Ladies and gentlemen, please welcome to the stage Vice President, Investor Relations, Mark Henninger.
Good morning. I'd like to welcome you all to Intel. I'd also like to extend a welcome to everyone who's joining us via the webcast today. We're excited to have you here. We have a full agenda. I'll share the schedule for the day in just a minute. Before we begin that, I'll touch on a few logistical details. First off, if I can ask everyone to check their phones and make sure that those are set to silent, that'll help us make sure that everyone can hear the presenters. We do have a guest Wi-Fi network. You probably saw it up on the screen here earlier. It's named IM2017, and the password is all lowercase investor. IM for Investor Meeting 2017, password, all lowercase investor.
If you need to charge a device, a few of the rows actually have charging blocks down at the base of the chairs. You're welcome to use one of those if you don't have a charging block near you. There are charging stations in the back of the room here, as well as out in the lobby, in the cafe where you checked in earlier this morning. We'll have drinks available throughout the day out in the lobby, feel free to take a quick break if you need one. We'll have coffee and water out there. If you have any questions over the course of the day, we have a team of people around the auditorium and the lobby that are here to help you. You can identify them by the green scarves they're wearing.
Feel free to grab any one of them to ask questions about schedule or logistics. All right. With that out of the way, I'll read the risk factors. We'll move on to the agenda. Today's presentations contain forward-looking statements. All statements made that are not historical facts are subject to a number of risks and uncertainties. Actual results may differ materially. Please refer to our most recent earnings release, 10-Q and 10-K for more information on the specific risk factors that could cause actual results to differ. If we use any non-GAAP financial measures during the presentations, you'll find the required reconciliation to the most directly comparable GAAP financial measure on our website, intc.com. Okay.
With that out of the way, I'll take a few minutes now just to walk through the agenda briefly. We'll start with our CEO, Brian Krzanich, kicking things off with a strategic overview. He's going to focus on the role data plays in our strategy and importantly our growth, as well as the significantly expanded TAM opportunity that's resulting from the investments that we're making in research and development. He'll be followed by Dr. Murthy Renduchintala. Murthy will build on Brian's themes, touching on a wide range of topics from client computing to Moore's Law to our work in 5G. We'll invite Diane Bryant up onto the stage. She'll be walking through our data center strategy and importantly, a rich look into the trends that are playing out in her business. As Diane wraps up, we'll break for about 30 minutes.
We'll have coffee and snacks out in the lobby. As is our tradition, we'll have the executive team circulating for informal conversation and networking. As we come back from the break, Rob Crooke will give us an overview of the opportunities that we see resulting from having a pair of differentiated and disruptive technologies in both 3D NAND and Optane. Rob will be followed by Doug Davis, who will talk through our IoT strategy, emphasizing the four big verticals that we're focused on in that segment. Our CFO, Bob Swan, will take the stage with the final keynote. He'll be summarizing what you've heard over the course of the day, and importantly, putting all of that into a financial context.
We'll wrap things up by inviting all of the speakers to join us up here on stage for a roughly 30-minute or so Q&A session, again, with all of the speakers. At that point, we'll conclude the webcast. We'll break for lunch in the cafe. Again, the cafe is where you checked in first thing this morning. We'll have execs at each table. That'll be another opportunity for informal networking. You should have your table assignment on the back of your name tag, which you should have received when you checked in this morning. If you didn't receive that or can't find it, we'll have hosts at lunch that'll help you find your seat. All right. With that all done, I'd like to invite our Chief Executive Officer, Brian Krzanich, up on the stage to kick off the day. Hey, Bob.
Thank you. Okay, good morning. Welcome to our Investor Day 2017. I also want to just take a second and welcome Bob Swan. Let's see. Where is. Oh, there he is, right there. This is his first Investor Day as CFO of Intel. He's really helped us craft this agenda and really try and answer the questions that you guys have around this company and exactly where we're headed. I think Bob's done a wonderful job of guiding us through this. I also think you'll see the same transparency and clarity that you're used to from when Stacy and even prior to that Andy were here, and Stacy and Andy are both here. We have three lineages of CFOs in the room today with us. With that, I've got three things I want to do with my time here this morning.
The first thing I want to do is go through our 2016 results. I think 2016 was just a great year, I want to spend a little time talking to you about that. The second thing I want to do is really talk to you about the data, how data is really driving innovation and growth at Intel, how we think about that, how we look at the expanding TAM that we have available to us as a result of that. Finally, really take that and turn that into how is that shaping our strategy. We have what we call our virtuous cycle of growth that really is driven by that data-centric view of the world. Those are the three things I'm going to spend my time with you this morning on. With that, let's go right into our 2016 results.
I think if you take a look at 2016, all I can say it was just a phenomenal year, a record year across many, many segments. Revenue at $59.5 billion, gross margins at 63.2%, earnings per share at $2.72. The key messages here is that we really truly had a record year. I'm really proud of what the organization has delivered as a result of this. Now, record years need to be turned into record results from a shareholder perspective, I think if you take a look at our three-year shareholder return, it's really done very well as well. It's outperforming the market. We've had 56% return over the three-year period that we're talking about. It's driven through consistency across our businesses, contributing really to this fantastic performance.
We're very proud of this, both our financial results, but also our shareholder return results, which are demonstrated by this chart. What I'd like to do now is kind of go through it by segment, go through by business group. The first one is the Client Group. The Client Group, all I can say is wow for 2016. At every point, you have to take a look at this and say, "What a great year." On a unit level, it was a down year on units. They were able to, through a great strategy of segmentation, of great leadership products, of strategically partnering with the OEM partners, focusing on the key areas that are driving performance-related products like virtual reality and gaming and the enthusiast brands. We were able to actually show great growth. ASPs were up. Profitability was up.
All of this in a market that actually continued to have down units. This is really, in my opinion, great engineering combined with great business acumen to deliver a record year in the Client Business for 2016. It was also a great year for the Data Center Group. If we take a look at the Data Center Group, it was really defined by three primary areas of growth. We continued to grow very well in the cloud area. We continued to grow, it's averaging somewhere in the 20-ish%. It's 25, 26, 22, but it's somewhere in those mid-20s growth rate for cloud growth rate. The adjacencies, I'd tell you, this was the year that the adjacencies that we've been talking about started to hit the ground running. Silicon photonics, Omni-Path Architecture, you'll see this year 3D XPoint or Optane also kick in.
These adjacencies are what we think is one of the big fuelers of growth as we move forward now. You saw them start to hit the data center in 2016. The area where we really have a lot of opportunity, we continue to believe, is our networking space. Networking is actually the fastest-growing segment of the data center. As the infrastructure continues to move to software-defined network function virtualization and really preparing for 5G, which if you understand 5G, it has to live in a software-defined environment. It has to be able to morph between workloads and spectrum as it moves through the day. You have to have a software-defined network in order to do that. The networks are preparing for that, and you can see the business growing from our standpoint.
If you take a look at some of our other businesses, beside these two core businesses, the IoT Group continues to grow. It continues to grow double digits. It's really driven by three key areas: retail, industrial, and security cameras, basically. Those three areas are driving the growth. We presented at NRF in January, which is the National Retail Federation. It was amazing the interest we saw in both in brick and mortar in order to use IoT to actually compete with the online and actually use data to their advantage. What they're finding is that people who come into the stores, there's fewer people, but those people who do come in have intent to purchase, and so data can really help them in that space. It help them close those sales.
We continue to think that IoT is a great opportunity for us for growth in those three segments. One of the ones I'm most proud about is PSG, formerly known as Altera. Altera was an acquisition we closed just in January of 2016. We've done a phenomenal job, in my opinion, of integrating Altera into Intel. If you go over to the Altera buildings right now, they're about two miles away from here, I think it would be a seamless transition between Intel and Altera. You'll find many Intel faces over there, and a tight integration between the engineering teams. Dan McNamara is in the audience here today. Dan's running this group. What I'm really proud of for this group is two things for 2016. One, the first 14-nanometer FPGA shipped to customers, which is on Intel technology.
Lastly, NSG. NSG is our memory business. You're going to hear Rob up here on stage this morning. We look at this as an opportunity driven by the leading-edge technologies that we have. The 3D NAND, which we believe to be some of the most highly efficient, cost-effective, and powerful memory technology. We've started up our Dalian factory, Fab 68. It's come on with yields and productivity equal to or better than the existing factories. We have our Optane or 3D XPoint technology, which is a game-changing technology which we've started shipping in the first quarter here.
Will really shift the architecture between memory and storage moving forward. We'll see this across many, many platforms, from client systems to the data center. We see memory really gaining momentum as we go out through 2017 and build out Fab 68, and really launch the remaining products of Optane technology. One of the ones you saw up here on one of the videos I was watching just before we all sat down, which was the drone light show, and if you watched the Super Bowl halftime, you saw our drones as well. This is really about new technology and testing and making sure we understand where is technology heading, and really where do we want to make sure that we're touching, and part of those new emerging technologies.
You see our RealSense cameras that are going into drones and robotics, our virtual reality systems like we announced at IDF this summer and showed again at CES, our Project Alloy, which is a tetherless system that could really do merged reality, bringing the real world into the virtual world and vice versa. We think these are emerging technologies that we want to be a part of and we want to be a leader in. We'll talk about how do those fit into the strategy here in a little bit, but I'll give you a hint. All of the things that we look at in this space all either produce or require large amounts of data. They acquire or produce data that has to be computed on. There's some interaction between artificial intelligence and these areas as well. That's really how we think about these.
You'll also see if you watch the Super Bowl, the 360 replay and the player view is another one that's a massive data acquisition and manipulation problem. That's how we look at the new technology group. If I take a look at 2016 overall, I look at it as it was one of the big years of transition, probably the biggest year of transition in Intel's history. We had executive leadership changes, and if you take a look in the room today, it's going to look very different than it did, say, last year even at this meeting. Some of the faces are familiar, like Stacy, but they're doing something different now. They're doing something that is causing them to grow as leaders, but it's also bringing new ideas and new ways of thinking into those existing areas. You're going to see new faces.
The newest I use is Bob Swan, who's our newest CFO, but you're going to see it in IT. You're going to talk to Murthy this morning. You're going to see it across the board. If you go one level down, that same kind of leadership change is occurring there as well. It's really bringing both an outside-in perspective to Intel, but it's also changing the inside perspective as people move around into new areas. We did the restructuring and reinvestment, which you saw in the spring timeframe. Around April, we announced, and we are continuing to execute through. Remember, that was both a combination of restructuring how and where we're investing, and Bob's going to talk to you a little bit about that, but also shifting where we're putting our resources and our energy into as a company.
We'll talk about that a little bit in my presentation, aligning to the strategy. One of the ones that I'm most excited about is focusing our sales force. In the past, our sales force was broken out by geography, and just mainly focused on, I'll call it client and data center. Things like IoT or software were kind of afterthoughts. They were other areas that if you could get into that OEM or you could sell an existing OEM, those products, they were done. What we've done is restructured all of our sales force, the largest transition we've done in sales, aligning all of the sales teams to the various business groups. Now you have a sales team specifically focused on IoT, for example, and their job is to win designs and win product wins in IoT.
You have a dedicated data center team now, and that data center team is now focused and reporting to Diane on sales. You see a much tighter link between the sales and the business units now. To me, as I look out into the future, this is going to be one of the biggest drivers as we've talked about this expanding TAM and new opportunities, aligning the sales force so that they're driven to those new areas is going to be very critical to the growth of those. I'd like to shift now to what's driving our growth, and I want to really spend some time with you talking about data and how we think about data. You're going to see an underlying discussion throughout almost every presentation today about data.
How data is thought about and how it's growing, but also what's going to happen with that data, what makes it useful. What makes it useful is when you apply analytics to it, when you gather that data in the cloud, apply analytics, and then push that back out, whether that's an autonomous car, the wearable on your wrist, the robot in your house or in your factory, whatever that is, it's about applying those analytics and then pushing that data back out. We believe that data is truly going to be the driver of growth and innovation at Intel, and you're going to see continued changes and focus this year on this specific target. First I thought we'd do is let's talk about where data is headed. Now, these numbers are 2020 numbers.
If you take a look at today, the average person produces about 650 megabits of data per day. That's you posting on Facebook and tweeting out your tweets and sending out your emails and all of those things that we typically do, 650 megabits a day. By 2020, the estimate is that'll grow to one and a half megabits a day. More of us will put out video. More of us will put out more complex and high-definition photos. You may be creating your own virtual reality systems at some point, or virtual reality 360 images or videos as well. All of those things are going to cause your data output to more than double over that time.
That pales in comparison, what I tell you today, or what I talk to people about is the cloud of today is built mostly on people's data, the cloud of tomorrow is going to be built on all of these machines connecting to the cloud. The average autonomous vehicle will put out approximately 4 terabytes of data per day. Just think about all of the images that it's going to be collecting as it's driving around, understanding where all the reference points are, updating the maps that have to be precision maps that are accurate to less than 10 centimeters. All of that data has to be pushed back up into the cloud, analytics applied to it, then pushed back into either the mapping or the learning systems that go into the car. The average connected plane will be 5 terabytes a day.
A smart factory, like one of Intel's factories, is about a petabyte of data a day. If you look at the cloud video providers, depending on who you want to look at, Netflix or whoever, they're pushing out about 750 petabytes a day by 2020. Things like Intel's Replay or the 360 Replay like you saw in the Super Bowl, or if you watch the NBA All-Star Game next weekend, you'll see that as well. It puts out about 2 terabytes of data a minute. A minute. Okay? As all of these new models and new methods are being applied to what we like to do, watch, work, data is going to explode across this whole network. We are a data company. We are a company that will grow off this data explosion that's going to happen over the next four or five years.
Every single part of this strategy and every area that we look at, you should be able to ask yourself or one of us in this room, how does that apply to data? Why does data fit into that? We should be able to answer to you, whether it's a robot, whether it's a data center appliance, whatever it is, we'll be able to explain to you how it fits into that data explosion. That is how we define what we do, what we invest in, and what we move to. We look at this as a massively expanding TAM. Let's talk about our strategy. Okay. Our strategy, we like to call this the virtuous cycle of growth.
At the center, what's really fueling this virtuous cycle. For those who don't know virtuous cycles, it's really a term that's used that when you could take any part of this circle, when you add something to it feeds all the rest of the parts, and the rest of the parts speed up or grow or accelerate. That's how we really think about this, what feeds this is really data. That's where people are really focusing on. You start down at the bottom with the things and devices. Those things and devices, whether it's the autonomous car that you're driving around in or the wearable that's on your wrist, the PC in your office, the virtual reality system, all of those systems are going to be generating or using large amounts of data.
Those systems are going to require a high degree of connectivity. Most people think autonomous cars will require a 5G connection in order to push the maps out to the car, considering the density of the maps, and in order to be able to understand traffic patterns and what's going on. Most people think robots that are going to be ubiquitous around factories and neighborhoods in the future are going to require 5G connections. We need to have this high-speed connectivity. That will be an important part of the accelerant of this. Lastly, all of those things, whether it's your virtual reality system or whether it's your robot that's running around or your autonomous car, are really not functional, not useful without a cloud. The cloud is going to take that data, apply analytics to it, and push it back down to whatever that device is.
We believe that there are accelerating technologies, things like FPGAs, things like 3D XPoint, things like silicon photonics, things like Omni-Path Architecture, that can accelerate within here and drive performance in these various segments. Something like 3D XPoint will drive acceleration both in the things and devices. It allows for large amounts of memory or storage-like capacity to act like memory and performance. We'll talk to you today about how it can apply to client devices. It'll also apply up into the cloud as big data and analytics are used. You're allowed to bring terabytes of data to memory-like performance in the cloud, and that really accelerates the big data applications, the AI applications, all of those types of heavy users of data in their applications. FPGAs are the same way.
It's a way to bring those algorithms that are used for AI right down into the silicon and speed those up. Also on the client in the car, it can be used as well, for example. Most of the autonomous cars are combinations of Xeons and FPGAs, where the FPGAs are really helping with the software load on the learning system and really applying the algorithms that are driving the car, basically, targeting where the car drives. We look at this data explosion as an expanding TAM. TAM that is either not a place where Intel is a large player but exists today and it's going through a transformation. We talked about connectivity and the networking there, but also new areas that are nascent and there aren't market leaders, and the technology's just being developed, and the autonomous cars are an example of that.
A traditional market view of Intel may look like this. Our TAM is mainly driven by the client and the data center systems. The data center is going to be mainly, when you think of Intel, our Xeon processor. When you take a look at that TAM, it's about a $45 billion TAM. That's probably how people think about Intel today. The way we look at the world, though, is a bit different. We look at the world across all of these segments saying there's a consistency across here. There's a consistency in each one of these segments that allows us to take the technologies, the knowledge from those core areas of the data center and client and expand them as data really is driven into these areas, and as data becomes more and more important through this decade.
