Thank you very much everybody for coming to our Analyst Day this year. I'm literally going to stand here for not very long, explain the context of the day this morning and what you're going to hear. You're normally used to hearing from me and from Tim and Ian, you're going to hear less from us this morning and more from the Arm team. We're going to kick off with Simon, who's going to be highlighting the importance and key benefits of the Arm business model, which our business model is a key differentiator for us, and that's what enables the Arm world to deliver innovation. The business model itself also delivers serious economic benefits to our partners and to their customers.
He'll also look at the next five years or so of market evolution and some of the opportunities there for Arm and the partnership, and how we're going to bring real products to market to realize that. The business model is crucial, also the technology that underpins it is crucial, Tom is going to talk about how Arm is so good at low power, and power efficiency in particular. By that time, we'll all need a short break. After the break, we're going to look at a couple of areas that are contributing to growth over the next several years for Arm, and that are of strategic importance for us. One from an Arm product perspective. Pete is going to talk about our graphics offering and how we're developing the potential for leadership in graphics.
You're going to hear from Ian, who will be talking about server design with data centers as an entry point for server design. The whole significant opportunity that's both presented to and enabled by the Arm architecture being used in servers. Before we close, we'll have our Q&A session. Actually, before we get to the Q&A session, Tim is going to summarize the growth opportunity and what that means through a financial lens. Talking about opportunities for revenue margin and earnings growth over the coming years. With that, I will hand over to Simon to kick off. Except the slide isn't advancing, Simon. I'm sorry.
Let me help.
See if you can make it work. There we go.
Thanks, Warren. Good morning everyone. Hope you're doing well. As you saw from the agenda, we're going to talk about some of our technology. Tom's going to talk about our CPU roadmap, what we're doing about low power. Pete's going to talk about our graphics roadmap and what we're doing there. Ian's going to talk about how some of this technology comes together in a product. Before we get into all of that though, what I'm going to do this morning is talk about Arm's approach to business, and Arm's approach to working with our customers, and how we go about developing some of those products. At the heart of all of that is partnership. Partnership has been the way that we've grown this company over the last 21 years, and remains our view on the right way of doing things going forwards.
Before we go look at the future, let's just go back a bit. Arm is 21 years old. I was fortunate enough to join the company just after we got started. I've witnessed firsthand the evolution of the company and the business during those 21 years. It's been a phenomenal evolution over those two decades. Interestingly, we take mobile phones for granted today, and at Arm, we spend a lot of time thinking about mobile phones and where they're going, and you'll see that as a trend this morning. When I joined the company, actually nobody else in the company had a mobile phone. Apart from our CEO, who had this big clunky old analog thing, nobody else had a mobile phone.
Which is quite strange when you think about today, where there are enough mobile phones in the world for everybody on the planet to have a couple. At the time you could buy a mobile phone. The original Motorola phone shown there were about $4,000 to buy new at the time. They were outside the price range of most people. Fortunately, over the last 21 years, these things have evolved enormously into the devices that we carry around and take for granted today. Computers have evolved a lot as well, and these two things have come together into what is now barely recognizable as a mobile phone, is really a mobile computer. Today, of course, we have a phenomenal amount of compute power in our pockets, and there's, I'm guessing, a huge amount of that in the room right now.
Functionality has gone up, and through economies of scale, through the way process technology has evolved, and through the way that Arm's business model has evolved, we've been able to deliver all of that massive increase in functionality at a much lower cost. That's been one of the key things that's helped the industry as a whole evolve, and technology evolve to what we know today. The industry has evolved a lot in the timeframe of Arm, and if we go back to the '70s, before Arm was around and certainly before I was at work, what you saw was fully integrated companies. Companies that did absolutely everything themselves. They did design, they did marketing of their products, they sold direct to consumers, and they manufactured everything. They were completely vertically integrated companies.
The problem with that was if you want to design anything, you had to do everything. That's a very costly way of approaching the world. Over that period, since the '70s to today, what we've seen is the industry disaggregate. First of all, if we look at semiconductors, you had the era of the ASIC model, where there were companies who specialized in getting your chip built, in doing the manufacturing, and in handing you back a finished working device. That went into systems integrating companies. I used to work for telecoms company that operated in this way. We did ASICs with LSI. They gave us chips, we built telephone exchanges. That model continued to disaggregate as well. What we saw going after that was the advent of EDA.
