computing for some time. We've been evolving our GPU platform so that it's better at general purpose computing than ever. Almost every single data center in the world and every single server company in the world are working with us to build servers that are based on GPUs, based on NVIDIA GPUs for high-performance computing. One of the most important areas of high-performance computing has been this area called deep learning. This deep learning, as you know, as you probably are starting to hear, is a brand-new computing model that takes advantage of the massively parallel processing capability of a GPU, along with the big data that many companies have to essentially have software write algorithms by itself.
Deep learning is a very important field of machine learning, and machine learning is now in the process of revolutionizing artificial intelligence, making machines more and more intelligent and using it to discover insight that, quite frankly, is impossible otherwise. This particular field was first adopted by hyperscale companies so that they could find insight and make recommendations and make predictions from the billions of customer transactions they have every day. Now it's in the process of moving into enterprises. In the meantime, hyperscale companies are now in the process of deploying our GPUs in deep learning applications into production. We've been talking about this area for some time, now we're starting to see the broad deployment in production. We're quite excited about that.
Our next question comes from the line of Mark Lipacis with Jefferies. Please go ahead.
Thanks for taking my questions. First question, the growth in the NVIDIA Tesla business is impressive. In looking back, it seemed like that business actually decelerated in 2015, which was a head-scratcher for me. I wonder, do you think that your customers in that business paused in anticipation of Pascal, or do you think it's the AI apps and deep learning applications that are just hitting their stride right now?
Well, decelerating, I guess I'm not sure I recall that. The thing about HPC, about GPU computing is, as you know, this is a new computing model, we've been promoting this computing model now for close to seven years. A new computing model doesn't come along very frequently. In fact, as I know it, I don't know if there's a new computing model that's used anywhere that has been revolutionary in the last 20 years. So GPU computing took some time to develop. We've been evangelizing it for quite some time. We developed robust tools so that make it easier for people to take advantage of our GPUs. We have industry expertise in a large number of industries now. We have APIs that we've created for each one of the industries.
We've been working with the ecosystem in each one of the industries and developers in each one of the industries, as of this time, we have quite a large number of industries that we accelerate applications for. So I guess my recollection would be that it has taken a long time, in fact, to have made GPU computing into a major new computing model. I think at this point, it is pretty clear that it's going mainstream. It is really one of the best ways to achieve the post-Moore's Law era of computing acceleration, it's being adopted by all kinds of applications. The one that, of course, that is a very big deal is deep learning and machine learning. This particular field is a brand-new way of doing computing for a large number of companies, we're seeing traction all over the place.
Our next question comes from the line of Stephen Chin with UBS. Please proceed.
Hi, thanks for taking my questions. Jensen or Colette, first of all, I wanted to see if you could help provide some color on some of the drivers of growth for fiscal 2Q, whether most of it's coming from Pascal, possibly in the gaming market or in the Tesla products, or if there's also some amount of growth in Tegra automotive as well for fiscal 2Q.
Yeah, Stephen, I would expect that all of our businesses grow in Q2. It's across the board. We're seeing great traction in gaming. Gaming, as you know, has multiple growth drivers. Partly the gaming is growing because the production value of games is growing, partly because the number of people who are playing is growing. Esports is more popular than ever. Sports spectatorship is more popular than ever. Gaming is just a larger and larger market, and it's surprising everybody. The quality of games is going up, which means that seems to go up.
High-performance computing is growing. The killer app is machine learning and deep learning. That's going to continue to go into production from the hyperscale companies as we expand our reach into enterprises all over the world now, companies who have a great deal of data that they would like to find insight in. Automotive is growing. We're delighted to see that the enterprise is growing as well.
Great. As I follow up, maybe for Colette, on the gross margin side of things, you guys are guiding margins up nicely for the quarter, and just kind of wondering, looking out further across the year, whether or not the levers that you have available to you currently, if there's further room for expansion, whether it's from product mix, higher ASPs, and/or maybe even some of the platform-related elements such as software services. I was just kind of wondering, especially on the software side, how much that can continually help margins from a platform perspective.
Sure. Thanks, Stephen. Yes, our gross margins within the quarter for Q1 did hit record levels, just due to very strong mix across our products on the Maxwell side, both from a gaming perspective as well as what we have in enterprise for pro-visualization and data center. As we look to Q2, a good review of where we also see gross margins, and those are looking at a non-GAAP at about 58%. Mix will again be a strong component of that as our launch of Pascal will come out with high-end gaming and with data center, and the growth essentially across all of our platforms will help our overall gross margins. As we go forward, there's still continued work to do.