We look at the world as a much, much larger TAM, as a TAM that's expanding at an even faster rate than probably I predicted on that foil a while ago. My guess is my 2020 numbers, the one thing I know about them is they're wrong, and they're probably an underestimate. Right? Let's talk about these a little bit. What you're going to see is we're making investments in each one of these segments that we believe will drive our market presence, market share, and position in each one of these segments. In data center, which you might say, "Well, where are you going to go there?" Actually, if you take a look at it, we're shifting from a server CPU company to a rack company. We don't think of the data center as a CPU company.
We think about how do we optimize that entire rack, how do we optimize between the racks, and how do we optimize across the data center. We are a data center company that builds high-performance racks, silicon photonics, Omni-Path Architecture, right? 3D XPoint, FPGAs, Xeon Phi, Nervana. All of those things combine to make the complete system that we believe we are now architecting at the system and data center level, expanding our TAM, expanding our opportunity, and delivering customer performance and value well beyond what we deliver today. The third place is memory. Memory, you've heard about, we believe we have leading edge, and Rob's going to come up and talk to you in a passionate, energized way because he loves this business. It's about two technologies. We believe we have leadership in 3D NAND.
3D NAND is, I'll call it, the standard of NAND as we move through 2017 and 2018. We believe our density and our architecture gives us a differentiating advantage over the competition. Rob will talk to you about it, but we made technical decisions over the last few years that give us a leadership that's different than what everybody else has. The big investment and the place that I believe will pay off very well over the next few years is our 3D XPoint Optane technology.
This is a technology that really shifts the entire architecture between memory and storage and really starts to push the limits of where computing technology really wants to go, where you have large amounts of data right next to the CPU, allowing for very high performance in these types of new applications in artificial intelligence, machine learning, deep learning, and even in just some of our existing technologies. I believe 3D XPoint, if you take a look at gaming systems on client, in the future, every gamer is going to want to have 3D XPoint. Every single gamer, right? From top to bottom, I believe 3D XPoint will shift the architecture of memory and storage. Our modem and mobile technology continues to grow, and you saw us get into phones in a big way in 2016.
We have plans and programs to continue to grow this segment through 2017 and beyond. We believe that as the world converts to 5G, the investment and the technology required are connectivity between the modem and the base station and network is going to be a critical advantage that we have. When we walk into customers and talk about 5G, we don't talk about a modem, and we don't talk about a base station or a data center. We talk about end-to-end, and that's why we're winning designs in 5G. It's because we can talk from end to end, and we can solve problems from end to end, and that's what our customers are looking for. Lastly is IoT. IoT, it's a big word, and it's used all over the place, right? I mean, it's from your furnace to your refrigerator to your factory.
We focused on a few key areas, right? Retail, industrial, security in mostly video. You see us making big advancements in ADAS. ADAS is the same thing as the autonomous driving. We talk about the fact that all of these things come together. The high-performance computing in the car, the connectivity with our 5G modems, the data centers, because everybody who wants to build out an autonomous network of cars must have a set of data centers. We are the one company who can walk into these partners and talk from end to end again across that spectrum and bring all of our products, the full suite, into the discussion and solve problems and take ownership for delivering a product.
That's how we look at this much, much bigger TAM than I think people see Intel as, and that's how we're investing over the next few years. Each one of the guys are going to come up here and talk to you about it and give you their view into these areas. Bob will, as I said at the end, kind of wrap that up. Our investment priorities are pretty simple. Cloud, artificial intelligence, and the network. That's first place we go.
We believe that as we look out to the rest of this decade, this is going to be the growth engine of the company, and I call it turns us from a single-cylinder engine, where we're mostly a client-based company, to a two-cylinder engine, where the client and the data center, and when I talk about the data center, a much bigger data center than probably you're thinking about prior to today, is really driving this company. We have what's next, memory, FPGAs, 5G, and data-rich things and devices. That's really how we think about our investments as we go through this. We believe that when we execute on these, we will have access to that expanding TAM, and that's where the growth of the company comes from. Can you go back one? I hit the button twice. Sorry.
When we think about this, let's break it down a little bit more refined. The cloud and data center, what we really think about is really three areas within that. The cloud itself, and you're going to see really what that is about our data center architecture. You're going to see how Diane, when Diane talks about Omni-Path fabric, Intel Rack Scale Design, 3D XPoint, silicon photonics, our Xeon, our Xeon Phi, Nervana, all of that fits into how we build out the cloud infrastructure in a much more performance and efficient way. AI, which we believe we have a very unique end-to-end set of assets for AI, which we'll talk about this morning.
Networking, which as I said earlier, was one of our fastest-growing segments in the data center, and an area where if you take a look at our market share, currently, it's relatively low. But is growing at a double-digit rate, and we believe is a huge opportunity for us as the networks become 5G software defined and network function virtualization. There we go. The next place that we're making these investments is around the 5G memory and FPGA. I'm tying this all back to our strategy in this virtual circle of growth. Rob, as I said, is going to come up and talk about Optane and how Optane plus 3D NAND is going to really give us a differentiating position in memory. We've got our FPGAs. We said we grew faster than market in 2016. We just introduced our 14 nanometer Stratix 10.
Technology at the end of last year. We've got more design wins and customer interest than any product in Altera's history on Stratix 10. There's a series of 14 nanometer products that will come out throughout the rest of this year and into next, and we're already starting to look at 10 nanometer products for FPGAs as well. We believe we can continue to win share and grow faster than market in FPGAs. 5G, which we believe is going to be a transition that occurs as we go towards the back half of this decade, and we're the one company that's well-positioned from end to end. You're going to see many network trials and first runs this year on 5G systems around the world, most of them with Intel. There.
As I said, the last thing is around these edge devices, the retail, and ADAS, we believe these are opportunities for us to continue to grow. Again, it's not everywhere we're going to go. We're going where we believe that there are large data, that data is required. That's where you'll see us in all of these edge devices, why we're choosing retail, industrial, and ADAS, because they require large amounts of data. In retail, for example, at the National Retail Federation, we just announced a $100 million fund to really build out in stores this whole connected store that really gives brick-and-mortar stores the same kind of data the online stores have had.
It'll show what people pick up when they pick up something in the store, what goes into the changing room, but never comes out, similar to what goes into your basket online, but you never end up doing a purchase. Right now, the online guys have had this data for years and put it as an advantage, but the brick-and-mortar guys have not had this. We have a program and a process and technology that gives them that equal advantage. All of these things, we believe, will allow us to expand into these new areas. Our strategy and how we win, we believe we have chosen the right markets with the right timing. We're at a unique position in time where computing, memory, and the requirements for large amounts of data are all converging.
3D video, virtual reality, autonomous driving, all of these usages coming along with things like 3D NAND, 3D XPoint, artificial intelligence, and 10 nanometers and seven nanometer silicon, all of them coming together as we move through the rest of this decade, is a unique time in our history, we believe. We're expanding into that larger TAM, that $220 billion TAM, using domain expertise. We're not going into areas we're not familiar with. This is all about compute. This is all about memory. We know those areas. These are not areas we're new to. They're new applications. We're doing partnerships and M&A. You see something different with 5G where as compared to when we were looking at LTE and we tried to be different, with 5G, we're partnering. We're partnering with all of the top equipment manufacturers. We're partnering with all of the top networks across the world.
We're doing trials together. We're partnering. We're doing M&A like Nervana, like Movidius, to bring new technologies and new computing platforms onto our silicon, like Altera. We believe that those are different than what we've done in the past. We're really thinking about this as an end-to-end solution. We think we're unique when we walk into a partner like BMW or Verizon or anybody else almost on the planet. We talk end to end. We don't talk about a segment. We solve problems. We own problems. We tell them, "We'll come in and we'll help, and we'll build out your autonomous car network. We'll help with the data center. We'll help with the 5G connectivity. We'll help with the in-car compute. We'll do all of that." When we talk about 5G and we walk into a service provider, same thing.
We talk about the modems, we talk about the base stations, we talk about the data centers. We're able to talk all of those. We believe we bring to that on our top silicon and software leadership that's unique. We have the ability to go in and customize parts. The big cloud provider wants to move their algorithms onto silicon, we could help with either FPGAs or custom silicon. We can provide any level of opportunity that they need. Our top priorities for 2017, I challenge the leaders with a few. First is growth in the data center. Growth in the data center through our core products and through our adjacencies. It's the number 1 priority for this company. We must build the second cylinder of this engine.
Second is to continue the great work that the team's done in keeping our Client Business, which I believe is one of the most unique and best businesses just about on the planet, keep it strong and healthy. I think Murti will come up here and talk to you about the roadmap and strategy they have in this space, and we're very excited about it. Second is growth in IoT and these devices that are connecting, from autonomous cars to retail, to industrial. This is the time to get in, do the design wins, and start the growth for the next several years. We need to flawlessly execute in memory and FPGAs. This is about bringing new products that differentiate. As I said, first 14-nanometer product in the world on the FPGA, the most design wins and customer interest in Altera's history, we must execute.
We must deliver those. If we do, the growth we believe in FPGAs can be much, much faster than market. With memory, you've seen the 3D NAND start to grow. Rob will talk to you about the expansion that he has and how he's going to grow that through the rest of this year. 3D NAND is critical. It brings us cost and efficiency and performance in a standard, more commodity memory business that we believe we can actually make money and grow in the future. Bringing out 3D XPoint, the Optane technology, that's how we really shift this whole industry. It's important that we flawlessly execute in these two product areas. This is the team that's committed to go do this, and they're all here today, I think, for the most part. I don't think anybody was traveling today. Maybe a couple are.
I'm really excited about the team that we have in place. We've made a lot of changes over this last year. People are excited and energized about their businesses. They're here today for your access. We really are here, not for us, but for you. We want to make sure you ask your questions, we get you your information that you want, and that you understand this strategy and how it all plays together. Hopefully, the presentations will make that clear, but if not, we have all of these people here today to make time and to be available to answer your questions. With that, I have two minutes left. I'm going to use two minutes to introduce my good friend, and somebody else who's new to this company, but actually feels old to this company now relative to the rest of this organization.
He comes to us from the outside, he's already considered an insider, he really brings us that outside-in perspective. He's brought a lot of new talent to the company. I'm really excited that he's here. He's somebody I'd been recruiting for quite a while to have come here. I'd like to invite Murti up on stage to talk to you about the Client and his whole organization.
Thank you. Good morning, everyone. I think I've seen a lot of familiar faces in the audience. I've also seen some that I don't know. To those that I know, hi. To those that I don't know, we've got a lot of breaks, so let's get to know each other. I've been part of the Intel management team now for about one year. The most prolific question I always get asked is, Why did I join Intel? When you're working for the number 3 semiconductor company in the world, you have two choices if you want to move up in the world. Do I want to move to Korea? Or do I want to work for the number 1 semiconductor company in the world, which Intel represents?
I also wanted to be part of a team that is made up of the best technical experts and technical assets in the industry. I wanted the experience to learn about new markets and new technologies that I was previously unfamiliar with. I wanted to learn firsthand what it was really like to drive Moore's Law and be part of an IDM team. Now, the role I have at Intel, and that I discussed with Brian, provides me with an exhilarating challenge to drive both business and technology. Now, I'm blessed with a fairly cumbersome title under my name, as you saw. It's almost a sentence. It's not a group. What I thought I'd do is spend a little bit of time decoding what it is I actually do and I'm actually responsible for.
On the business side, I'm responsible for Intel's client, IoT, connectivity, and ADAS businesses. Within connectivity, we're talking about both the modem business, and also our connectivity business that includes Wi-Fi and other unlicensed spectrum technologies. Together, they represent about 60% of Intel's revenues. On the technology side, I'm responsible for the company-wide systems architecture organization, and that covers everything from silicon engineering, platform development, and software enablement. But another very important part of my role is to work alongside my colleagues in our technology and development organization within TMG with Stacy and Sohail to really align our process roadmaps with our product roadmaps. That's one of the most vibrant discussions that Intel engages in, matching our roadmap and leadership expectations in process with the kind of great things we want to deliver in our product businesses. Now, given that remit, I'm going to cover many areas today.
By the end of my talk, I'd really like you to take away three key messages. First, the pursuit of Moore's Law continues to deliver time-to-market leadership position in both performance and area. Second, it's the client business that provides Intel the scale to invest to keep Moore's Law moving. Second, as Brian has already very clearly enunciated, the client business is being managed in an efficient and disciplined manner, growing operating profits and focusing R&D investments in high-return areas. Third, in driving the data revolution era, Intel is a leader in defining industry standards and delivering leadership in connectivity products. This is probably best exemplified in Intel's efforts in 5G. Let's take a look at some of the results in 2016 for the businesses I'm responsible for.
I'm going to focus my comments on the client businesses, because Doug, who's coming up later, is going to speak in a lot more detail about IoT. As Brian said, he stole the adjective I was going to use, so I'll use a different one. 2016 was just an absolutely hit it out of the park year for the client group. We were able to deliver a 30% year-over-year increase in profit. That's on a business that's generating over $30 billion a year. As Brian said, there are almost no franchises I can think of that are of that scale, yet we were able to achieve such a significant operating profit. That wasn't done just by cutting spending. We clearly pruned our investments in areas where we felt we were not going to get the return on that investment.
I think we did a fantastic job in really understanding the growth segments of the market and targeting our R&D to deliver leadership products that delivered unique and incomparable experiences in those areas. That really drove up ASPs. It was a mixture of efficient investment, really focusing on segmenting the market, understanding where people were going to pay for performance, driving up ASPs, then working really well to bring the product costs down to efficient levels. As I said, that generated 30% year-over-year operating profit improvement. We did that in an environment where the TAM was uncertain. We didn't know whether we were looking at high single-digit declines or low single-digit declines. Impervious to that, we delivered stellar business results. As I said, our disciplined approach focused on targeting our investments at growth areas.
Let's look at some of these growth areas and how we attack them to grow both market share and higher ASPs. It's very important to understand that the PC business is not monolithic. I'm learning this as a freshman on the team, and it was very educative to me to understand just how complicated and diverse the market is. Sometimes we talk about IoT, and we're really talking about a business that has a dynamic range of DC to light. The PC business is not that different if you look at how much we put under that banner. By differentiating on performance, form factors, and great experiences, we continue to drive an exemplary segmentation strategy in this area. Our overall business is comprised of a number of components that are both solid and many growing in the years to come.
As I've already stated, it's in these growing segments that we've targeted our investments. Gaming is a great example of where we see growth. Let's talk about esports for a minute. In-stadium events like the Intel Extreme Masters and groundbreaking online platforms like Twitch have driven explosive growth in this area. Esports enthusiasts, when they leave the stadiums, when they leave that online experience, are craving to indulge in the same experience they've just witnessed their gaming heroes indulge in. To satisfy that need, Navin and the team came up with the Broadwell-E platform. We defined it and we launched it this year, really targeting those gaming enthusiasts. That was a 10-core product with an ASP of over $1,000. That was really driven by, as Brian said, understanding where the gaming environment was going, looking at the growth of online gaming.
I gave a talk at IDF this year, I had one of the key players from Twitch with me. Yeah, she's got probably more followers than Cristiano Ronaldo or some other huge star out there. I really was blown away at how many people were looking at this young lady playing a game for three, four, five hours. After they'd done that, they didn't want to take a break. They wanted to play the game themselves, but they couldn't get the kind of performance of machine that they needed. That's what we provided when we delivered that Broadwell-E 10-core product. Turning to our adjacent businesses, which consist of modems, gateways, Wi-Fi, and Thunderbolt. In aggregate, these businesses contributed over $2 billion in 2016. That's a 40% year-over-year growth. As Brian said, really, really pleased with our progress in the modem space this year.
We launched our first large-scale activity this year, and we also prosecuted an extremely exciting roadmap, the evidence of which you'll see in the next few months and next few years. I'm really pleased with the integration of our Lantiq acquisition that drives our gateway business. I think we're making large strides there and capitalizing on the connected home, and its growth in terms of capability and functionality in the home. Our Wi-Fi business goes from strength to strength, and I'm delighted with the share we've been able to gain in the PC space. Our Thunderbolt strategy is continuing to provide wireless transfer experiences that no other technologies can match. People are recognizing that and now integrating that into their product. In our adjacent businesses, we've grown tremendously because of targeting our R&D at those growth areas.