ASIC companies couldn't afford to do all of the software development for tools design themselves. There just wasn't the economy of scale there. You needed specialized EDA companies who could amortize the cost across the entire industry. Same with IP. When Arm came along, if you wanted a microprocessor, you had to build your own. There was no other choice. It limited the number of people who could build CPU-based chips. With the advent of Arm, again, we were able to design CPUs, design other IP, and amortize the cost across an entire industry. It made it very cost-effective for lots of people to do design. What that has led to is this industry where companies can specialize, companies can achieve economies of scale, and deliver a much more efficient solution from an industry cost perspective.
That has in turn enabled very sophisticated designs to be done by lots and lots of different people. We've seen design costs going up along the way for sure, and I'll come back to that. We've seen the number of transistors that you can put into a device go up enormously, and the sophistication of the device go up enormously because of the way that this specialization has occurred across the industry. The downside of this disaggregation is, or potential downside, is that if everybody concentrates on what they do alone, then the whole thing becomes actually a bit inefficient. If I only worry about CPU design, and I leave you to worry about putting your chip together, then it may be that I make the wrong decisions based on the next problem that you have to solve.
Smart people have recognized this, personally, I like to think of this as not a stack of people operating a supply chain, actually more of a kind of circle of companies where the smart ones have kind of worked out that actually communicating a lot with each other can reduce and remove the potential inefficiencies between these slices of this disaggregated chain. A couple of good examples. As I said, IP has been really an approach to design that's been really transformational. It's allowed people to take very complex building blocks and build really sophisticated chips without having to do everything themselves. At the end of the day, you've got to build that. The way you build that is to use EDA tools, and you need a process at the end of it to actually go manufacture the transistors.
By IP companies and EDA companies and foundries working together, we can look at some of those issues up front so that a designer can then take all of that knowledge, safe in the knowledge that when he comes to bring it all together, it's actually going to work. Through collaborations like that, we've been able to reduce some of these inefficiencies. The key to it is an approach to business that is around partnership and openness. That is what Arm has been doing, and it's what we intend to continue to do as we go forward. At the heart of our business model is about helping the efficiency of the industry. What we do is design IP that we license to people who build chips.
Those chips in turn get sold to OEM companies who build their own products, and every time that they sell one of those, then we extract a small royalty from that. Our remit is to create those designs to work very closely with our customers, to work very closely across the supply chain, to understand the needs of today, the needs of tomorrow, and the needs of next week to make sure that our technology is going to be well suited for the future needs of those end products. As I said a moment ago, in 1990, if you wanted to design a CPU-based chip, your only choice was to do everything yourself. You had to design the processor, all the software tools. If you wanted an operating system, you had to write it, and you could only amortize those costs over your end products.
If you have a lot of them, that's great. You can afford to do that, very few people can. The IP model of providing processors, and Arm has been enormously successful in that, has proven to be a very successful way of enabling lots of people to do chip designs without having to reinvent the wheel every time. We are not alone in that. We have a very active partnership of over 900 companies. We call this the Arm Connected Community. Part of what we do is taking the profit that this business generates and obviously investing it back into our R&D roadmap, and you'll see some of that this morning. We also invest that in developing this community of companies who are going to help with the usage of Arm technology by anybody who wants to use it.
Whether you're writing code for an Arm processor, trying to get your chip taped out or tested or packaged, trying to run an operating system or applications on top, through this network of companies that we actively develop and engage with, there is somebody who can help you. That is a really, really important thing. We work hard to ensure that no matter what you're doing around Arm, there is someone where you've got a problem who can help you solve that problem and do it in a way that is economically viable for them as well as for you, as well as for Arm. That community is a great thing. The partnership is a great thing, but it does only work if everybody can make a profit from that. We recognize that and work hard to ensure that the community around Arm is as vibrant as possible.