We're here to guide just one quarter out, but we do have a large TAM in front of us on many of these different markets, and the mix will certainly help us. We're in the initial stages of rolling out what we have in software services in our overall systems, so I don't expect it to be a material part of the overall gross margin. It will definitely be a great value proposition for us for what we put forth.
Our next question comes from the line of Deepan Nag with Macquarie. Please proceed.
Yeah, thanks, guys, and congratulations on the great quarter. For Q2, can you kind of talk about how much of a contribution you expect from Pascal, and also maybe give us an update on where you think yields are progressing right now?
Yeah, thanks a lot, Deepan. Well, we're expecting a lot of Pascal. Pascal was just announced with 1080 and 1070, and both of those products are in full production. We're in production with Tesla P100, and so all of our Pascal products that we've already announced are in full production, so we're expecting a lot. Yields are good, and building these semiconductor devices are always hard, but we're very good at it. This is now a year behind when the first 16-nanometer FinFET products went into production at TSMC. They have yields under great control. TSMC is the world's best manufacturer of semiconductors, and we work very closely with them to make sure that we're ready for production. We surely wouldn't have announced it if we didn't have manufacturing under control. We're in great shape.
Our next question comes from the line of Ambrish Srivastava with BMO Capital Markets. Please proceed.
Hi, this is Gabriel Ho calling in for Ambrish. Thanks for taking my question. When you recently launched a new GTX GPU product, looks like your pricing, your MSRP appears to be higher than your prior generation. How should we think about your ASP and even gross margin trend as you're ramping this product for the rest of the year?
Yeah, thanks. The thing that's most important is that the value is greater than ever. One of the things that we know is games are becoming richer than ever. The production value's become richer than ever, and gamers want to play these games with all of the settings maxed out. They would like to play at a very high resolution, and they want to play it at very high frame rates. When I announced 1080, I was showing all of the latest, the most demanding games running at twice the resolution of a game console, at twice the frame rate of a game console, and it was barely even breathing hard.
I think one of the most important things is for customers of this segment, they want to buy a product that they can count on and that they can rely on to be ready for future generation games. Some of the most important future generation games are going to be in VR. The resolution's going to be even higher. The frame rate expectation is 90 hertz, and the latency has to be incredibly low so that you feel a sense of presence. I think the net of it all is that the value proposition we delivered with 1080 and 1070 is just through the roof. If you look at the early response on the web and from analysts, they're quite excited about the value proposition that we brought.
Our next question comes from CJ Muse with Evercore. Please go ahead.
Good afternoon. Thank you for taking my question. I guess two questions around the data center. I guess first part, how's the visibility here today, and I guess how do you see perhaps the transition hyperscale to ramp in HPC? I know you guys don't like to forecast over the next couple of quarters, but looking out over the next 12 to 24 months, this part of your business has grown from 8% to 11% year-over-year, and curious, as you look at one to two years, what do you think this could be as a percentage of your overall company? Thank you.
CJ, thanks a lot. I think the answer to a lot of your questions is I don't know. However, there are some things I do know very well. One of the things that we do know is that high-performance computing is an essential approach for one of the most important computing models that we know today, which is machine learning and deep learning. Hyperscale data centers all over the world is relying on this new model of computing so that it could harvest, it could study all of the vast amounts of data that they're getting to find insight for individual customers to make the perfect recommendation, predict what somebody would anticipate, would look forward to in terms of news or products or whatever it is.
This approach of using computing is really unprecedented, and this is a new computing model, and GPU is really ideal for it. We've been working on this for coming up on a decade, and it explains one of the reasons why we have such a great lead in this particular aspect. The GPU is really the ideal processor for these massively parallel problems, and we've optimized our entire stack of platforms, from the architecture to the design, to the system, to the middleware, to the system software, all the way to the work that we do with developers all over the world so that we can optimize the entire experience to deliver the best performance. This is something that's taken a long time to do. I have a great deal of confidence that machine learning is not a fad.
I have a great deal of confidence that machine learning is going to be the future computing model for a lot of very large and complicated problems. I think that all of the stories that you see, whether it's the groundbreaking work that's done at Google and Google DeepMind on AlphaGo, to self-driving cars, to the work that people are talking about in artificial intelligence recommendation chatbots to Boy, the list just goes on and on. I think that it goes without saying that this new computing model in the last couple of years has really started to deliver very promising results, and I would characterize the results as being superhuman results. Now they're going into production, and we're seeing production deployments not just in one or two customers, but basically in every single hyperscale data center in the world, in every single country.