The bottom line for the client business is we're able to address opportunities where customers are demanding performance and are willing to pay for it. As a result of that, we grew core mix in 2016 to an all-time high, over 70%. Concurrently, we drove Core i7 mix to another all-time high as we exited the year. That mix, I'm sure you'll agree, represents stellar results for a client business that, as I said earlier, was coping with a fairly uncertain TAM environment. Going forward, we'll continue the same strategy that has worked so well this year. We'll continue to innovate and target our research and development at areas we believe are going to represent opportunities of growth, and do that in a way that is exemplary.
We'll accelerate areas where we basically believe growth is going to be fueled by our innovation and what we can bring, particularly in the areas of VR, enthusiast gaming, smart home, two-in-ones. We're confident that with that pipeline of leadership products, that we're going to have a performance lead over our competitors in this space. I think the year is going to be difficult to beat that we just passed, but the pipeline of products, the attitude of wanting to repeat success and victory, and I think the fusion of our development cooperation with our process technology colleagues, I think is going to make 2017 just as good, if not better than 2017. Navin's got a large task ahead of him. Having done it in 2016, I know he can do it in 2017.
The foundation of our success in the client business, and for that matter, in all the other businesses that Intel has delivered successfully, is our process technology. I'm profoundly proud to be part of the company that both defined and continues to prosecute and drive Moore's Law. To Sohail and the rest of the TMG organization, it's actually incredibly inspiring to work with technologists entrusted with delivering on that responsibility year, after year, after year. At Intel, I've come to learn that we focus on a metric known as effective logic cell area. Different to gate pitch, which having spent the last decade working in the fabless environment, people really use as the normal vernacular. We believe that when we talk about effective logic cell area, we're talking about a much more understandable and appropriate attribute of how to describe scaling in logic and SRAM technology.
The effective cell area is pretty simple, really. It's the product of the cell width and the cell height. If you tabulate that metric against time, it's very clear that Intel has been able to reduce the logic cell area at a rate of 50% every two years. When you compare that against others, three important points become very clear. First, Intel scaling rate puts us on a more aggressive trajectory. Second, this scaling rate provides us with a three-year lead over our competitors. Third, if we project these current trajectories forward, we will continue to maintain a three-year lead even after our competitors deliver their 10 nanometer plans during the course of this year. The economic dimension of Moore's Law is also critically important.
The charts behind me, while simplistic, show very clearly that the economic side of Moore's Law is alive and well despite rising wafer costs. Our projections for seven nanometer give us the confidence that this trajectory, in an economic sense, will continue. Node transitions are only part of the story. One of the things we've worked really hard this year to do, and will continue to do, is to optimize our process and products within a technology generation to deliver meaningful performance and maintain an annual product cadence that's so important to our client and other businesses. As we move through our sixth, seventh, and eighth generation cores, the last of which we'll launch in the second half of this year, we're delivering significant performance improvement on industry standard benchmarks such as SYSmark.
The kind of improvements that we're generating are very meaningful, and they're also being delivered with a substantial reduction in power as we move through that 14 nanometer portfolio. We believe that the combination of performance improvement and power reduction are going to basically be a key factor in another critical strategy for us, which is to drive the annual cadence and refresh of PCs. We certainly believe that these kind of performance numbers are going to be a stimulus to delivering that. The improvements we're delivering have been achieved through a combination of continued process improvements and, just as importantly, continued design innovation in our microarchitecture and the efficiency of our implementations. These results fundamentally come from the advantage of being part of an Integrated Device Manufacturer, as does the other businesses within Intel. It's a clearly differentiating element of what we're able to do.
In an integrated device manufacturing environment, we're concurrently developing process and product architectures. In the fabless industry, which as I said, I've spent over a decade being part of, process technology and product development are largely separate domains with some alignment very late in the marriage, it's very limited with what you can do at that stage. In the fabless industry, a lot of the transistor technology that's being developed is being targeted for a very wide dynamic range of applications, therefore you're constantly having to make compromises between, for example, the peak performance you can achieve versus the power that you need to operate at. Those constraints to maximize the volume aperture that a particular node can service naturally bring to the surface compromises and loss of execution and optimizations that certain specific applications may very well benefit from.
At Intel, there's a fusion between process technology and product development, resulting in customized transistors and IP libraries for specific product segments. The one thing I can say about Intel's view on process technology is we are not a jack of all trades and master of none. Now, I wanted to give you an analogy to bring it to reality of what I really mean by that. I went through a lot. I picked one. Tell me later if you think it's cheesy or not. Basically, the best analogy I could pick was that of a baker who prides himself in creating artisanal bread. Now, most bakers order their wheat on a wholesale basis, it's pretty much the same wheat that's delivered to every baker. At Intel, the baker partners with the farmer in the husbandry of that wheat. We walk the fields together.
We look at where the sun's shining. We look at where the crops are best suited. We look at what the irrigation is like. We look at the soil quality, we select and incorporate the best wheat germ we can create as a result of that dialogue into products of incomparable quality. Now, it's really all about being able to work at the very elemental level of the key attributes that ultimately are what our product's built on, that gives us that edge. We can look at the transistor physics three years before we want to deliver a product on those and start thinking about how the decisions we make at that point are going to lead to product leadership three years later.
When you have those discussions, when you look at the interaction of ideas, when you look at the quality of the intellectual debate that goes on, it's really clear to me why Intel where it's at. As I said, I've spent 10 years in the fabless industry, we've had great discussions, nothing to that elemental depth. I think that really is the real nuclear heart that drives Intel's product leadership. It's the tailoring of our product process technologies to our product requirements that provides us the performance advantage, as I've said, that few in the non-IDM environment can match. Let me give you an example about that. We're able to use different flavors of transistors to meet the specific power versus performance versus area requirements for our CPUs and GPUs in our cores.
We're actually using different types of transistors for targeting graphics functions versus CPU functions because of the difference in the load they bear, and therefore seeking to get differentiated performance between how the CPU performs and how the GPU performs. That's a level of granularity that really is very difficult to achieve when you don't have the intimacy of cooperation between our process and development engineers. This tailoring of performance leadership encompasses the entire product portfolio of Intel. I've given you one. Over coffee, during dinner, over lunch, we can share many, many others. Sohail and Stacy can give you many more examples to supplement the one that I gave you.
One thing I want to make a shout-out for, on the whole technology development side and its integration with our product development process, Stacy and Sohail are going to be arranging a full day on technology development and manufacturing during the first half of this year. I think that's something that will provide much more detail, a much more incisive description that lie behind some of the data that I presented today. To conclude my talk today, I'd like to address how leadership products are propelling Intel's position to take unique advantage of the data revolution that Brian talked about. Now, the data explosion referenced by Brian is already beginning to envelop our world today. It's this data explosion that underpins Intel's strategic framework of the virtuous cycle of growth, as Brian clearly identified.
This explosion of data will usher in an era of profound transformation, and that transformation, I think, will be as tectonic as the shift from analog to digital was. The demand for high bandwidth, low latency communications will create a need for new infrastructure and will drive the advent of what we call the interconnectedness of everything. Three key factors will define this new era of connectivity. One, computing will become pervasive and ambient in everything and everywhere. Second, compute analytics storage will be distributed into the very fabric of the network as data becomes aggregated locally to its generation, its ultimate use. Third, low latency, high bandwidth connectivity will be the lifeblood that binds everything together. The fusing of this compute and communications capability will define the next era of innovation for our industry.
I hope the next few charts in the forthcoming presentations will show you that Intel is well positioned to take advantage of that shift. Key strategic bets within the virtuous cycle framework are artificial intelligence, virtual reality, 5G, and ADAS. Diane's going to talk more about ADAS-- I'm sorry, AI, in her presentation. I know Doug's got a great session lined up on the autonomous vehicle in his speech. I'm going to focus my comments today in the areas of virtual reality and 5G. Virtual reality itself is a segmented market. Brian highlighted the size of the opportunity in some of his remarks and also talked about Alloy. Some of the most profoundly immersive VR experiences will only be delivered in conjunction with high-performance PC platforms. In fact, the purchase of PC VR headsets is growing at a 70% rate year-over-year.
In 2017, the TAM for these devices is expected to be in the millions of units. This will drive upsell to higher performance CPUs as consumers make virtual reality a primary purchase criterion. Let me share some really interesting data with you. We used an industry standard benchmark called VRMark. It's a benchmark that measures via a number of attributes, the degree of immersiveness of the virtual reality experience. Against that standard benchmark of the industry, what we're able to show as we go through our core portfolio is a substantial improvement in the VR experience. This data is from our seventh generation Kaby Lake range, and what we're showing is that going from a Core i3 to a Core i7 drives a 36% improvement in the VRMark score. Very clear benefit, and one that we believe will drive upsell in the PC space.
We think that the evolution of VR is going to really stimulate that upsell of the PC core. Going forward, we see VR continuing to evolve as we go from tethered headsets with increasing functioning and higher speed connectivity through technologies such as Thunderbolt, and they'll be augmented by lightweight tetherless versions as things like technologies like WiGig start penetrating the VR experience. Intel is investing in and delivering platform technologies such as Thunderbolt, WiGig, and RealSense, that will be essential ingredients to the future of VR headsets. Now let me turn to 5G, and this is an area I'm really passionate about, and Intel is investing with an ambition for leadership here.
I spent the last 25 years of my career in the wireless industry, and one thing I've learned is that scale in wireless, and in many other industries, comes from participating in industry-wide standards and collaborating with partners. Different from the past, and I really think this is true, different from the past, this has been a position that Intel has taken from the inception of its 5G development program. Intel is perhaps the only company that, as Brian said, can provide that end-to-end 5G solution, all the way from the cloudification of the RAN through the delivery of consumer devices. Somebody asked me is cloudification a word when I was putting my speech together. My excuse is English is a living language, so we can make whatever word tailors to our meeting. I basically say that today cloudification gets added to Webster's Dictionary.
As a demonstration of this leadership, Intel has made a number of announcements across the spectrum. I'd like to highlight a few, because when you put them together, I think it gives you a real sense of tangibility in terms of what we mean by that end-to-end capability. In the cloud space, Intel is a member of the XRAN initiative to standardize a software-based RAN, based on the feedback we've got from operators who are really demanding that extra level of performance. In the core network, we're partnering with AT&T, Ericsson, Nokia, Cisco, and a large ecosystem of others to deploy solutions using NFV and SDN. In the access networks, we're authoring standards together with Docomo, KT, SKT, and Verizon. We're working with telecom equipment manufacturers to deliver to their very demanding requirements.
In the wireless technology space, Intel is playing an active role in standards bodies that are defining millimeter wave and the new 5G NR specifications. With respect to devices, at CES, we announced Intel's first global 5G modem, with samples shipping by the end of this year. What's really exciting to me about that event is that that product will use Intel's 14 nanometer technology. Starting in 2018, we expect our 5G product to ship in the millions, this will be followed up later in 2018 with LTE products that will ship in the tens of millions using the same indigenous 14 nanometer technology. Lastly, on devices, Doug Davis will talk more about the Intel GO platform and our partnership with BMW and Mobileye.
Let me sum up Intel's play in 5G like this: we're perhaps the only company that can have a meeting with a telco or an infrastructure partner, simultaneously converse with their CTO, their CIO, and their CMO. That comes only because of the breadth of our portfolio. To wrap up and to quickly review the key points of my talk today. First, our pursuit of Moore's Law continues to deliver time-to-market leadership in both performance and area, it's the client business that provides the scale to keep that investment going. Second, the client business is being managed in an efficient and disciplined manner, growing operating profits and focusing R&D on high-return opportunities. Third, in driving the data revolution, Intel is a leader in defining industry standards and delivering leadership connectivity products. As I said, this is best exemplified by our efforts to lead in 5G.
In closing, I really do believe that Intel is well positioned to take advantage of the imminent data revolution because of our unique set of process and product assets. Before I say thank you and goodbye, my last duty is to introduce Diane, who will talk about our data center business. Thank you. Okay. Good morning. Let's see if this works for me. I'm going to give you an update, obviously, on the data center business. Hopefully, give you lots of data to answer all of your questions. I'll start with what you're going to hear. You're going to hear, first of all, that the fundamental drivers of the business are unchanged. It's, number 1, this pervasive move to cloud computing as the most efficient way to deliver information and communication technology.
The second is the transformation of the network, moving from proprietary fixed function boxes onto industry Intel standard servers, as the way to deliver the reach and the breadth and the low latency and the high bandwidth needed in 5G, as BK talked about. The third is this massive amount of data that we are processing, enabling, delivering through artificial intelligence, creating new insights, and delivering new services.
Second, as was clear in our 2016 results and also in our statement regarding 2017, enterprise IT as a whole is in a period of significant transformation in the move and adoption of cloud computing. The tech companies that directly serve enterprise IT are also going through a substantial change. That transformation is creating a bit of a headwind for us in the near term, but we'll get through that, as you'll see in the growth projections out to the future. Third, what gets us all very excited is that we are in a very unique position to win in the data center as the world's dependency on information and communication technology just continues to grow. We have an array of unparalleled assets to address the demand and an array of capabilities, including Moore's Law and our process technology, as Murthy clearly talked about.
Okay, that's what you're going to hear. To start, I want to orient you on the way we drive our business. Okay? Because I know it gets a little complex and complicated for us as well as everyone else. We look at the end users that actually procure the systems, that's either enterprise and government or the cloud service providers or the comm service providers. We look at the systems that they actually procure, and that's either server, storage, or network infrastructure. We look at the wide array of products that we sell into those systems. The way we describe our world is who bought the box, what is the box, and what's in the box. Okay? That's how we think about it. When you heard BK and Bob give Q1 earnings, you heard them talk about who bought the box. Okay?
Enterprise was clearly the challenging segment, as I talked about. We saw a 3% decline across server storage and network systems, and the driver of that decline though being servers. Comm service providers continued to show great growth, volume and ASP, as we'll discuss, up 24%. I'm sorry, the cloud service providers. Comm service providers also showing outstanding growth, 19%. Net that out for the year was an 8% growth year-over-year for 2016. In the second half of 2016, enterprise fell below 50% of our total revenue for the first time ever. We are at 49% of our total revenue was contributed by the enterprise segment. Looking forward to 2017, our expectations on enterprise servers, so the box, remains the same as 2016. We're not expecting any recovery. We're expecting that continued decline in that segment.
As was noted in the earnings call, the total year results we're projecting for 2017 are high single digits. Okay? Okay. Now if you look across the planning horizon, which is 2017 through 2021, and we look at who buys the box, enterprise and government, we're projecting now is low double digits. We do not forecast recovery in the server space across this horizon, but we do see continued strong growth in network and continued strong growth in the adjacencies. The cloud service provider market continues to grow nicely, growing at greater than 15% over that horizon. The comm service provider also continues to grow very nicely, over 10% growth across CAGR across that horizon as well. Okay? That then translates to low double-digit CAGR out through the horizon.
I know I've heard how badly some of you just want me to say 10% growth, but I'm going to say low double-digit growth out through the horizon. Okay. Much of that future growth is in the adjacencies, and you can see that there. The adjacencies for us is anything that's not a CPU. Those adjacencies are growing at a 30% CAGR. Very nice growth as you can see then out through the horizon, contributing four points plus of that low double-digit growth. On the revenue CAGR decline from historical, we have historically talked about a 15% growth CAGR, so I just want to give you some insight into how do you go from 15% to low double digits. The big driver is clearly the stall in enterprise servers. That's the biggest contributor to that.
Last Investor Day, just to refresh you, last Investor Day, we said servers purchased by enterprise were projected at mid-single digits, so positive growth, mid-single digits. Now we're projecting negative 5% out through this horizon. Significant change there in expectations that drives us to the low double digits. The second driver of the change is there has been some push-out to some of our new product lines, the new adjacency businesses. These are big investments, high innovation level, high complexity level. We have had some delay in that ramp of the new products. That's the other smaller factor. Okay. The other area I wanted to talk about though was operating margin. Operating margin, because we've had this conversation before, it will continue to grow clearly. It's a great business. Operating profits continue to grow.
It will grow slightly slower than the revenue growth, but for some really good reasons, it's the same reasons that I've talked to most of you about before. The first is, as a corporation, with BK's strategic move earlier last year to the virtuous cycle of growth, we call it data center first. As someone that has been in the enterprise business at Intel for 18 years, it's pretty exciting to be first. We're data center first, and that means that each of the functional groups inside of Intel look at their business and their investments and their strategies in the context of making the data center a priority, the number 1 priority, as VK said. That includes being first to launch on a next-generation process technology node, and that's a big deal. The data center, we clearly value performance.