This model, the way the industry has evolved, the way Arm has evolved, has helped create lots and lots of different products. If we look at how some of these have evolved over the years, going back to 1990, as I said, there was big old analog phones which didn't have an apps processor inside them. Nobody really thought of apps. The internet didn't really exist back then, or the web didn't, certainly. Making phone calls was just about all you could do with it. I actually have one of those phones in my office, it is so big, the instruction manual is actually printed on the inside of the battery. If you're wondering what to do, you can take it off, remind yourself, put it back together, off you go.
In the first decade of Arm's existence, what we saw was phones get a lot smarter. Arm processors came into these devices, they moved from being machines that you could make a phone call on, to machines you could start to organize your life. We saw the first smartphones, really feature phones with Arm processors in them, with enough processing power to run other applications to help you with your calendar and your contacts, et cetera. In the same time period, what we saw with desktop PCs was them go from being big, gray, and ugly to be big, gray, and ugly. Pretty much doing the same thing they ever did, running the same software that they ever did, enabling you to do what you did before, without really any significant reduction in the cost of the raw materials.
The CPU still cost you a couple of hundred GBP. In the last 10 years, we've seen smartphones get even smarter. The number of processors integrated into those smartphones go up. The amount of technology integrated into one single chip go up and up, with more connectivity, more functionality, such as graphics, as you'll hear about later on, all integrated into one chip. Where, through economies of scale, through the manufacturing industry that exists around the fabless community, you can get that chip for as low as GBP 15. That is not a lot of money to spend for a hell of a lot of transistors. On the CPU side, on the desktop PC, on the PC side, things have changed a bit. Laptops are much more common now than they were 10 years ago.
Still, though, you do pretty much the same things with pretty much the same old software. Whilst some Arm technology has got in there, in the disk drive and in some of the connectivity, the main CPU is still going to cost you GBP 100. Still very expensive compared to all the functionality you can get on the left-hand side of this slide for GBP 15. That doesn't look particularly satisfying. Going forwards, though, I think the next five years or so is going to be where it gets really, really interesting. We're going to see, on the left-hand side there, more and more functionality, higher levels of data connectivity, higher levels of graphical and video interface.
Where we have a new opportunity is on the right-hand side, where the Arm technology now is sufficiently powerful in terms of delivered performance, that we can start seeing it in more conventional clamshell form factor and other form factor devices. With the very low power consumption, the high integration of the functionality into these chips, that can sell for GBP 20. This opens up, I think, a new realm of devices, new opportunities for new form factors, very low cost, no fan, very thin, very lightweight. I think we're going to see a broadening evolution of the type of devices available, all based on Arm technology.
I think the next five years is going to be really interesting as we see this evolve, as what you do with the device that you see on the right-hand side there can suddenly change and how you interact with the web and data services at large can change. It's going to be really interesting. As we look at how we've got ourselves to here, the question then we need to ask ourselves is: Are we organized the right way for success going forwards? Before I go into that, what I'm going to look at, though, is some of the technologies that we look at within Arm as key drivers of the next five years or so. Excuse me. The key technology areas that we look at are, obviously, mobile computing.
As I said, we spend a lot of time in Arm thinking about phones and tablets and thin form factor devices. Also servers. If you had a chance outside before this event started, we have a Calxeda box there, which is really interesting. Connectivity, how all these devices are going to talk to each other. The Internet of Things, which may seem a bit less glamorous, but personally, I think is a really exciting technology area. Let's look at some of these in a bit more detail. Now, mobile computing. You may think it's just more of the same, but I actually think that the way in which mobile computing is being driven has changed. It's gone from "What technology can I put in this device?" to what the users actually do with these devices.
The usage model is much more driving the specification and the requirements of mobile computers now than they ever were. Devices have changed, as I've been saying, enormously from machines you make a phone call on, to machines that you operate your life with, that you interact with the web, you interact with data services. Some of us were talking about that just now, just before the event started. How you organize around your work life and also your social life, your home life. Increasingly, a lot of the way you're using this device is about how you interact with others, how you interact in a social way, which doesn't necessarily mean goofing off at work surfing Facebook. It means having the right data and right interactions with the right group of people based on the context that you're currently in.