I think this is a very big deal, and I don't think it's a short-term phenomenon. The amount of data that we process is just going to grow, and so those are some of the things I do know.
Our next question comes from the line of Mark Lipacis with Jefferies. Please proceed.
Hi. Thanks for circling back and for a follow-up. Sometimes when you introduce a new product, and this is broadly for technology, there's kind of a hiccup as the transition happens where the supply chain blows out the older inventory before the new products can ramp in. Some people call that an air pocket. I was wondering, is that something that you can manage? How do you try to manage that? Do you account for it when you think about the outlook for this quarter? Thank you.
Yeah, thanks, Mark. Well, product transitions are always tricky, and we take it very seriously. There are several things that we do know. We have a great deal of visibility to the channel, we know how much inventory is where and of which kind. Secondarily, we have perfect visibility into our supply chain. Both of those matters need to be taken into account when we launch a new product. Anything could happen. The fact of the matter is we are in a high-tech business, and high tech is hard. The work that we do is hard. The team doesn't take it for granted, and we're not complacent about our work. I think that I can't imagine a better team in the world to manage this transition. We manage transitions all the time, we don't take it lightly.
However, you're absolutely right. It requires care, and the only thing I can tell you is that we're very careful.
Our next question comes from the line of Joseph Moore with Morgan Stanley. Please proceed.
Great. Thank you. I guess along the same lines, can you talk a little bit about the Founders Edition of the new gaming products? How is that different from sort of previous reference designs that you've done, and is there any kind of difference in economics to NVIDIA if you sell Founders Edition?
The Founders Edition is something we did as a result of demand from the end user base. The Founders Edition is basically a wholly designed by NVIDIA product. A reference design is really not designed to be an end product. It's really designed to be a reference for manufacturers to use as a starting point. A Founders Edition is designed so that it could be manufactured, it can be marketed, and customers can continue to buy it from us for as long as they desire. Our strategy is to support our global network of add-in card partners, and we're going to continue to do that. We gave everybody reference designs like we did before.
In this particular case, we created the Founders Edition so that people who would like to buy directly from us, people who like our industrial design, and people who would like the exquisite design and quality that comes with our products that we can do. It's designed to be extremely overclockable. It's designed with all the best possible components and if somebody would like to buy products directly from us, they have the ability to do that. I expect that the vast majority of the add-in cards will continue to be manufactured by our add-in card partners. That's our expectation and that's our hope. I don't expect any dramatic change in the amount of shifting of that. That's basically it. Founders Edition, the most exquisitely engineered add-in card the world's ever seen, directly from NVIDIA.
Our next question comes from the line of Harlan Sur with JP Morgan. Please go ahead.
Good afternoon. Solid job on the execution. At the recent Analyst Day, I think the team articulated its exposure to developed and emerging markets and the unit and ASP growth opportunities around EM. I am just wondering, what are the current demand dynamics that you're seeing in the emerging markets? Clearly, I think macro-wise, they're still pretty weak, but on the flip side, gaming has shown to be fairly macro insensitive. It would be great to get your views here.
Well, I think you've just said it. Depending on which one of our businesses that you're talking about, gaming is rather macro insensitive for some reason. People enjoy gaming, whether the economy is good or not, whether the oil price is high or not. People seem to enjoy gaming. Don't forget, gaming is not something that people do once a month, like going out to a movie theater or something like that. People game every day. The gamers that use our products are gaming every day. It's their way of engaging with their friends. They hang out with their friends that way. It's a platform for chatting. Don't forget that the number one messaging company in China is actually a gaming company. The reason for that is because while people are gaming, they're hanging out with their friends, and they're chatting with their friends.
It's really a medium for all kinds of things, whether it's entertaining or hanging out or expressing your artistic capabilities or whatnot. Gaming, for one, appears to be doing quite well in all aspects of the market. The second thing is enterprise, however, is largely, or hyperscale, is largely U.S. dynamic. The reason for that is because U.S. dynamic as well as the China dynamic, because that's where most of the world's hyperscale companies happen to be. Automotive, most of our automotive success to date has been from the European car companies, and we're seeing robust demand from the premium segments of the marketplace. However, in the future, we're going to see a lot more success with automotive here in the United States, here in Silicon Valley, in China.
We're going to see a lot more global penetration because of our self-driving car platform.