We get paid for our performance. We continue to deliver leadership products out into the industry. That is reflected then in our share of those markets. We're moving now to being first to the new node of process technology on seven nanometers. Data center will go first on seven nanometers, and we're actually going first on what we call the third wave of 10 nanometers as well. There's three waves of 10-nanometer products with incremental performance improvements. We'll be the first on the third wave. You can say, "Hackney, we haven't done this before." Right. The data center's always valued performance. Historically, there were two limiters to us going first, two reasons why the PC always went first on a new process technology. One was volume.
You need a lot of wafers going through the factory in order to get a new process technology up and running. We didn't have the volume back then. We clearly have the volume now. The other was die size. You need a small enough die size to ramp a new process technology, given the defects are naturally higher at that time. The engineering teams, Murty's organization and Sohail's organization, were challenged with figuring out how to get around the die size issue, and there's been a great innovation. It's called EMIB technology. It's a high-speed packaging technology that allows us to cut up the Xeon die and debug just a piece of it at a time. That combination, which is pretty exciting for us, is allowing us now to go first on a next-generation process technology node.
We will say those early wafers on a new process node are very expensive, that reflects back obviously in operating margin. It is an overall win for the data center group and a win for the corporation. Okay? That's operating margin. Oh, I'm sorry. I missed the second piece there. The second piece of operating margin impact is with the adjacencies. We've talked about this before, that as we grow the adjacent product lines, the gross margins of those products are different than the Xeon CPU processors, which are obviously very nice gross margins. As the adjacencies become a greater portion of the total business, that will obviously have an impact on operating margins as well.
The last one is that the obvious statement is as the data center business becomes a greater and greater portion of Intel's overall business, we take on a greater percent of the allocations, the allocation of the corporate expenses. Those three factors then play into operating margin. The result then is historically our operating margins have been between 45%-50%, and over this horizon, our operating margin percent will drop to mid-40s to low 40s. Okay? What is going on with enterprise IT, since that was the big driver of 2016 and also part of our projections on what we're going to see in 2017. These are the folks that buy the boxes, the enterprise IT, CIOs. One thing we do know, obviously, is that the demand for compute continues to grow at a relentless rate.
Year-over-year, it's very predictably a 50% CAGR in demand for compute capacity. However, the way businesses are consuming that compute capacity is clearly changing with the move to public cloud as a viable option. How an IT organization decides where to land those enterprise applications, whether they put it in a public cloud environment or they leave it on-premise in a private cloud deployment, it's not a random event. There's a lot of key decision factors that go into that decision, as listed on the slide. As you can see on the chart on the left, what we've done here is we've taken all of the enterprise IT applications and put them into five general buckets. The yellow is new enterprise applications that were deployed in 2016 on-premise, and blue is new deployments, new enterprise application deployments that were deployed into a public cloud.
What you see is in 2016, there was significant investment in moving collaboration apps, things like email, Google Docs, Office 365, web solutions, a significant move of what we call the low-hanging fruit, out of the on-premise data center and into the public cloud. The use of public cloud for collaboration solutions grew 15%, while the on-premise deployment of those applications declined 21%. You go to the other end of the spectrum, modeling and simulation, which is basically high-performance computing workloads. They have a high affinity to being local. They need the performance, the low latency. What you see there is deployments of high-performance computing on-premise in the private cloud deployment grew 8% in 2016. In 2016, about a third, 35%, of enterprise compute capacity was in the public cloud, and two-thirds remained on-prem.
We believe that over time, there'll be an even distribution of those workloads. We believe it'll settle out at about 50/50. 50% of the enterprise workloads will be on-prem in a private cloud deployment and 50% in a public cloud. Okay? Now looking at the private cloud. Private cloud obviously has had a slow start. That's a pretty obvious statement. We have been investing very targeted in accelerating the adoption of private cloud. The tech companies that serve enterprise segment, as I noted, they have been restructuring to also address the new enterprise market realities of what solutions are needed. We do know what is required to grow enterprise IT and grow private clouds. It is the fact that we've got to have a broad availability of enterprise-class, easy-to-deploy cloud orchestration solutions. That's what's needed.
The really good news, the exciting news is, the market has confirmed that this is the end state. The end state is going to be hybrid cloud. You're going to have some workloads running on a private cloud and some in a public cloud. There have been some very exciting announcements made late last year around offerings to support this, and that's the Amazon VMware announcement that they announced back in November of creating a hybrid compatible public-private cloud environment. As well, Microsoft announced earlier last year that they would be deploying Azure Stack for an on-premise private cloud solution that's fully compatible with Azure cloud. Those are two big announcements that we believe will really help drive the adoption of private cloud and get us to that equilibrium of 50/50. Private cloud deployments have grown at a 25% CAGR over time.
We are looking at 2013 to 2016. Getting those hybrid cloud deployments ramped is clearly critical, we are currently assuming that this, as I said, the server deployments into enterprise will continue to decline at about a -5% CAGR over the planning horizon. Okay? We're not expecting any immediate rebound in adoption of these solutions. Although they are critical, and we're quite excited that they are coming to market this year. Because enterprise server is our legacy business, and because it did decline more than we had predicted in 2016, I wanted to show you the sensitivity analysis that we have done. We expect our future growth to be increasingly buffered against the uncertainty in enterprise IT. That's for two reasons.
One is because more and more of our total business, the total revenue, is driven by comm service providers and cloud service providers, not the enterprise IT folks, and because of the growth beyond CPUs into the adjacent markets. Okay? We have a data center double-digit growth forecast that assumes a 5% decline in enterprise server CPUs. If that decline is worse, at 7% decline, we still have a double-digit growth out through the horizon, and we believe there is a scenario of -1% growth or contraction based on the fact that we do believe those new workloads could be adopted more rapidly by enterprises, things around artificial intelligence, data analytics solutions. We think -7% to -1% is a reasonable range, and we've set the base towards the conservative end of that range at -5%. Okay?
In all cases, we get to double-digit growth. Now shifting to the public cloud service provider market. I'll start in the top right. I showed the move of workloads, enterprise workloads, from on-premise to the public cloud. What I want to highlight here is that it's not all conversion. It's not all cannibalization of a workload that was on-premise into the cloud. New services are being adopted by enterprise IT as they leverage the public cloud, this is TAM expansion. Things like high-performance computing as a service or artificial intelligence solutions for those first-time business units. When you have a very complex IT solution and you're not a very mature IT house, this is an opportunity for you to use a technical solution that you would otherwise not deploy.
On the left, you see that the gap between enterprise ASPs and cloud service provider ASPs, which has been a topic amongst us for a while, that gap continues to narrow. The drivers of that reduction in the difference in ASPs are a couple things. One is the cloud service provider ASPs are up. As you see, 10% up over the past four years. That's because they see value in our high-end products. They're buying up the stack, and they're continuing to adopt custom CPUs to get a very targeted solution against their workload and against their data center. Also, with the shift in our corporate priorities, as BK talked about, we've increased the data center sales force, and it was clear in BK's talk as well. We now have a dedicated data center sales force focused on our customers.
That's an increase of 20% in data center sales headcount, a significant increase in 2016. What we have seen is with direct account coverage of the next wave of the cloud service provider, beyond the Super 7. The next 100 cloud service providers, the ASPs increased 32% in Q4 year-over-year. When we have Intel going in to the cloud service providers, helping them understand how to drive up performance and lower their total cost of operation, we see our ASPs go up as they move up the stack, and we also see our share of wallet, the portion of products that they buy Intel, also go up. This direct data center sales force is having a very clear impact, very positive impact.
The last point to make on this slide is that business cloud services deliver high value, high value to the enterprise customer that's procuring them, and hence they command a price premium compared to your basic consumer cloud service like Twitter or Snapchat. In 2016, business cloud services had an ASP that was 11% higher than consumer cloud services. If you back out the Super 7, who obviously have large volume, and there's the economies in driving pricing. If you take out the Super 7, then business-directed cloud services had a 25% ASP premium versus consumer services. What this means is that we are increasingly indifferent and less affected by where an enterprise IT actually chooses to place their application, whether it be public cloud or private cloud. Next is the comm service provider space, a very exciting space.
We have been talking about winning the network infrastructure over to Intel architecture for at least a decade, and I'm very happy to say that we have crossed the chasm. Adoption is happening. The network is moving from Arm and MIPS and PowerPC and all those other architectures, all of those custom ASICs. It's moving now onto Intel architecture. Moving from fixed function, purpose-built appliances over to Intel-based servers, virtualized and delivering those capabilities. 5G operates as an accelerant to that transformation, because 5G is going to require the convergence of compute and communications and the need to move that compute capability closer to those connected devices. Combining our existing business with the design wins that we solidified in 2016, we are now number one in wireless access. That's a combination of macro base stations and cloud RAN deployments.
You add that up, and we are number one in wireless access, and those design wins that we won in 2016 will ramp in the market over the next four years as 5G ramps. We're very proud of being number one, but if that doesn't impress you maybe this will. 2016 was the first year where we have market segment share leadership in the overall network market. Number one in that $18 billion silicon opportunity. That share gain is ahead of our plan, we are doing very well against that. That is an Intel share statement, that's a combination of the data center business and Dan's business in PSG, previously known as Altera. The combination of our FPGAs and our data center products has gotten us to number one.
We've obviously gotten here because we've made very targeted investments to make sure that the Intel Xeon and Atom CPU product lines run all network functions very well. We've been integrating wireless and network features into those core product lines. The other thing that we talk a lot about amongst us folks is the ASP. I wanted to show you, on the right is our weighted ASP by processor family. Okay? The lion's share of that network revenue comes from Xeon. As you can see, the Xeon ASPs are growing very nicely. Year-over-year, 2015 to 2016, it was up 18%, our Xeon ASPs. That's because customers are buying up the stack as the workloads, the network workloads, are converging onto Intel.
All those disparate pieces of silicon, custom ASICs, and MIPS products, taking all of those silicon compute capabilities and converging it onto a single Xeon. With the SoC product line at the bottom, we have, with that product line, the ability to serve the lower power, lower performance, lower cost requirements of the network infrastructure space. The SoC uptick that you see in the ASP in 2016 is a result of the introduction of our very first Xeon-based SoC. Historically, those were Atom-based. We have our new Xeon-based SoC, based on Broadwell, that we launched. That uptick reinforces the fact that performance in this space is truly valued, and the customers are obviously willing to pay for that performance.
This is a huge competitive differentiator for Intel in our ability to address the network infrastructure customer's complete portfolio of products from top to bottom, from the core network to the edge, out to the access layer, all running on one architecture, gives them one tool chain, one broad ecosystem to pull from, and that commitment of a continuous improvement in performance generation after generation. Okay, AI. As mentioned, artificial intelligence is the fastest-growing workload in the data center. It is still nascent. To put it in perspective, if you look at all the servers shipped in 2016, 7% of them were running AI workloads. Okay? Majority of that is classical machine learning algorithms. Credit card fraud detection or stock trading or recommendation engines, all of those artificial intelligence predictive solutions is classic machine learning.
40% of it was deep learning, which is predominantly used today for image processing and language processing. For classic machine learning, we just build upon the strength that we have in traditional analytics workloads. We have the right technology to address the space, and we have 97% share of the servers running Intel architecture. For deep learning, and I'm talking both training and inference, 91% of the servers today are Intel architecture only, two-socket Xeon servers or Xeon 5. However, we recognize very clearly this is the space where GPGPUs have value. GPGPUs were not explicitly designed for deep learning, but the math happens to be the same between graphics and deep learning. Low-precision, dense matrix multiply is the math for graphics, and it's also the math for neural networks in deep learning algorithms. By units, as you can see here, the number is relatively small.
It's about 20,000-30,000 servers out of a total of 9.5 million that we shipped last year. These are expensive systems, right? These have expensive GPGPU cards in them. Because we are known for delivering performance at scale, although GPGPUs are there today, our customers are asking us to deliver a solution against this workload. They all want to have a homogeneous environment, having Intel end-to-end across the entire workload, as well as the fact that GPGPUs don't scale. With the Intel architecture, you can scale out to tens of thousands or hundreds of thousands of servers to serve those very large data sets, very large analytics problems. It's clearly our objective that we are going to run all of AI best on IA, and we say that, "AI on IA." This is fun. AI on IA. See? He likes it.
It's catchy. We do know how to win architectural battles. We've done this many times over the years. I've been here now 32 years. We've been winning architectural battles many, many times over our history. We have addressed GPGPUs and the competitive threat before. We addressed it in supercomputing and high-performance computing. We addressed it with Xeon Phi. As an example, with the latest top 500 list that was published in November, you can see that we have one share away from NVIDIA with the launch of our Xeon Phi processor. We now have 84% of the total flops of the top 500 supercomputer list. There are many different approaches to artificial intelligence, as I talked about on the prior slide, they do have one thing in common, which is scale. They want scale.
The more data you can compute, the more accurate the model is going to be, the more compute capacity you can apply to the problem, the faster you will get the result, the faster the model will train. Okay? To win in AI, our approaches will provide the full portfolio of products, as BK talked about, will span all artificial intelligence workload implementations. One architecture to serve the many algorithms from training to scoring, from development of the model to deployment of the model, from general purpose algorithms to highly targeted deep learning algorithms. We have the full portfolio of products. We call it the Nervana portfolio of products. It's Xeon. It's a new version of Xeon Phi that is coming out this year that's targeted for deep learning. It'll deliver a 4x improvement in deep learning performance gen over gen.
We also have Xeon + FPGA for those customized algorithms. We also have the Lake Crest silicon that is coming out this year that came through the acquisition of Nervana, this is silicon that was developed explicitly for deep learning workloads. We will integrate that technology into Xeon. You'll have the best-in-class Xeon processor with that very targeted acceleration for deep learning. No one else can deliver this breadth of integrated solutions end to end. It is one of our core capabilities and core opportunities, we are very passionate and confident that we will continue to win in the world of artificial intelligence as it grows. I know that it is easy to think about us as having 98% share, 98-plus % share of a server CPU market. As BK showed you earlier, that's not how we see ourselves.
We see ourselves as operating in what was a $46 billion market last year, growing to $65 billion in 2021. We have just a third of that market today. There's huge opportunity for the data center business to grow. We're addressing that growth with these new product lines. To hit on them quickly to give you an idea of the scope, we launched Omni-Path in 2016, early in the year. This is a high-speed fabric, high-performance fabric for high-performance computing. In the first year, we captured 42% of the 100G HPC market, we're growing into a $1.1 billion SAM by 2021. This opportunity for us is not just the fabric controller and switch silicon, we are also able to integrate it into our Xeon and Xeon Phi processors for additional performance, lower power, and better cost structure.
We launched silicon photonics mid last year. That is now ramping nicely into volume. With that, we're addressing the 100G market. As you can see, it is small. The 100G market is just now being adopted by the market. Hyperscale data centers are the first to adopt. As that adoption occurs, we then are growing into a $1.8 billion market in 2021, comprised of both 100G and then moving to 400G photonic interconnect. We have a clear technology advantage in this space in silicon photonics, thanks to our integrated design and manufacturing capability, integrating the laser onto the silicon IC. That advantage just continues to grow over time as data centers adopt higher and higher speed optics, and as optics becomes the predominant interconnect versus cables in the data center. 3D XPoint DIMMs, this is a huge opportunity for us.
The technology, as you heard, will be coming into the market in 2017. Then will ramp into 2018 in a system memory DIMM form factor. This is a technology that allows us to displace DRAM, a market we clearly don't participate in today, an $8 billion market opportunity as we go out in time. With 3D XPoint, it is a very compelling value proposition to the end user because we can deliver twice the memory footprint capacity in a system at a 40% lower cost structure. Our expectations in that horizon are pretty conservative. We're assuming single-digit server attach rate of 3D XPoint memory out through the horizon. We're assuming servers that obviously require a large memory footprint, so servers that are deploying 512, half a terabyte of memory or above. It's a constrained market we're going after, a huge opportunity, an $8 billion SAM.
Finally is Rack Scale Design. As BK mentioned, this is a system architecture that we defined. We launched it at IDF last year. Along with the industry, we had all the OEM partners deploying Rack Scale Design architecture. We, again, are uniquely positioned to deliver the technology into Rack Scale Design due to the high level of integration that's going to be required in order to continue to deliver the step function improvement in ever-increasing performance at ever lower cost of operation. We can drive an integration across the CPUs, the fabric, Ethernet, solid state drives, FPGAs, memory, that entire portfolio, we can deliver a better solution than anyone else on the planet. To summarize, the growth drivers remain intact. The enterprise market is weak, but it is not the swing factor for our future growth. We are a data center silicon business.