It may be an event like today, blogging about what's going on at this event or the conference. It may be sharing photos of your son's broken arm, which happened to me last night, with the rest of your family while you're on the road. Using the device to interact with both work, with your personal life in a more connected, social way is really driving the evolution of these devices. The technology, though, still provides some underlying limiting factors. That usage model is driving some of the key things about data connectivity rates, about screen sizes, and about security, which is another really important area that we see. As more and more of your personal data goes onto this machine, then you're going to really, really care about how secure it is. Those are things that get in the way.
What's clear is that we're going to need a range of different solutions to address future needs. Now, I'm sure all of you in this room, given what you do for a living, are pretty well aware of how mobile phone volumes have grown over time, and the forecast out for what they're likely to do going forward. Here we have some data from Gartner showing over 1.2 billion mobile devices out in 2014. This is smartphones in all levels of smartness, because we are seeing a kind of tiering of levels of smartness. In the West, it's very easy to think about the very high-end phones, the so-called super phones with big screens and wizard graphics, and all the rest of it. There is a huge growing market for low-cost devices in emerging economies, and a big business opportunity around that.
Kind of 2009 was an interesting takeoff point for smartphones, and I think a couple of things kind of came along at the same time. We were able to deliver a processor that was powerful enough to enable different applications to be downloaded onto a phone. What really changed at that point was all the software that the machine ran was not on it when it left the factory. Now, that's the case of a feature phone and of a basic phone. With a smartphone, you can download applications. With a combination of a high-performance processor, of a touch screen, and of an open operating system, we've been able to put a platform in the hands of thousands of developers who can then take all that underlying hardware and innovate around it.
We aren't constrained by the person that designed it in the first place and the narrowness or restrictiveness of his imagination. We've been able to give this to thousands of people who can now think about it in different ways, and people regularly find ways of using the underlying hardware that none of us ever thought of when we were thinking about the Arm processor that's going to go out in the future and nobody in the device manufacturer thought about either. That's what makes it really interesting and means that there is an insatiable need right now for more and more compute performance in these devices. More CPU performance, more GPU performance, and using these in very creative ways. We see needs for more and more compute power going forward, and you'll hear about some of our roadmap activities later on this morning.
We're going to continue driving that forwards for many years to come. What's key, though, is that with all of these different devices, you need a different range of solutions. You're going to see single-core phones in low-cost markets. You're going to see quad-core phones in high-performance markets. Whilst it's easy to think about just the top end and just the apps processor, there's actually a lot of other technologies within your phone. I don't recommend you do this, but if you took your phone apart, what you would see is many different silicon devices in there. You'll see the big apps processor, and that's the kind of big, sexy thing that gets a lot of air time. It's manufactured on the latest, greatest process technology. You'll see a lot of other devices in there as well, interfacing with the outside world.
Whilst we like to think about all things digital, the outside world is annoyingly analog, and interfacing with it is actually quite difficult. What that requires is a range of different technologies, different process technologies, and it isn't all about 22 nanometers and funky transistors. Some of those older technologies are going to stick around for a long time. 0.18, 0.13 micron, higher voltage domains, analog devices to interface with the outside world. The power management, the touch screen require analog interfaces. Those technologies are very important. We have to keep driving the cost out of them, but you'll see them around in this device for a long time to come.
As we try and continually shrink the form factor, what's going to be required to put all of these technologies together is more creative ways of packaging devices, of stacking die together into what's called a 3DIC, where the connectivity between the different die is straight through the silicon. Lots of technical horror associated with that that we have to solve over time. Different technical challenges that aren't just about making the CPU clock faster. Lots of different dimensions on which we are evolving the technology for mobile devices. I forgot the slide was going there, I say. On top of all of those devices is the software that runs on it. Obviously, we have been working with Microsoft on their Windows on Arm initiative. This is carrying on from a long engagement that Arm and Microsoft have had.