Our next question comes from the line of Ian Ing with the MKM Partners. Please go ahead.
Yes, thank you. For July, looks like you've got some operating expense discipline, given some hiring activity in April, you're down sequentially. Is that related to the timing of some tape-out activity? As Pascal rolls out, what could the shape of tape-out speed do you think for the upcoming quarters?
Well, all of the Pascal chips have been taped out. We still have a lot of engineering work to do. The differences are minor. We're a large company, and we have a lot of things that we're doing. I wouldn't over study the small deltas in OPEX. We don't manage things a dollar at a time, and we're trying to invest in the important things. On the other hand, this company is really, really good about not wasting money. We want to make sure that on the one hand, we invest into opportunities that are very important to our company, but we just have a culture of frugality that permeates our company. Lastly, from an operational perspective, we've unified everything in our company behind one architecture.
Whether you're talking about the cloud or workstations or data centers or PCs or cars or embedded systems or autonomous machines, you name it, everything is exactly one architecture. The benefit of one architecture is that we can leverage one common stack of software and that base software it really streamlines our execution. It's an incredibly efficient approach for leveraging our one architecture into multiple markets. Those three aspects of how we run the company really helps.
Our next question comes from the line, Blayne Curtis with Barclays. Please go ahead.
Hey, guys. Thanks for taking my question and nice results. Just curious, two questions. Jensen, you talked about the ramp of deep learning, and you talked about that you're going to use GPUs for both learning as well as applying inferences. Just curious, you mentioned all these customers, which stages are all these customers? Are they actually deploying it in volume, or are they still more your sales for learning? You said all segments up, just curious, OEMs finally hitting some easy compares. Is that also going to be up year-over-year?
I think this question actually
The OEM business, will that be up year-over-year?
I think OEM business is down year-over-year, isn't it?
Right. On Q2, we'll probably follow along in Q2, along with overall PC demand, which is not expected to grow. We'll look at that as our side product and probably would not be a growth business in Q2.
Yeah. Blayne, you know that our OEM business is a declining part of our company's overall business. Not to mention that the margins are also significantly below the corporate average. That would suggest that it's just increasingly a less important part of the way that we go to market. Now, what I don't mean by that is that we don't partner with the world's large OEMs, HP, Dell, IBM, Cisco, Lenovo, all of the world's large enterprise companies are our partners. We partner with them to take our platforms, our differentiated platforms, our specialty platforms to the world's markets. Most of them are related to enterprise. We just do less and less high volume components devices. Generic devices like cell phones that we got out of, generic PCs that we've gotten out of. Largely, we tend not to do business like that anymore.
We tend to focus on our differentiated platforms. Now you mentioned learning and training and inferencing. First of all, training is production. You can't train a network just once. You have to train your network all the time. Every single hyperscale company in the world is in the process of scaling out their training because the networks are getting bigger. They want their networks to do even better. The difference between a 95% accurate network and a 98% accurate network, or a 99% accurate network, could mean billions of dollars of differences to internet companies. This is a very big deal, and so they want their networks to be larger. They want to deploy their networks across more applications, and they want to train their network with new data all the time. Training is a production matter.
It is probably the largest HPC, high-performance computing application on the planet that we know of at the moment. We're scaling, we're ramping up training for production for hyperscale companies. On the other hand, I really appreciate you asking about the inferencing thing. We recently, well, this year, several months ago, we announced the M4, the Tesla M4 that was designed for inferencing. It's a little tiny graphics card, a little tiny processor, and it's less than 50 watts. It's called the M4. At GTC, I announced a brand-new compiler called the GPU Inference Engine, GIE. GIE recompiles the network that was trained so that it can be optimally inferenced at the lowest possible energy.
Not only are we already 50 watts, which is low power, we can also now inference at a higher energy efficiency than any processor that we know of today, better than any CPU by a very long shot, better than any FPGA. Now hyperscale companies could use our GPUs for both training, and they use exactly the same architecture for inferencing, and the energy efficiency is really fantastic. Now, the benefit of using GPU for inferencing is that you're not just trying to inference only, you're trying to oftentimes decode the image. You could be decoding the video, you inference on it, and you might even want to use it for transcoding, which is to re-encode that video, and stream it to whoever it is that wants to share a live video with.
The processing that you want to do on the images and the video and the data is more than just inferencing. The benefit of our GPU is that it's really great for all of the other stuff too. We're seeing a lot of success in M4. I expect M4 to be quite a successful product in hyperscale data centers. My expectation will start to ramp that into production Q2, Q3, Q4 timeframe.