We are a $17 billion business today in a market that's $46 billion large. We have an unrivaled set of assets and a very unique ability to integrate and deliver greater value through that integration. We are now number one market segment share in servers, which is not a surprise, in storage, which shouldn't be a surprise. Now number one in network as well. We're also number one in Ethernet connectivity. We are delivering high single-digit growth in 2017, but continue to project out a double-digit CAGR out through the horizon. Very excited about this business. Thank you for your time.
Mark Henninger.
Yay.
All right. Welcome back, everyone. Thank you for rejoining us. I'd like to welcome up on the stage Rob Crooke, the general manager of our Non-Volatile Memory Solutions Group. Thank you, Mark.
Well, thank you, Mark, good morning to you all. I am Rob, the memory guy, here to talk to you for the next few minutes about what we have going on in our memory business and our strategy for that. I would like to start by re-sharing Brian's chart from earlier that shows what a tremendous opportunity we have in front of us and how we are so uniquely positioned as a memory business at this particular moment in time. We're an adjacent disruptive growth business with leading-edge technology and manufacturing capabilities that has a strong connection to our core platforms, we happen to be in a very fast-growing market segment that we really have participated in since its inception. Tremendous opportunity at this moment in time to be the memory guy for Intel.
Now, as BK stated, I won't go through how the data is growing, it is growing tremendously. We see it every day in our business as we interact with customers and users of the technology. Data is growing tremendously, to gain benefit from that, you have to capture, analyze that data, then operationalize it to get the results. Whether you're on the edge or whether you're in the cloud, to capture and analyze that data, you need big memory and fast storage. If we could put all of that data in the datasphere onto the CPU, we would. It would unleash the CPU with unlimited scaling and amazing new insights in its analytic power. As we move data away from the CPU, it causes delay and reduces the benefit of all that data.
Unfortunately, technology and economics keep it apart. The fact that this data is growing at such a rate makes this challenge harder. So does also the fact that the CPU is getting faster. The same delay to data tomorrow is actually greater than it is today. What do we do? We break up all that data into different tiers where we trade off cost versus delay. We have hot data as close to the CPU as we possibly can. That's for workloads that require real-time and constant access to the data. Think real-time analytics, transaction processing, machine learning, artificial intelligence. All want to be as close to the CPU with as much data as we possibly can get close to the CPU, as Diane said earlier.
Warm data workloads are big and valuable. We want to get them done as soon as possible, but they don't need to be done in real time. Finally, we have to put all of the data somewhere, right? That's out on the edge in cold data. Those are things like the incredibly exciting pictures of Mark's dinner last night that he's sharing on social media that apparently he wants to save forever and show us what a fine diner he is. He wants to be able to reach back and see those. We do have to store all that somewhere. The industry is going to continue to need innovation on all levels of that datasphere.
The hard disk guys have turned their R&D towards the edge, the cold tier, to get the best cost per bit, pushing outward in the datasphere to capture all of that data from Mark's dining. They're pushing outwards. They're doing that and trading off performance in order to get that. They're focused on cost per bit and to get as much in there as possible. The inner part of the datasphere for DRAM, we do still need continued innovation in the lowest latency for things like graphics frame buffers to continue to get the kind of performance we need in graphics. With much higher capacity and performance for hot and warm data in between those two tiers. When we look at that combined TAM for all of those markets in 2017, this is over $100 billion available market for us.
We've been focused on the SSD portion of this market that's gone from near zero in 2008 to $20 billion this year within the NVM TAM. Clearly, there was a gap in the hierarchy in between memory, DRAM, and storage that created that $20 billion market. We're really excited about that market. That market is growing at 25% CAGR between 2010 and 2020. That market alone is really exciting. We're excited about even more than that because we think the opportunity is greater than that. We think we can push non-volatile memory outward into both segments with the new technologies we're inventing. We are working on two new technologies that help drive non-volatile memory outward towards the outer edge and also inward on that data sphere.
First, 3D NAND is all about density and lower cost, pushing outward to capture more of the warm data on media that's 1,000 times faster than the next ring. Nonvolatile NAND-based SSDs are 1,000 times faster than a hard disk. That enables us to get more of the data closer to the CPU so we can get more value from it. Optane technology is about solidifying our position in hot data for real-time processing while pushing inward with memory media that's 10 times the density of DRAM, pulling more data onto faster storage and bigger memory. Now, when we look at our differentiation in 3D NAND, it has a number of different aspects of it. First, let me say, 3D NAND is a key transition for everyone in the market.
All players are driving to 3D NAND memory, and it helps us all extend Moore's Law scaling into the third dimension. It's visually obvious how important that is. Historically, we've been in two dimensions. Think of it as a checkerboard, where the checkerboard squares, we've been trying to get them smaller and smaller over time, and we've gotten to the minimalist dimensions there. Now we're moving into that third dimension. It gives us a cube of space in which to put memory cells. It's a big decision as we go down there. We're making technology choices focused on our target markets. Our focus, as you can see from my data sphere conversation, was about density and cost and getting further out into the data sphere. Our choices have resulted in leadership here. Let me show you a couple of illustrations of that.
On the left side of the illustration above, you'll see our competitor's technology. What it has, you'll see the trenches in the middle of the array field. Those trenches in the middle of the array field are necessary for the replacement gate technology they've decided to go on. The architecture that they've chosen does not allow them to put the support circuitry or the periphery circuitry underneath the array. Think of that as like you've got a skyscraper and you haven't made room for the parking spaces below the skyscraper, so you have to put the parking lot around the skyscraper. That's what you see in the illustration on the left, and it has consequences, right? The fact that we don't need the replacement gate enables us to get the smallest memory cell in the industry.
The architecture that we've chosen enables us to get a very small periphery around it, and it has consequences to die size. What you see on the chart on your right is the consequences of that. Those choices result in what we call areal density leadership. Areal density is simply just the number of memory cells that are on a die divided by the die area of a given design. What it means is if you have better areal density, you got more memory cells per wafer, which is really the first order impact on cost. We use it to simplify the comparison of technologies. If someone's got two bits per cell, we still want to be able to compare the density of that, for example, in the early 3D NAND.
What you see on this chart is the red dots are measured die area from the competitor's technology, the blue dot is our 32-tier, and then the blue oval is a couple of designs for us that we've measured on our 64-tier technology. The red oval is some recently published papers, although we haven't seen the die, from some of our competitors. You compare, if you notice the top of the oval is the 512 gigabit design, the bottom of the oval is both 256 gigabit designs. You'll see a couple of things going on here. One is that our density across time is greater than our competitor's. It's over 15% in the 32-tier technology. You'll see that it has potential to expand.
That die area, and hence first-order impact on cost, has the opportunity to be greater and expand over time because of those technology choices that I mentioned a moment ago. Our 3D NAND is also architected for rapid scaling and yield. You saw the fact that we had the factory up, Fab 2 in Singapore, and then we brought up Fab 60 up very quickly and got high yield on that immediately. That is not by mistake. That is because the technology is architected for rapid yield. It's also architected for rapid transitions. What you see is while our competitors are doing 24 and then 32, then 48, then 64-tier, we're actually going to leapfrog from 32-tier directly to 64-tier in just five quarters. We're actually accelerating Moore's Law with these technology decisions.
Actually, this presentation right here is being done on our 64-tier SSD that we plan to have in production for revenue in Q3 of this year. We can scale quickly because we're pushing the tooling advancement independent of the integration advancement. What we do in the first or second generation of this, we push the tooling hard to get it to 32-tier technology, then we double up with an integration strategy, then we can push the tooling again on the next generation of technology. We've made these architectural choices in our technology because of our focus on the solid-state drive market and getting high density. Now, Optane technology, of course, is pushing for high performance in the data sphere, pushing inward to solidify the position in hot data, and push inward towards increasing density for memory.
We recently just qualified our first data center SSD, and we're in an early ship program for that Optane SSD with our leading customers. As you know, Optane is a new class of memory. It's completely different from 3D NAND. Well, you hear the phrase 3D XPoint and 3D NAND, they both begin with 3D. That's where the similarity ends. They're actually completely different in the way the physics works and the fundamentals of the technology itself. The magic in Optane technology is the fact that the switch and the memory cell itself are matched, and they both can be manipulated with two terminals instead of three, like a transistor needs. That enables us to put it on a CrossPoint architecture. That CrossPoint architecture enables us to get fast access to the memory cell itself.
It also allows us to get think of it as byte-writeable, direct-writeable memory cells that allows us to use it as memory in addition to using it as storage. We have multiple generations of Optane going on under development right now. We see a very clear path to the first three generations of the technology and beyond. I'd like to talk briefly about what this means at the platform level of all these non-volatile memory technologies. We started with replacing storage with NAND technology. This is an older terabyte hard disk. When we're doing the first NAND-based SSDs, we have to fit in the hole in the box that existed from the thing before us. We use the same interfaces, same mechanicals, and this would be a terabyte SSD from a year ago.
You can see that we're getting into smaller form factors, but we're still using the legacy interfaces, and we're still a square, and you're a square because there's a circular thing inside of it that you're drawing a box around. When we are in client space, we got penetration high enough that we could then begin to re-optimize the platform for non-volatile memory. What you see here is a terabyte SSD for this year technology. Here what we've done is we're able to re-optimize this for small form factor PCs to enable us to get thinner, lighter, more robust platforms with the same performance and without compromising on capacity. Then as we go forward, this next year will be a terabyte SSD. We continue to re-optimize the platform, enable continued innovation at the platform level without compromising on capacity. Thank you.
If we take a re-look at this in terms of the data center, we can do the same kind of thing. What I've got here for you is a model of a data center rack with a petabyte of storage. This rack has in it 502 terabyte SSDs, and you can feel the weight of that on the stage. You can feel the power coming off of that thing, the electricity, the heat, the volume of the fans making noise. You can almost feel the angular momentum of the spinning disks right inside that thing. We said, "Hey, what if we thought differently about solid-state storage?
What if we took that same viewpoint in the way we re-optimize storage for the data center as we did for client?" Actually what we were able to do at that point in time, as you look at the chart on your left, is we turned that same petabyte with all of those data center hard disks into something that is in 1U. That right there is what you see in the cartoon on or the illustration on there. That is what we've done is we've re-optimized what an SSD looks like. We have done something that's got better cooling, better form factor, more serviceability. It allows us to get much higher density in there and take advantage of 3D NAND. Now we can also, when we've got this interface, we now look at the platform differently.
Now we take Diane Bryant's platform, and we add more direct connect PCIe ports. We add better, robust, direct support for hot swap virtualization, new features in the platform that take advantage of this new memory. Now it's exciting. That bigger, that non-volatile memory faster storage is driving new experiences in the client and data center efficiency that drive demand for the technology, and it is indeed driving more demand. The current market forecast is 300 exabytes of demand per year of non-volatile memory storage by 2020. That's a lot. That's a lot of anything. Gigabytes are big. We have a disciplined investment strategy that's aligned to market growth, and we have a three-pronged strategy for that disciplined investment strategy to keep our technologies on the leading edge.
In 2017, we'll invest in 3D NAND in Fab 68 to ramp that site to meet our customers' needs and drive towards more cost-effective scale. We'll have a sustained collaboration with Micron in Fab 10, and the result of those two things will give us more than 60% in gigabyte growth year-over-year. That helps us deliver on fast growth in our business, and we'll have over 90% of that supply will be 3D NAND, which will enable us to get our cost structure down much lower. We expect that even with these big investments, the core NAND business will return to profitability during 2017 and improve from there as we ramp to a higher percentage of our business being on 3D NAND and then second generation of 3D NAND over time.
2017 will remain an investment year for Optane technology as we're investing in Optane product R&D, and the factory ramp. The early ramp of the volume, that's a lot of the questions I'm getting, it's a new technology, and so people are excited about it, but it's uncertain how fast that will ramp. Our current expectation is for 2017, it will probably be less than 5% of our total storage revenue. Overall, we expect these two combined businesses to return to positive operating margin by the end of 2017 when you sum them together. What you can see for Intel's memory strategy is that we're an adjacent disruptive growth business with leading-edge technology and manufacturing capabilities, with strong connections to our core platforms in this incredibly fast-growing market segment that we participated in since its inception. Great time to be a memory guy at Intel.
With that, I would like to invite my good friend Doug Davis up here to talk to you guys about Internet of Things.
Thanks, Brian. Great segue. Thanks for the opportunity to spend a few minutes with you this morning. I wanted to start out commenting on something that Brian mentioned earlier in how we think about the Internet of Things. He even kind of indicated it can be lots of different things to lots of different people. Over the past three years, as I've been doing keynotes and traveling around the world, we've really described the Internet of Things in a very consistent way. We always say that the value comes from connecting things through the network to the data center or cloud, and then using data analytics to extract information from all the data that these things are generating. The value for Intel really comes from the computing that's required to extract that information from the analytics that's possible.
Of course, that happens primarily in the data center, we see it starting to move more and more to the network or the edge of the network and increasingly out into the things as well. That's important as we think about how this business is evolving and growing. Sometimes the question comes up is the IoT really just a relabeling of the embedded business that we've been in for years? As I said earlier, the Internet of Things is about connecting things and enabling open platforms, and then that enables you to deliver new products and new services with the information that you can derive as an owner of one of those things. You can see by the graph, this transition to connected things has been actually happening for years.
That's the distinction that's really the baseline that we think about when we talk about the IoT. The other thing that's important is we see real implementations happening, delivering real value. According to Gartner, IoT spending grew 18% in 2016. That was primarily led by manufacturing, industrial, transportation, utilities, right? That infrastructure kinds of capabilities where they're seeing real value because they have lots of things generating lots of data. They also said that at the end of the year, there were 6.4 billion connected things, and that was an increase of 30% just over 2015. We see this transformation really happening. Again, you saw in BK's slide earlier, we have a $30 billion opportunity when we think about the broad set of applications that make up the Internet of Things. I'm going to spend a few minutes.
I'm going to walk through the strategy, talk about some of these targeted sectors, you've heard several of them this morning, and really put it together for you. I wanted to start out and just summarize the business results from the past year and think about how this business has evolved over the past several years. If you look at the Internet of Things Group, it's grown about 95% over the past six years, while the aggregate of competitors in this space grew about 45%. Those companies include Qualcomm, TI, AMD, Microchip, NVIDIA, companies that are building computing that go into things. We define the SAMs in this case as 32-bit and above connected computing. To go compare that, the reported SAMs grew at a compound annual growth rate of about 12% 2013 to 2016.
Again, we're focused on this high-value 32-bit and above computing space. If you look at the other computing kinds of applications, that TAM contracted about 5% CAGR over the same period of time. The emphasis here is the part of the market that we're targeting is growing much faster than the areas that we're not focused on. That's important in terms of how we're oriented for growth going forward. Now, over the past three years, we've also defined that the Internet of Things is really moving through 3 stages, all kind of happening simultaneously depending on the kinds of things and how they're being deployed. We start out talking about connecting the unconnected. These are things that have integrated computing but aren't connected to each other or back to the data center or cloud, thus the efforts that we've been investing in around gateways.
In the second phase, as developers are creating new things, we've really been working with them and encouraging them to build in the computing and communications capacity to be able to get those things connected, as well as the ability to support data analytics. Finally, once you've implemented an architecture that gets the things connected to the network and back to the data center or cloud, you've enabled the ability for smart and connected things and Internet of Things implementations, you've also now set them up over time to become software defined and autonomous. Let's dig a little bit deeper into the business. To start, I want to give you some insight into the four verticals that we're focused on, you heard us talk about those earlier, and talk a little bit about Wind River software, which is also part of this business.
As you also heard, we recently made a change in our structure to break out automated driving into a separate sector, I'll talk about that in more detail in a few minutes. Tom Lantzsch was hired, brought into Intel. We're very excited to have him here. I think Tom is sitting here front and center, so if you haven't met Tom yet, take a few minutes to meet him. He's going to continue to run the Internet of Things Group while I go manage the automated driving efforts for the company. Now, in 2016, we were very deliberate in our vision, our investments, the platforms we're creating, strategic partnerships to ensure that we could create the market segment share and continue to drive growth as we go forward. Our vision is aligned where these end users want to go.
They want an industry based on software-defined open platforms. They want open standards, the ability to have control systems that are not proprietary and monolithic as they've been in the past. They want the ability to have that flexible infrastructure to derive the value from these things. The silicon portfolio that we have to apply to these market verticals is the best that we've had in this space for many years. That's pretty exciting for us. While the overall IoT silicon TAM is expected to grow about 12% CAGR in the future, we expect to grow faster than the industry. Within each of these verticals, there are high growth opportunities. If you look at retail, we see the growth in things like digital signage everywhere. That's one of those areas.
In the transportation space, infotainment is one of our fast-growing areas, as there are more and more interactive screens that are being integrated into cars. This is really important because the growth in that business also gives us the foundation and the understanding and credibility in the automotive space from which we'll drive our growth in the autonomous driving area. In the manufacturing space, there's a shift happening to more software-defined infrastructure, and things like energy expansion and alternative sources of energy are helping fuel fast growth as well. In the last area of this markets and channels organization, we also have some areas like video, we talked about it's in things like digital security and surveillance. That's a fast-growing market for us.
Finally, Wind River has opportunities in automotive, in networking, in things like device management as some of the opportunities that they're focused on. In order to capitalize on some of these opportunities, we've been investing for the future. The slide is admittedly busy, but you can see in the upper left, there are a broad range of technologies that have to come into play to create these kinds of solutions. I've always said there's no single company that alone can go off and create an Internet of Things type of solution. It takes consortiums. We've helped found the types of consortiums that are noted here and build those strategic partnerships to help achieve scale and drive standards where it's necessary. We have a broad range of assets at Intel.
We've changed the conversations we're having with customers and really talking to them about how they create these end-to-end solutions and get those things connected back to the data center. We have an unparalleled breadth of capabilities compared to anyone else in the industry, and that's even despite the consolidation that we've seen happening over the past couple of years. Let's dig a little bit deeper into each of these verticals, and I'm going to start with the retail space. This is a strong core business for us, but also one of the areas where we're seeing nice growth. It's really widely recognized in the retail industry that in brick and mortar, as Brian was talking about earlier, inventory inaccuracy costs the industry about $1 trillion per year.
We're developing technologies like our responsive retail platform that integrates RFID, but also many other sensors that are possible to enable the retailer to understand what's happening within the store and to be able to manage that environment. According to marketsandmarkets.com, retailers are also recognizing the benefit that the Internet of Things can deliver to them. They can create greater operational efficiencies, plus create better experiences for the customers that walk into their stores. We see digital signage, interactive kiosks, point of sale, and this responsive retail platform, all as areas that are driving fast growth. As a result, the TAMs that we have in front of us out in the year 2021 is about $7 billion in this particular sector. Next, I want to shift and take a look at the industrial space.
According to McKinsey, these segments represent the largest IoT opportunity and will be responsible for about half of the annual IoT value over the next decade. Obviously, an important area. We already have a significant silicon business in this area around industrial PCs and high-end PLC or programmable logic controllers. At the same time, though, we have pretty limited traction in the low-end applications like motor control or low-end PLCs because those applications require things like real-time processing. They need deterministic performance. We now are moving towards the consolidation of some of those kinds of technology, real-time determinism, combined with our Intel architecture platforms, that gives us the ability to now deploy more consolidated solutions that are software-defined into this industrial space. That opens up a big part of the opportunity for us.
This space of smart and connected in the industrial world represents about an $8 billion TAM by the year 2021. I mentioned earlier, this is markets and channels acceleration division. Video is one of the largest and fastest-growing areas. In the past, what we've seen in the space is really analog cameras feeding lots of data back to something called a network video recorder, where all that video gets stored and the data analytics are processed on the video feeds. Those then get deployed into airports, into cities, into buildings. The challenge becomes now as more intelligence is needed in these systems and those cameras become higher and higher resolution, feeding all that data back to the network becomes pretty cumbersome and expensive. The industry is starting to move that data analytics further out to the edge and out into the camera.
We can now help the developers move those Intel architecture-based video analytics algorithms directly into the camera, where the camera can make a lot of those decisions. These smart cameras and what we call video gateways really open up a high growth opportunity as we go forward, and that equates to about a $6 billion TAM out in the year 2021. Finally, let's talk a little bit about autonomous vehicles. You've heard that mentioned a lot today, and clearly, they'll become the most visible artificial intelligence application that we'll interact with on a day-to-day basis in the coming 5-10 years. However, they really are also one of the most dependent examples upon having the vehicle connect back to the data center or cloud to create that artificial intelligence type of implementation.
Intel really has the differentiating technologies to build the solutions from car to connectivity to the cloud. Let me break it down. Let's start with the car, what's happening in the car is you have all these sensors, camera, lidar, radar, that they're creating a picture of the environment around the car real-time. We need to be able to pull all that data together and fuse it. We're combining it all together, our FPGAs do that really well. We want to take that data, that environmental view, and combine it with a high-definition map so the car knows where it is at any point in time.
The computing in the car that the brain needs to take all of that, along with the machine learning and deep learning model that's built to determine its trajectory, to decide where the car is going to be headed next. That kind of brainpower demands a high level of computing that our microprocessors and our FPGAs can deliver. If we look at what's available today in the car from an Intel silicon BOM standpoint, it's $100-$200. It's what we do today around infotainment, instrument cluster, rear seat entertainment systems, and other computing functions. If we look forward a little further out into 2025, our opportunity grows significantly as the requirements grow to be able to provide autonomous type of features.
In fact, in a L4 or level 4 or level 5 autonomous vehicle, it'll require two high-end Xeon processors, two FPGAs, memory and storage, the communications for the car to be able to connect. As a result of that, we see about a 10-15x increase in the addressable BOM, which equates to about a $10 billion silicon TAMs out in 2025. The opportunity is even greater when we look beyond just the silicon TAMs. In fact, in other parts of our Internet of Things business, we're successfully deploying new capabilities and business models that enable us to get paid for more than just our silicon. If we think about all the elements necessary to deliver comprehensive artificial intelligence solutions beyond just the computing that I referred to, this creates a much, much greater opportunity.
When we look out into 2025, again, the TAMs for silicon software and services is much bigger. It's more like a $65 billion opportunity. We intend to explore the range of technologies and platform-level solutions that are possible as we deliver these capabilities for our customers. We're also developing all the building blocks that developers need to create these kinds of solutions. At CES, we announced the Intel GO brand to represent the activities, the products, the technologies that we have to support autonomous driving. As part of that announcement, we talked about the capability and the offering for a 5G development platform that allows developers to begin developing for 5G implementations in the car based on FPGA, so it can adapt and evolve as the standards become more defined over time.
We also announced two development platforms that support computing in the car, one based on our Atom technology, the other based on our Xeon technology. I happen to have one of them here. You can see this is our Atom-based platform. It's got all of the ruggedness and connectors and form factor that the developers need. It's about the size of that six-disc CD changer you used to have in your car, right? We'll continue to shrink this over time as Moore's Law enables us to do so. The third element is we announced an automotive development kit to be able to tie together the development environment developers need in the car all the way back to the data center or cloud.
Finally, we're working very closely with the Data Center Group to build the automated driving private cloud platform capabilities that the OEMs will need to be able to do development and then ultimately to get cars out on the road. We're doing this in such a way that it supports the level of artificial intelligence automation that they need for giving them the scalability that'll be necessary to start to put the vast number of cars that will be deployed over the years. All of these are becoming available this year to support that car to connectivity to cloud capabilities. If you didn't get a chance during the first break, during lunch, please walk out in front of the building. We have one of the cars that we've been working with from a development platform standpoint.
Take a look at it, get a chance to talk to the folks that are out there. The other thing that I thought maybe would be a departure from kind of the normal presentation is, Brian did a keynote at the L.A. Auto Show. We put together a video to kind of summarize this. We thought it might be fun to show you that video.
For more than a century, the evolution of the automobile has stirred our imaginations and ignited our passions. Never in that long history has the industry faced a more important moment of transformation, a transformation that promises to bring a whole new meaning to the phrase, "I wonder what this bad boy's got under the hood." The car of tomorrow will essentially be a data center on wheels, capturing and analyzing terabytes of data collected through hundreds of sensors and put to use by powerful in-vehicle processors. Of course, the technology that will enable fully automated vehicles won't be confined to just the vehicle itself. It will require a diverse array of flexible, sophisticated, and fully integrated solutions, from bumper to bumper inside the car, to end to end of the entire automotive ecosystem outside the car.
The result will be a technological wonder on wheels that leverages the intelligent use of data to enable exciting new innovations, make driving safer, and completely redefines the concept of high-performance vehicles.
I hope that was a nice summary for you. I talked about the three phases of the Internet of Things, and what's really interesting is to be able to get these things connected and be able to analyze the data that they generate. What obviously is important for us as well is that we see as that happens, we generate the need for more and more data analytics and more and more server capacity. We thought we'd put together a few examples, and we have lots of these around the number of endpoint things that can drive the need for an incremental server across all the different market sectors that we're focused on.
To pull it all together, Intel is well-positioned to continue to deliver rapid growth around the Internet of Things and in autonomous vehicles based on the breadth of assets that we have to offer. We have a disciplined focus on key verticals. As these devices and the Internet of Things become smart, they naturally lead to the capability to also deliver artificial intelligence. Of course, that's going to demand more aggregation, storage, and analytics of data. This is a highly profitable and fast-growing business. We delivered $2.6 billion in 2016. We grew 15% year-over-year. In front of us, we have a $32 billion silicon SAM in the year 2021. I know I'm very excited about this business. I'm very excited about what we're doing in autonomous driving, and I look forward to talking to you a bit later.
I now have the distinct privilege to be able to introduce Intel's new Chief Financial Officer, Bob Swan, to come up on stage. Bob.
Thank you, Doug.
You bet.
Appreciate it. Distinct pleasure. How cool is that? Good morning. Thanks for being here. For me, I'd probably first like to just thank Brian for the humble introduction for me. It's great to be here. It's a fascinating opportunity. Maybe as cool as being the memory guy, it's a pretty cool time to be the CFO here at Intel, so I'm really honored. I'm also a little bit intimidated because I have the challenge of filling the relatively big shoes of Andy and Stacy in terms of prior CFOs, so it's a little bit daunting. Even more than that, I've spent the last 10 years in consumer internet world. I haven't worn a jacket for more than two hours for 10 years. The idea of being up here in a jacket, having worn it for four hours already is kind of a little bit constraining for me.
I actually thought before I came up to just take the jacket off and throw it over my chair, I reflected back on a prior role about 13 years ago. I was the CFO of a company called EDS back in Plano, Texas. Some of you may have known the company at the time. I had been on board for about two months. We had our analyst day at the time, and it was a fairly formal place. I had my jacket on. Before I went up on stage, I took my jacket off, threw it on the chair, and went up to speak. The next day, the headlines of The Dallas Morning News was, "Blue Collar CFO Arrives on Plano Campus." My wife found that to be a little bit embarrassing, I have to be a little bit constrained for the next several minutes.
Look, Brian said, I want to provide a little bit of a recap, and then put what you've heard in the context of a little bit of a financial expression, if I could. So far, just quickly, you've heard about Intel as an industry leader, poised for growth as the flood of data coming from smart and connected devices fuels the cloud. You've heard from the general managers about how they are protecting their core businesses, but also investing to capitalize on this expanding TAM that we see in front of us. How these industry trends of more and more data are going to require more and more compute, more and more data analytics, and more and more storage, and more and more memory, and they want that memory and data, and they want it now.
We're in a very unique position, we believe, to capitalize on these opportunities in front of us. As Murthy said, it kind of starts with three fundamental cores that have been part of Intel forever. The IDM advantage, Moore's Law, and world-class products. The IDM advantage, simply, this won't be nearly as eloquent, Murthy. Simply, process technology, product architecture to deliver higher performance and scale and ramp products faster than anybody else on the planet. That's the value of the IDM advantage for us. For the less technically inclined, I love the artisan story. The artisan and the-
The farmer and the baker.
The farmer, the baker. I understand that. I think for this story, this for me, is about market share. The IDM advantage is about market share. Moore's Law, continuing to lead three years ahead of others on Moore's Law allows us to get to next node quicker. It drives performance and lower costs. For us, this means value capture and really attractive gross margins. Third, to leverage the IDM advantage and Moore's Law in this extensive product portfolio. For me, that means return on capital. These are the core competitive advantages that we've benefited from for a long time. This is the set of businesses that leverage those core capabilities to expand and to grow. You've heard about the role that CCG has played over the years.
Great products for its clients, massive scale for the fabs, and very high gross margins and cash flows. CCG, this will be a growth vehicle for us for years to come. Great core competencies, expanding TAM. For Doug, with the IoT business, this business simply leverages the core technologies that have been built for the data center and the client business and takes those core technologies into new vertical industries that we otherwise may not reach. Automotive, industrial, retail. Big verticals that we otherwise may not reach. Memory and PSG, these are the accelerant technologies that'll drive faster, more rapid adoption in the virtuous cycle of growth. Those fundamental three strategic advantages have allowed us to build really strong market positions across these five businesses. Collectively, these businesses leverage core IP.
They leverage a manufacturing footprint to drive higher and higher return on capital for you, our investors. We are going through a fairly significant transformation, as you know, as a company. From a PC-centric company to a company that powers the clouds and billions of smart and connected devices. In that transformation over the last couple years, what it has meant is the client business, as Murthy explained, the PC TAM is down during that two-year timeframe 15%. 15% volume down. Through new products, new adjacencies, better core mix and market segmentation, they've been able to manage this business with only 3% volume down. More importantly, dramatically improved the profitability of the business. What I'm going to call the growth businesses, simply put, this is the data center business, the memory business, the IoT business, and the PSG business.
During that same two-year timeframe, this collection of businesses that are taking advantage of the increased compute demands have grown by 13% a year. We've continued to invest in these businesses to take advantage of what we see in the future. In order to capitalize on this dramatic transformation, we have shifted how we spend money in a fairly significant way. Reducing our cost base, reallocating our costs into the higher growth opportunities, investing in what we characterize as big bets that can generate attractive returns down the road, and repositioning. Repositioning is the pending exit of our minority position in McAfee and the addition of Dan's business, Altera or PSG into the portfolio, which we think is better positioned to capitalize on the dynamics that you've heard about during the course of the day. From a reduction standpoint, we announced a restructuring last year, as you know.
We took out $700 million of cost in 2016. We expect to get another $400 million reduction from the actions that we've taken this year. We have reduced the cost structure of the business. At the same time, we reduced CCG even more. We've taken those lower costs from CCG to redeploy into the higher growth businesses. We are making investments for the future. You've heard about them this morning. We are making investments in ADAS. I'll talk more about that. We are making investments in this great memory business because it's great to be the memory guy, and we are making investments in ADAS and 5G. Over the course of this two-year timeframe, we think our cost structure today is much better positioned to take advantages of the opportunities we see in front of us.
We also have our direct spending as a percentage of revenue down two points during this timeframe. More importantly, it's how we're spending the money to take advantage of what you've heard during the course of this morning. I won't reiterate this. Massive floods of data that need more and more compute, that need more and more analytics, that need more and more memory, and that is incrementally where our invested capital is going to take advantage of what we see in front of us. We're no longer that business that's just going after data center CPU and PC CPU. We're leveraging the core competencies, the IDM advantage, Moore's Law, and the IP to extend our served market into these broader and broader opportunities. This is a wonderful opportunity for me to be the new CFO of this fascinating company.
At the same time, we know that you have some questions about some of the things we're doing. We know that you're questioning our level of direct spending. We know you question our ability to integrate acquisitions. We know you question our leadership in Moore's Law. We get it. What we hope you leave today here with is why we're doing what we're doing and the acknowledgment that we understand what's top of mind for you and the steps we're taking to improve the growth and the earnings horsepower of the company. With that, let me just talk about what I call delivering on the present, so a little more micro on 2017, and creating the future. Illuminating some of those bets that we're making that we think are the right things to do. We had the chance to talk to you a couple of weeks ago.
After a record 2016, we laid out our plans for 2017. Revenue growth in low single digits once you exclude the impacts of McAfee. Gross margins at the higher end of our historical range, 63%, roughly flat year-on-year, and earnings of $2.80 a share, which if you allow me to exclude McAfee for a second, it implies 5% EPS growth on low double single digit growth, during the course of the year. And at the end of the year, after a record 2016, our intentions are that 2017 will be yet another record for your company. And underneath the covers, the transformation continues. We've taken a fairly, what we hope to be a cautious view on the growth of the PC TAM in 2017.
Our guidance assumes low PC TAM down to mid-single digits and the implications on the revenue growth of our client business to be down in the low single digits. At the same time, we're going to continue to grow our other growth-oriented business that we've talked to you about earlier, and we're going to continue to invest in these businesses because of the opportunities that we see in front of us. Let me take them each one by one. You heard Diane talk about revenue growth for the year. We're expecting high single-digit revenue growth year-on-year, while continuing to expand our TAM and invest in new opportunities.