Our next question comes from the line of Ross Seymore with Deutsche Bank. Please go ahead.
Hi, thanks for letting me ask a question. On the automotive side, I just wondered, Colette, in your CFO commentary, you mentioned product development contracts as part of the reason it was increasing. Can you give us a little bit of indication what those are and is the percentage of the revenue coming from those increasing? Maybe finally, is that activity indicative of future growth in any way that can be meaningful for us to track?
Sure. Thanks for the question. In our automotive business, there's definitely a process even before we're shipping platforms into the overall cars that we're working jointly with the auto manufacturers, startups, and others on what may be a future product. Many of those agreements continue and will likely continue going forward. That's what you see incorporated in our automotive business. Yes, you'll probably see this continue and go forward. It's not necessarily consistent. It starts in some quarters, bigger in other quarters, that's what's incorporated in our automotive.
Colette, let me just add one thing. The thing to remember is that we're not selling chips into a car. You know that DRIVE PX is the world's first autonomous driving car computer that's powered by AI. It's powered by deep learning. We're seeing a lot of success with DRIVE PX. As Colette mentioned earlier, there are some 80 companies that we're working with, whether it's tier ones or OEMs or startup companies all over the world that we're working with in this area of autonomous vehicles. The thing to realize is you're not selling a chip into that car. You're working with a car company to build an autonomous driving car. That process requires a fair amount of engineering. We have a development mechanism that allows car companies to work with our engineers to collaborate to develop these self-driving cars.
That's what most of that stuff that Colette was talking about.
Our next question comes from the line of Craig Ellis with B. Riley & Company.
Thanks for taking the question and congratulations on the revenue and margin performance. Jensen, I wanted to follow up on one of the comments that you made regarding Pascal. I think you indicated that all Pascal parts had taped out. The question is, if that is the case, will we see refresh activity across all the platform groups in fiscal 2017? In fact, will some of the refresh activity be taking place in fiscal 2018? What's the duration of the refresh that we're looking at?
First, thanks for the question, we don't comment on unannounced products, as you know. I hate to ruin all of the surprises for you. Pascal is the single most ambitious GPU architecture we have ever undertaken, this is really the first GPU that was designed from the ground up for applications that are quite well beyond computer graphics and high-performance computing. It was designed to take into consideration all of the things that we've learned about deep learning, all the things that we've learned about VR. For example, it has a brand-new graphics pipeline that allows Pascal to simultaneously project into multiple surfaces at the same time with no performance penalty. Otherwise, it would degrade your performance in VR by a factor of two just because you have two surfaces you're projecting into.
We can do all kinds of amazing things for augmented reality, other types of virtual reality displays, surround displays, curved displays, dome displays. There's all kinds of holographic displays. There's all kinds of displays that are being invented at the moment, we have the ability to now support those type of displays with a much more elegant architecture without degrading performance. Pascal is, whether it's AI, whether it's gaming, whether it's VR, is really the most ambitious project we've ever undertaken, it's going to go through all of our markets. The application for self-driving cars is going to be pretty exciting. It's going to go through all of our markets. Of course, we have plenty to announce in the future, we've announced what we've announced.
Our next question comes from the line of Romit Shah with Nomura Research. Please go ahead.
Yes, thanks very much. Jensen, I was hoping you could just share your view today on fully autonomous driving because Mobileye's chairman has said very recently that the technology basically isn't ready and that fully autonomous cars won't be available until, I think he was saying 2019. I guess my question is, well, one, I'd love your view on that, and two, whether cars are fully autonomous or autonomous in certain environments, say one or two years out, does it impact the trajectory of your automotive business?
First of all, working on full autonomy is a great endeavor, and whether we get there 100%, 90%, 92%, 93% is, in my mind, completely irrelevant. The endeavor of getting there and making your car more and more autonomous. Initially, of course, we would like to have a virtual co-pilot. Having a virtual co-pilot is the way I get to work every day. Every single day I drive my Model S, and every single day I put it into autonomous mode, and every single day it brings me joy. I'm not confessing necessarily, but texting a little bit is okay. I think that the path to full autonomy is going to be paved by amazing capabilities along the way. We're not waiting around for 2019. We'll ship autonomous vehicles by the end of this year.
I understand that we're three years ahead of other people's schedules. However, we also know that DRIVE PX 2 is the most advanced autonomous computing car computer in the world today, and it's powered by AI fully. DRIVE PX 2, there will be a DRIVE PX 3, there'll be a DRIVE PX 4, then by 2019, I guess we'll be shipping DRIVE PX 5. Our roadmap is just like that. That's how we work as you guys know very well. I think there's a lot of work to be done, which is the exciting part. The thing about a technology company, a thing about any company, unless there's great problems and great challenges that we can help solve, what value do we bring?
What NVIDIA does for a living is to build computers that no other company in the world can build. Whether it's high-performance computers that are used to power a nation's supercomputers or deep learning supercomputers so that we can gain insight from data, or self-driving car computers, so that autonomous cars can save people's lives and make people's lives more convenient. That's what we do. This is the work that we do, and I'm delighted to hear that we're three years ahead of the competition.
Our next question comes from the line of Suji DeSilva with Topeka Capital Markets.
Hi, Jensen. Hi, Colette. Congratulations on the impressive results here. On the data center business, is there an inflection going on with deep learning with the software maturity that's driving this at this point? Can you give us any metrics, Jensen, for how to think about the size of this opportunity for you? I know it's hard, but things like server attach rates, what % of servers you could attach. Will be an M4 in the high end in every box, or maybe the number of GPUs a single deep learning implementation has, something like that would help. Thanks.
Yeah. The truth is that nobody really knows how big this deep learning market's going to be. Until a couple of two, three years ago, it was really even hard to imagine how good the results were going to be. If it wasn't because of the groundbreaking work that was done at Google and Facebook and other researchers around the world, how would we have discovered that it was going to be superhuman? The work that recently was done at Microsoft Research, they've achieved superhuman levels of inferencing, of image recognition and voice recognition that's really kind of hard to imagine. These networks are now huge. The Microsoft Research network, super deep network, is 1,000 layers deep.
Training such a network is quite a chore, it is quite an endeavor, this is a problem that high-performance computing will have to be deployed, this is why our GPUs are so sought after. In terms of how big that's going to be, my sense is that almost no transaction with the internet will be without deep learning or some machine learning inference in the future. I just can't imagine that. There's no recommendation of a movie, no recommendation of a purchase, no search, no image search, no text that won't somehow have passed through some smart chatbot or smart bot or some machine learning algorithm so that they could make the transaction more, make the inference or request more useful to you. I think this is going to be a very big thing.
On the other hand, the enterprises, we use deep learning all over our company today. We had the benefit of being early because we saw the power of this technology early on. We're seeing deep learning being used now in medical imaging all over the world. We're seeing it being used in manufacturing. It's going to be used for scientific computing. More data is generated by high-performance computers and supercomputers than just about anything. They generate it through simulation. They generate so much data that they have to throw the vast majority of it away. For example, the Hadron Collider, whenever the protons collide, they throw away 99% of the data, they're able to barely keep up with just that 1%. By using machine learning and our GPUs, they could find insight in the rest of the 99%.
There are just applications go on and on, people are now starting to understand this deep learning. It really puts machine learning and puts artificial intelligence in the hands of engineers, it's understandable. That's one of the reasons why it's growing so fast. I don't know exactly how big it's going to be, but here's my proposition, that this is going to be the next big computing model. The way that people compute. That in the past, software programmers wrote programs, compiled it, in the future, we're going to have algorithms write the software for us. That's a very different way of computing, I think it's a very big deal.
Our next question comes from the line of David Wong with Wells Fargo. Please go ahead.
Thanks very much. In automotive, what products are your revenues coming from currently? Is DRIVE PX at all significant, or are your sales primarily NVIDIA DRIVE CX or something else?
The primary parts of our automotive business today comes from infotainment and the premier infotainment systems. For example, the virtual cockpit that Audi ships. The vast majority of our development projects today come from DRIVE PX on those projects. We probably have 10 times as many autonomous driving projects as we have infotainment projects today, and we have a fair number of infotainment projects. That gives you a sense of where we were in the past and where we're going in the future.
I'm showing no further questions at this time, Mr. Chanda. Please, I'll turn the call over to you.
We've had a great start to the year with strong revenue growth and profitability. Pascal is a quantum leap in performance for AI, gaming, and VR and is in full production now. Deep learning is spreading across every industry, making data center our fastest-growing business. With growing worldwide adoption of AI, the arrival of VR, and the rise of self-driving cars, we're really excited about the future. Thanks for tuning in.
Ladies and gentlemen, that concludes today's conference call. We thank you for your participation, and we ask that you please disconnect your line. Have a great day, everyone.