The growth drivers for the business, as Diane laid out this year, are continuing to win in the cloud, to expand our presence in comms, to grow our adjacencies, and to capture value for the wonderful set of products, particularly Skylake, that we'll be delivering and ramping during the course of the year. The margin drivers, the margins of this business are going to continue to grow in terms of dollars. The margin percent we expect to come down a little bit. We're investing. We're migrating to 14-nanometer, which at the early stages, the costs are a little bit higher. And I'm saddling Diane with more of our corporate-oriented costs. It's not a reflection about her business, or definitely not a reflection on Diane, but it's just how we allocate our costs. Great business, high single-digit growth, expanding TAM. Our client business, delivering on the present.
As I indicated, down mid-single-digit growth. TAM declining, disciplined investments, continuing to grow its adjacencies and segment the market to get higher core mix, and growing profit faster than declining revenue. Good execution by Navin and the client business in dealing with a fairly challenging market environment. I hope that our collective view of PCTM turns out to be cautious, but we think it's prudent to think about the growth in this way and make sure our costs are aligned appropriately. If it's better, then we'll all benefit as a result. NSG, a $2.6 billion business last year, growing 20% year-on-year. Key product deliverables during the course of this year for SSD in 2017. Margin drivers, we're still in investment mode in this business this year. The nature of the investments are a little bit different, as Rob said. We're scaling the fab.
We believe as we exit the year, the core NAND business will be a profitable business. We have significant investments this year for that next wave of products that we think are going to disrupt the market, particularly Optane. IoT, anticipating greater than 10% growth after 14% growth last year. We've had some big design wins where Tom, as a new member of the team, even newer than me, reaps the benefits of some key design wins in each one of those verticals that Doug talked about. With BMW on the automotive side, with Levi's on the retail side, and with GE on the industrial side. Important design wins for the business, and we're expecting good 10% growth year-on-year while we invest in the ADAS platform.
PSG, continued revenue growth, significant client wins in getting this new product out in the market to be trialed with hundreds of customers we feel very good about. Dan and the team are driving real synergies in this business. I'll talk a bit more about this later, in the mid-single digit growth, we expect margins to grow faster during the course of the year. Delivering on the present. That's our commitment to you in 2017. At the same time, we are making big investments for the future. Again, you heard about this from Murthy, from Doug, and from Rob. Memory technology is a disruptor. Playing in a market which hasn't been a big market for us in the past. A $100 billion market. NVM is the beneficiary of the dynamics going on in this market.
SSDs will get a disproportionate share of the memory game going forward. Most importantly for us, our 3D technology and our Optane technology, we believe will be gaining real share over the long-term horizon. Big investment, big capital going into this business. We think with technology and a leadership cost position, we believe this is going to be a big part of our future. Autonomous vehicles. Over 100 million cars built every year. An increasing portion of those cars are going to have ADAS technologies incorporated within them over the next decade. The silicon component in those ADAS-generated cars is going to be significant, and the opportunities to play even a broader role in the digitization of the automobile is big. This is a big market. It's relatively important for us, and we're investing behind it.
We believe that this will generate good returns for you going forward. Murthy talked about 5G, and in this connected world with over 50 billion cumulative connected devices, where this next generation of technology is all about processing data faster and faster. We believe it's important for us to establish and maintain a leadership edge in the transition from 4G to 5G, and this will be a part of our investment profile in 2017 and beyond. Delivering on the present and investing for the future. Let me shift gears a little bit and talk about capital allocation. Overall, no dramatic change in what you've been accustomed to over the last several years in terms of how we think about capital deployment. Simply put, two things, invest in our business to strengthen our competitive position and return value to our shareholders along the way.
How we go about doing that, investing in the business, we spend a lot of R&D. We realize that. We think those investments are worth it. We're excited about our core business and our opportunities to grow. CapEx, we consider it's large, but it's a competitive advantage that we drive a higher return on capital. Strategic M&A. We've made acquisitions along the way. I expect we'll continue to make acquisitions. That's not a sign to say we're going to do more. It's not a sign to say we're going to do less. It's just been a part of our portfolio, and we expect it to continue. Shareholder returns. Over the last 10 years, we've returned 100% of the free cash flows that we've generated back through the form of dividends, roughly 40%, and share buybacks, roughly 60%.
It's been a big part of our investment story to you over the years, and we expect that to continue. Our intentions are to grow our dividend in line with non-GAAP earnings growth. Our desires are to keep that around 40% of our free cash flows. As capital goes up, particularly for memory, we have to rethink about the level of dividend. Our challenge is and our commitment to you is we will continue to look at how we increase our dividend in line with non-GAAP earnings growth. We'll also continue to use our strong balance sheet and our strong cash flows to offset dilution from our comp-based program. Through all of this, we'll do it while maintaining our financial flexibility and a strong credit rating so we have the capacity to continue to invest and grow in this virtuous cycle of returns growth for you.
Let me take each one of these in unison. CapEx. In this year, we're going to step up our CapEx over last year, $12 billion. It's up about $2.5 billion, driven by two things. One, new capacity for 10 nanometer in our logic business, and two, incremental CapEx for memory. That's the growth year-over-year in CapEx. For the logic business, this is still in line on a percentage of revenue of what you've come to expect in terms of the capital intensity of the business. No real dramatic change. The real change here is the level of investment that we made last year in memory and again this year, and we expect those investments in memory to continue. R&D. I talked earlier about the reallocation of our R&D to the higher growth businesses.
This is an attempt to just show you how that reallocation of R&D is aligned with the expanded TAM that we've been talking about during the course of the day. First, Murthy and Navin and Aicha have dramatically reduced the level of R&D going into our client business. PCs down 5%, mobile's down 55% during this kind of two-year timeframe. Underneath the covers, they are still making the investments on segmentation for these new products of two and ones and products for the enthusiasts that bring higher ASPs. Dramatic shift in our R&D spend to the broader TAM in the higher growth businesses. You've heard from each of the GMs about where the incremental capital is going. For Diane, spending's up about 25% over the two-year timeframe.
As we invest in artificial intelligence, as we invest in expanding our share in the networking space, and we build competencies in adjacents like silicon photonics and Omni-Path. In memory, our spending's up quite a bit, this is primarily bringing our 3D NAND product to market, but also Optane. Last not least, in IoT, we're making significant investments to take this business and expand its TAM into these three big TAMs that Dov walked you through, while also starting to seed some growth to be well-positioned to participate in the nascent ADAS market. M&A, strengthening our integration. A year ago, we announced the acquisition of Altera.
At that time, I just simply pulled a page out of the deck that we use when Brian and Stacy announced the combination. What we said at the time is we would generate real product synergies from the combination, real manufacturing synergies from the combination, and that this business would be non-GAAP EPS and free cash flow accretive in the first year of the acquisition. A year later, check, check. We did what we said on this acquisition, and we feel great about how well it's positioned to continue to gain share in this increasingly important FPGA space. We've also made several smaller acquisitions during the course of the year. These acquisitions we characterize as how do we accelerate our technologies and our time to market in critical areas that we think are big differentiators for one of our businesses, if not all.
Movidius, and you've heard a little bit about this, but a low-power, low-cost, high-performance machine vision solutions to complement our Intel RealSense technology. Second, Nervana. Diane talked about Nervana in the AI world, and particularly in deep learning. A great addition to our data center business that we're reaping benefits from now, and that whole suite of AI products that Diane talked about earlier. We also acquired Soft Machines during the year, and this transaction really gives us a team of skilled engineers to further enhance our CPU-oriented products. All three of these acquisitions that were made in 2016 are on track or better than the case that we put together when we made this acquisition. We continue to make acquisitions, and we get better and better on these acquisitions to make sure that they achieve the intended purpose when we make them.
We feel good about kind of the last 12 to 15 months, our improved ability to execute and integrate acquisitions to make us stronger. An additional use of capital has been our ICAP portfolio. We haven't talked about this a lot to you in the past, but what a wonderful portfolio. Under Wendell Brooks' leadership, this portfolio continues to be more and more strategically connected to either our TMG and manufacturing world or to the individual business units to keep their eyes on emerging technologies that may be important to them going forward. This portfolio is roughly 350 companies we're invested in today. We derive benefits strategically from these investments. Our portfolio companies benefit from being part of the Intel ecosystem. You benefit from the attractive returns that this portfolio generates.
It's roughly 350 companies today, and our estimate of the valuation of these companies is roughly $10 billion. What I said. Just kidding. CapEx, R&D, deploying capital, and returning to shareholders. Our intention is to grow our dividend in line with non-GAAP EPS growth. The guidance I flagged for you earlier, in essence, for 2017, is that our non-GAAP EPS growth, if you set aside McAfee, is going to be roughly 5%. Our intentions are consistent with our stated policy, is to effective in the second quarter to increase our dividend by roughly 5%, or up 5% on an annualized basis. Subject to our board's approval, we will make this change in the second quarter for shareholders of record. This has been an important part of our investment story too over the years, and we will continue to make this an important part going forward.
I've talked a bit about some context, about delivering on the present while also investing for the future, and I've talked about our capital allocation philosophies to strengthen our business and return capital to you, our shareholders. Let me just briefly touch on what it means for us over the course of the next three years. We expect our revenue over the next three years to grow in the low single digits. The same equation that we've talked about so far, we will continue to be cautious about our outlook for the PC TAM business, roughly low single-digit growth in PC TAM, while our growth-oriented business will grow greater than 10%. That's kind of our outlook for the next three years for revenue growth. I should say that's my outlook. The ambitions of my peers, as you've seen from their presentations, are for higher growth.
I'm with them. For you, what I would suggest is you think about this business in this transformation to be growing in the low single digits over the three-year timeframe. I hope I'm being conservative. Secondly, operating efficiently. Our intentions are to grow our operating income dollars slightly higher than our revenue growth. The components of that are we expect gross margins to continue to be in the high end of that 55%-65% historical band we've given you, in the high end. But we do expect them to be a little bit lower than they were in 2016 and what we suggested for 2017, or roughly 63%. Marginal decline. At the same time, we also expect our R&D efficiency as the big bets materialize and our SG&A structure to gain leverage with our growth.
Third, we expect to grow EPS slightly faster than operating income growth, which we expect to be slightly higher than revenue growth. The components are, we generate a lot of cash over time. This is, in essence, an organic plan, so I'm not predicting M&A, but this is an organic plan over this three-year timeframe. As you know, we'll generate significant cash flows. That net cash position will earn returns and/or reduce share count. Secondly, our ICAP portfolio has generated attractive returns over time. We do expect that to continue. As you know, that tends to be more lumpy over time, but over the last 10 years and we expect over the next three, we'll continue to get good, attractive returns from our ICAP portfolio. Third is our effective tax rate. We've assumed it's basically going to be flat over this timeframe.
If and when there's tax reform and we're a beneficiary of that, obviously, we'll let you know, but for planning purposes, we've assumed our effective tax rate will stay relatively flat to this year's guidance of 22%. As I talked about earlier, our intention is to grow our dividend over time in line with non-GAAP EPS growth. In the short term, because of the higher capital intensity as we scale memory, what this implies is our dividend as a percentage of free cash flow will be higher than the 40% range that we've targeted in the future. We're okay with that because it's a sign of our conviction that the investment in memory is going to pay out over time.
In the near term, our dividend payout will be a little bit higher as a percentage of free cash flow as we embark upon growing our dividend in line with our EPS growth, which we believe will be slightly higher than our operating income growth, which we believe will be slightly higher than our low single-digit revenue growth. Again, the team's plans are more ambitious, but this is what we're committing to as we go forward in the next three years. Let me just wrap before I invite my peers up on stage, and we open up for your questions. We're in the midst of a transformation from a PC-centric company to a company that powers the cloud and billions of smart connected devices. There's a flood of data coming our way.
We think it plays to our strategic advantages and provides us the opportunity to expand the TAM in which we serve and compete in. We're making some big bets, but we will also improve the R&D efficiency or the efficiency of those bets over time. Last, and certainly not least, I know our capital return policy has always been important to you. It will remain a high priority for us in terms of the capital that we redeploy. With that, thank you for your patience.
Thanks, Bob. All right. I think we can step over here. We'll bring some chairs up onto the stage, we'll do a little bit of Q&A. After we get the chairs up, I'd like to invite the rest of the speakers to join us here. We will have some mic runners. You'll notice members of the Investor Relations team around the auditorium. If you have a question, if you can raise your hand good and high. We have spotlights in our eyes, it's a little bit tough to spot folks. We'll ask folks just to limit themselves to one question until we have an opportunity to make our way around the auditorium. We'll take just a minute here to get settled in.
Are you sitting?
No, I'll be standing. We'll jump into some Q&A.
Great. Why don't we start right over here?
Thanks. Blayne Curtis. Diane, maybe you can just clarify on the op margin guidance going down, you said a 40% op margin. Obviously, there's some one times there, so it should be a little higher. Maybe you can just talk about that impact. Keeping it at that lower range, maybe you can just talk about the puts and takes. Obviously, the adjacencies are growing faster. Maybe you can talk about whether that's a headwind or tailwind to gross margin and why is it at this lower range?
Yeah. First of all, our Q4 results, we were at 40% operating margin for the year. We were 44% for 2016. To your point, that 40% included a couple one-time events, and hence they're one-time events. Now if we move forward and look at 2017 and beyond, my point was the operating margin for the data center business comes down for three reasons. One is the good news, we get to go first. I'm more than happy to take that operating margin hit if I get to go first and be the first one to launch new products on Intel's process technology. Those early wafers are expensive, and so that has an impact. The second one is the adjacency businesses. Our beyond CPU product line, which has historically been the vast majority, 88+% of our total revenue, those have very high gross margins.
The adjacency businesses, whether it's silicon photonics or Omni-Path or 3D XPoint memory or Ethernet, all of those businesses are wonderful businesses. We're growing share, but the gross margins are lower. As they become a bigger portion of our total revenue, obviously, operating margins will come down. The third one was what Bob mentioned, which is not because he's trying to penalize us, but we do get the greater portion of the overall corporate allocation, whether it's the fact that because I'm on the leading edge of a process node, I take on more of the R&D burden for the process technology. Which is good, right? Someone has to carry it for the company. I'm happy to do that. Or whether it's just the SG&A allocation has a greater % of revenue, I take a greater %.
Those are the three reasons that pretty equally create about a 5% shift in operating margins. Going from 45% to 50% to 40% to 45% over the horizon.
All right. Then over here.
Oh, thanks. Vivek Arya from Bank of America Merrill Lynch. I just wanted to take a step back. You mentioned that Intel is in the process of a transformation. Transformation has positive connotations, but at the same time, at what point should we see Intel outgrow the semiconductor industry? When you look at all the R&D investments you're putting, look at all the CapEx investments you're putting, this year, you have low single-digit growth. You're talking about low single-digit growth. How do you put that in the context of a semiconductor industry where many of your peers are able to grow faster? They have higher profit margins. They have higher free cash flow margins in many cases. They might have slightly different business models, but how would you put yourself in the context of the semiconductor and the technology industry?
Are you being as productive as you can be with all these investments you're making? Thank you.
Yeah, it's a great question. Look, as we think about our business and the capital we're deploying, one of the challenges is how do we get the benefits of that capital being deployed, whether it's the R&D or the CapEx, across a broader range of products so we can grow faster than the industry. The dynamics are where we are in the stage of the transformation now is we have, again, a series of growth-oriented businesses that we believe are going to grow greater than 10% for the foreseeable future. We're very excited about the returns of those growth-oriented businesses. We are dealing with the practical reality of our historical success in the PC business, and it's still over 50% of the revenues of the company. We've assumed over this horizon that the PC TAM will continue to decline.
While we do have over 50% of our business declining by mid-single digits, at the same time, our capital being deployed towards higher growth is a growth rate that is much higher than I think what the industry experiences in the greater than 10% levels.
Great. Why don't we hit this side of the room over here? Stacy, please.
Hi, guys. Stacy Rasgon, Bernstein. I guess, number 1, a little disappointed that my quotes weren't up on the board.
Next year.
Next year. I wanted to ask a little about investment levels. Obviously, investment is higher today as you're looking at some of the making these big bets. You talked about R&D coming down a little bit. A little bit of leverage. I'll be honest, I still have a hard time seeing a scenario where you really get the kind of leverage that you used to talk about back in Stacy's days when he was giving the kind of longer-term operating models. What do we need to believe to see Intel actually getting a significant amount of operating leverage? Just knowing the kind of bets that you guys tend to make, I'm not sure I see I guess I can imagine big bets happening all the time. What do we need to believe that these are working and we can see them roll off?
A, I don't think you want them to necessarily roll off. You'd like me to find the next big bet.
Exactly.
Right? That's a good thing. What do you have to believe, though, to believe the bets, and how do we get the extension of our technology? Let's talk about that for a second, Stacy. First, you'll see every single thing that we talked about today talks about data And it's using a high level of compute. Whether it's the same architecture that's going to go in across autonomous driving or into networking and that infrastructure, all of the growth areas we're talking about are using this same high compute, high data, high analytics technology. A lot of what we're able to develop is able to go across all of these. You see that our accelerator technologies, like 3D XPoint or FPGAs, they fit into every one of those, too. 3D, if you take FPGAs, they're in networking boxes.
Depending on the design of the car, there's one FPGA for every Xeon in every one of the autonomous cars, and you just go through every one of those, and you see the synergy among these technologies. What do you have to believe? You have to believe that, A, the data rates we showed, you believe it, that the data is going to continue to explode, and hence, it's going to go to the cloud, and our technology and our leadership will continue in that space. If you believe that's one thing. Second, you have to believe that there's going to be all these use cases at the edge that are going to use that analytics of the cloud.
Whether it's retail and IoT or wearable devices or robotics or autonomous cars, that there are those use cases at the edge that also require a fairly high level of compute. Everything's not a dumb thermostat that sits at the edge. You have to believe that. The last thing I think you have to believe is that we could execute, that we can execute to these technologies and this leadership, which Diane in her discussion talked about. We've had these architectural battles in the past, and we've won. To the point, if you take a look at our growth businesses, we're growing double digits in every one of our growth businesses, which is as equal or faster than the semi industry. We're held down by the burden, yet the gift of the PC industry. It's declining.
It's huge, but it's a large cash flow, and they're managing it flawlessly, in my opinion. That brings the average of that growth down. That's not a bad place to be, and I'd rather be there and have that cash to be able to invest than a lot of the other companies in this industry, and have the power of our factories and everything else aligned to that. Did that answer the questions? I think you hope that I find the next thing, whatever's after autonomous cars and whatever's after robotics and IoT, whether it's quantum systems, which we're investing heavily in quantum computing, whether it's neuromorphic computing. We are investing in those edge-type things, the things that are way out there, that I think my daughters will run this company and be in charge of.
Thanks.
Chris
Okay.
Hi, it's Chris Caso from CLSA. Actually, to follow on from what Stacy said, I was actually happy that you didn't use my quote, so thank you for that. My question is regarding the overhead allocation, and for different scenarios for the PC business and I guess the enterprise server business, how does that affect the overhead allocation and therefore the operating margin of the business if you should see those businesses grow faster or grow slower than your assumptions? If it grows slower, are you able to cut costs so that those allocations don't have to be absorbed by some of the growth groups? By extension, if it grows faster, does that mean better operating margin?
Yeah. The allocations are relatively simple and straightforward, that we have a set of costs that are generic across all the portfolios, and we allocate those costs, in essence, based on the size of the business. That's an allocation game. It's how we allocate costs, not necessarily reflective of the underlying performance of the business. For example, I'll use Diane's example. Her margins, she's growing top line, she's growing operating income, not quite as fast as revenue. She has a changing mix in an expanding TAM that doesn't have the higher gross margins that our current business does. For me, that's all goodness. For Diane, she has to absorb a bigger chunk of cost. For Brian and I, and for you, it should be no different because it's a zero-sum game.
On your second aspect to your question, look, we're going to be looking at how do we become more efficient as a company. We've done it in the past. We've just taken a fairly large step to reduce the infrastructure of the company while still making important bets in the future. To go back to Stacy's question a little bit, if the ambitions of the team up here achieve higher growth rate than what I've assumed, that will drive higher return on capital and higher earnings growth. We're going to continue to challenge the cost structure of the business, but don't penalize these guys on how I end up allocating the existing costs that we have.
Thanks.
Next time, we'll get your quote in.
Why don't we get to the back of the auditorium here, John?
Thanks, Mark. It's John Pitzer from Credit Suisse. I'm going to perhaps ask Stacy and your next question a little bit differently. Brian, you really haven't shied away from giving us aspirational growth targets, whether it's been the DCG growth target of double digits or the TAM number you gave out today of $220 billion over time. Why not give us a more aspirational operating margin target as well? If you look at the history of Intel, you've always been in the top decile of operating profits for the semiconductor industry. You're now not even close. You look at some models running 40%, 45% op margins. Why not come out and give us a 35%-40% longer-term target? What's the reticence against putting profitability targets out there on a longer-term basis?
I think, when you think about putting a profitability versus a growth or a margin in an existing business, you run the risk of limiting investment. I believe that Intel needs to make. There's big transitions going on in the industry right now. If we don't hit 5G now, I believe actually over the next 18 months, we'll not be in a leadership position, for example, and we need to invest in 5G. Our Rack Scale Design architecture, we have an opportunity to really transition how the data center is architected and how it's operated on a day-to-day basis. If we put out operational margin, op margin aspirational goals, I worry will we take away from those investments that are going to be long-term growth. I'm more about top and bottom-line growth than just necessarily targeting the operating margin.
All right. Let's come back over to this.
Do you have a different opinion, Bob?
No. Look, for the last 2016, we expanded operating margin by 6/10 of a point. Our guidance for this year is to improve operating margins by another point. In our next two-year outlook, what I suggested is we'll continue to improve the operating percentage of the company such that income will grow faster than revenue. In that domain, and getting a higher margin percent in the grand scheme of things, I'd put in the relatively easy category. It's continuing to be more efficient while making the bets that we're making and ensuring we have the milestones in those bets such that they pay off over time when they may not be beneficial today. That's the more challenging part. We don't want to use an artificial percentage per se to make those decisions.
We want to use good analytics about the long-term inherent value in the investments we're making, while also continuing to be more efficient and expand our margins.
Great. Let's come back over to this side of the room. Joe?
I've already offered to Mark that I'll give you a more inflammatory quote for next year if you want to use it. In terms of 10-nanometer, you made a pretty good case for the importance of that. Can you talk about how that rolls out? Because it seems like you will have products this year, but your flagship eighth-generation client chip is still based on 14. You said you'll lead the way with new processes with servers. Is that something that happens at 10 or is that something that happens at 7? Thank you.
We're not changing the guidance that Brian's given on earnings calls of previous vintage. We still expect to be shipping our first 10-nanometer products before the end of the year, with volume in the first half of the year. Again, I think it's very, very important to make sure that you took away one of the key precepts of my presentation, which is in addition to these eonic node transformations, we're going to continue to exploit intra-node improvements. What you'll see going forward is a fairly fluid blend of those big technology improvements together with very, very focused, really extraction of goodness out of our existing technologies. As you saw, at the end of the day, what the end user will see is a better than 15% improvement on industry standard benchmarks.
That's really what they're buying, and that's what we believe will continue to stimulate PC refresh. I think the same is the case for the server business, that what we're really doing is focusing our technology to continue to improve use cases that really matter to the end user, whether they be in delivering peak performance or whether they be delivering that performance at lower power.
I'll just add, there's a significant lead time for the design engineers to target a process technology. When VK flipped the corporate strategy and made Data Center first, many of the product ships had already sailed. The first time we have on a new node to intercept is seven nanometers, but we are backing it up, and as tectonic-- What did you say? The internode of 10 nanometer, there's a third wave. We have the original launch of 10 nanometer, then there's a second wave and a third wave. We'll be first on that third wave and then first on seven nanometer.
Great. Thank you. Back over here.
A question on Foundry. I spoke with Sohail during the break, and he was very bullish on the Foundry, but I was just wondering, is it going to be an approach more getting smaller customers at the edge or getting share from larger customers, just like in the modem strategy with the recent announcements?
I'd say on the Foundry, our approach is very strategically specific.
We're certainly not positioning our foundry ambitions to be a general purpose merchant foundry. What we're really looking at is where there are complementary sets of benefits between our partnerships. Just like, for example, our partnership with Altera started as a fab arrangement because essentially the FPGA logic structure lends itself towards being able to support the ramp of new technology. We're looking at those kind of areas in other partnerships. For example, low power, high-performance silicon is an area, and we publicly announced some partnerships there that really help us in that area. I think we're really looking at how we can strategically partner to improve the depth and breadth of our IP portfolio. We talked about recent IP licensing agreements we've already made.
We're looking at this very strategically and how our foundry customers can provide indigenous benefit to the products that we're also developing internally.
Great. Back over here to Ambrish.
Thank you. Ambrish with BMO. Bob, I loved what you did first analyst day when you started a hashtag, here's my quote contest for next year. My question is for you, Diane, DCG. Thanks for all the details. One thing I didn't quite catch was the ASP trend. We've seen the last three years, units have grown 8% three years in a row. ASPs have gone +10%, +4%, down 1% on an annual basis. Should we be expecting the same kind of trend in ASPs? I get it's a bit of a mix as well as, I don't know what's the next part. Also embedded in your forecast for the 3-year cadre, 3 to 5-year cadre, are you anticipating any competition? At x86, you have 99% market share, and there's no also ran. There's no ran. Is there a change that you're anticipating?
Thank you.
Yes, thanks. First on the ASPs, and that's why I went to the trouble of putting together the slide that showed the breakout between Xeon and SoCs. First I'll say our Xeon ASPs continue to rise and have risen over that horizon that you talk about. DP Xeon has risen, MP Xeon has risen, UP Xeon has risen. The Atom SoCs have risen. All of our ASPs continue to go up. As we've been winning in some of the low end of the network markets with our Atom SoCs, it's high volume, it's low ASP. That has had a dilutive effect on the net ASP. We're trying very hard to break out our ASPs in the future.
We talk about the Xeon, and you can see the continued growth there, as well as the low-end CPUs that are deployed in some of the very low power, low performance demanding parts of the network. High volume, low performance means low ASPs. Margins across all of those are great, right? There's no difference in the margin that we get on an Atom SoC than a Xeon SoC than a Xeon full-fledged product. Margin is all good margin dollars. The second one on x86, we've always had baked into our plans some decline, whether it be market segment share decline or ASP decline, given a competitive environment. We've always had that baked into our long-range plans.
If you go back five years, it was concern about Arm coming in, the microservice space. It was open power. Now AMD coming back with an x86 processor after a five-year hiatus. We've always had that baked in. The competitive threat comes and goes. We always have competition. That is included in our model.
Great. Thank you.
Up front.
Yeah. Romit Shah with Nomura Instinet. Bob, question for you. If data center and memory and IoT are going to be receiving a bigger allocation of the operating expenses, should investors be prepared for more quarter-to-quarter margin and earnings volatility, just given the inherent lumpiness of those businesses?
To the earlier question, the cost base that we have is the cost base that we have going into the year. During the course of the year, at the Intel level, the costs are going to be allocated by business. There may be a little more lumpiness, but not dramatic. At the Intel level, it is what it is in terms of our overall cost. I wouldn't be too concerned about how costs get allocated in terms of driving the lumpiness of Intel's earnings.
Great. Let's come back over to this side of the room. Is that Chris?
Yeah. Thanks. Excuse me. Brian, I'll ask you the same question I asked your predecessor, Paul, several years ago. You're up here, you're taking us through all the growth prospects. Some of the growth markets are admittedly lower margin than your current or core business. In terms of your main competitor, not necessarily on the processor side, but for mega cap investment dollars, TI has had a different approach. They say, "Hey, we're not a growth company. We're just going to stick to our knitting and give all the cash back to the investors." It's done pretty well. Maybe take us inside the boardroom or the planning sessions, does this ever come up and say, "Hey, let's just stick to the DCG PC business that could grow low to mid single digits.
Our margins would go up, give all the cash back to the investors." Does that ever come up? Why or why not, and what's the reaction there?
Well, actually, we have two board members here, and you could actually even ask them. Sure, we have the discussion: Are we a growth company? Aren't we a growth company? Do we want to keep investing? I think in every case, it comes back to these are extensions of our technology. We need to have this growth in order to continue to fuel our factories, right? If you take a look at the business model of Intel, we lead even in the data center, in the PC, based on our silicon leadership to start with. Our silicon leadership starts with the ability to invest in Moore's Law, to invest in those factories. Those are going to continue to get more expensive.
The minute you stop growing, you're basically saying over time, likelihood is you will run out of the ability to afford those factories, and then that leadership will slowly diminish. That may take 10 years, but that will happen if you believe in that. We believe that we're going to continue to invest. This is the discussion we have is there a strategy behind this investment or are you just going out and targeting random areas? No. We're targeting areas that are data-driven, and that can take the technology and the IP that we generate in those core businesses and extend it. If we believe we can grow with that, then that's the right way to try and grow this company. That is literally how the discussion goes.
You do have that discussion up front of understanding the basic business model of Intel. That basic business model is the reason we have the leadership and the product quality that we have in the PC and the data center space, it starts with the silicon leadership in our factories. You have to be able to continue to afford that investment. By the way, the same thing happens in memory. If I didn't think we had leadership technology that could allow us to grow, share, and grow that business, if we think of a commodity business, that's probably not the right business for Intel because we're just not architected as a company to really drive that. We're making those investments because we have differential technology in silicon again that we think we can leverage. That's what's going to fuel that growth in that.
That same kind of silicon to products to profitability model is living in that space as well. That's, again, exactly the discussion we had in the board.
All right, back here to Tim.
Thanks, Tim. Back here again. One question for Bob. I know that we used to get a target of 30% long-term OpEx. My question is there a target you have longer term for OpEx as a % of revenue? In the near term, in the event that revenue is better or worse than you think, how fungible is OpEx?
At this stage, my long-range horizon is three years. What we've talked about today is how do we take our operating expense from 2015 to 2017 is down two points. Then over the next from 2015 to 2017, and then in 2018 and 2019, implied in what we talked about is that it will come down even more. In between now and then, we'll continue to look for opportunities to make that investment more efficient on the R&D front, i.e., get higher growth and continue to look for ways to get more and more leverage on the SG&A front. That's the horizon we have right now. I wouldn't say that over the long haul, we can't get to 30, but right now our focus is we got core businesses we're executing on.
We got some big bets that we're making, and we're trying to make those investments work such that we can grow both top line and bottom line fast and top line over the near three-year horizon.
Great point. Back over to the far edge here. David, please.
Thanks so much. David. Question for Diane. Could you fill us in a bit, Diane, on accelerators? When will we expect to see the first Xeon chips with integrated PLDs? When do you expect first revenues from accelerators for AI type applications? What's Xeon Phi running at the moment, in terms of revenues?
We have Xeon plus FPGA integrated solution that is out in the market today as a software development vehicle. Giving the end users the opportunity to figure out how they would use that combination to accelerate their data centers. That as a software development vehicle, it'll go to production towards the end of this year into production ramps using the Skylake processor with the Stratix 10 FPGA product. That'll kick off and go well. Xeon Phi, your question was how is it?
Revenues for it.
Yes, we don't break out the Xeon Phi product line revenue from the rest, but you can see we're continuing to win share in high-performance computing with Xeon Phi. Xeon Phi is increasingly being used for artificial intelligence solutions as well. Deployment is not just for supercomputer and scientific workloads or for R&D modeling and simulation, but also being deployed now for artificial intelligence. That product line will continue to get more and more robust, targeted at artificial intelligence with the Knights Mill launch at the end of this year. We expect great growth in the Xeon Phi product line.
All right. We're going to take one last question back here. Back center of the room, Mike.
Yes. Bob, I was wondering, why do you expect gross margins to go down modestly over the next three years? Diane, what was the product delay you talked about on your slide on your DCG deck?
The modest decline in gross margin was primarily a function of the higher growth businesses, particularly memory, modem and some of the adjacencies will have a negative mix impact on gross margin. That's what we see over the near term.
On the products, as I showed you, when we look at the future growth of the Data Center business, the growth of those adjacencies become a significant portion. That's everything that's not a CPU. Those are big new inventions. They're product lines that didn't exist before that we're inventing grounds up and that have taken many years. Two of them that we talk a lot about, silicon photonics, that was a 16-year development investment. As Isaac will remind me, just as we were about to launch, we had a snafu and we had to delay the launch by a year. That was a year launch in the ramp of silicon photonics. 3D XPoint is another one, over 10-year R&D investment. This is tough stuff.
We had a plan to launch it, time to market with the Skylake platform, which is our platform that we launched this year. That's going to be delayed to end of year, ramping into 2018. We have these massive invention of new products, new product lines, many years of R&D. In a few cases, they've pushed out a bit, and that just simply delays the ramp of those adjacent businesses.
All right. Well, I'd like to thank everyone who joined us on the webcast today. We'll wrap the webcast up with that. For those of you who are with us here in the auditorium, I'd like you to join us in the cafe for lunch with the leadership team. Your table numbers, again, should be on the back of your name tags. If you can't find your table number, we'll have hosts in the cafe who will be happy to help you find it.