Arm has had people up in Redmond working on Windows devices for a long time. Windows Mobile's been running on Arm for many years now. We're expecting to see that deployed into these very high-end mobile devices, again, in a different range of form factors. We think that's going to deliver some really interesting technologies. Those clamshells that I was showing on one of the earlier slides, different tablets, different form factors running Windows, is going to offer some new opportunities for people making Arm-based devices. We think this is a really interesting time. That's going to roll out obviously over the next little while. We're working to get ready for that. We're working with our partners to get ready for that. We're very excited about the prospects of the kind of device that we're going to see.
You don't need to sell many of those devices before somewhere in the world somebody needs a new server. Later on, Ian's going to be talking about the work we're doing around servers. All of these new devices, as they grow, as they consume vast amounts of data, is going to lead to an order of magnitude increase in the amount of data flowing around the internet. That just does require lots of servers. That's a great business opportunity in its own right, servers are annoyingly power hungry. Unless you want something the size of the Three Gorges Dam power station in your back garden, we need to do something about that. We are doing something about that, both in Arm and with the Arm partnership and with the Arm community around all of that.
Servers are a big drain on the world's power consumption. There is an opportunity to take all of the goodness around the Arm model, very low power processors, very efficient manufacturing and design, to build SoCs for servers. Instead of conventional multipurpose chips that fuel servers, what you find is if you know what you're doing, then it's always most power efficient to build some dedicated hardware to solve the problem. Servers have themselves evolved from being general purpose things that calculate the weather in Tokyo next week to farms of machines which are all running the same application over and over and over again. When you've got a context like that, an SoC may well be the best approach for achieving very low power. That is something that's very interesting to us. It's interesting to our partners.
I think over the next few years, you're going to see different approaches to building servers, which are going to fundamentally deliver more performance in a very, very power efficient way. That needs an ecosystem around that. Again, I don't want to steal Ian's thunder, I'll let him talk about that. At the other end of the computing spectrum is the Internet of Things. What the Internet of Things has in common with servers and cell phones and everything else is it's about getting the right amount of compute power in a very low power implementation and then deployed in massive quantities.
The Internet of Things is about combining sensors with processors in very power efficient networks to gather data from everywhere in the environment, from machines talking to each other, embedded in buildings so that you can control the lighting and the heating, embedded in the road so you can work out whether there's a traffic accident, and generating vast amounts of data. Data in itself is not particularly interesting. What is interesting is the information that you can glean from that data. The data needs passing to a server so that we, as humans, can take action based on it. There's some good examples about how this may help us all in our day-to-day lives. Yesterday, driving around London, turns out there was a garden party at Buckingham Palace. Our cab driver knew that, so he took some different route.
If I'd been driving around London and I didn't know that, I'd have got stuck in a traffic jam. Were the Internet of Things deployed and traffic information was being recorded live, maybe the satellite navigation in my car could get reprogrammed on the fly to divert me around it. I can tell you, a lot of people did not know that was going on yesterday, and there was a lot of traffic congestion that this could have helped with. Other things, there's some example out there with a lighting application. If your dishwasher tells your washing machine not to run the spin cycle right now because I'm about to, then you'll get a smaller power spike into your house. That helps with energy deployment, helps you with your electricity bill.
There are lots and lots of ways in which intelligence embedded into everything around the planet can help make the world a more efficient place. We think there's a big opportunity for Arm technology in there. Connecting all of that together is mobile infrastructure. We're seeing the amount of data flowing around going up enormously. As a result of that, the architecture of mobile infrastructure is changing from being very centralized to being more decentralized. We're seeing the growth of micro servers, micro cells, femtocells deployed closer to the edge to gather the data, to get it into the internet, give the answer back to whatever mobile device you happen to be carrying around with you. Change in architecture here. Again, big opportunity for more machines and more deployment of Arm technology. As technologists, we look at this and go, "Fantastic.
Big, hairy problems here. Intellectually interesting problems to go and solve from the technology point of view. From the business point of view, what's clear is the demand for semiconductors is not going down anytime soon. All of those devices have in common that they need high performance, low power, low cost. That is what the Arm ecosystem is really good at delivering. We've used this partnership model to get from where we started to here. We strongly believe, I firmly believe, that the partnership model is the way that we're going to address some of these key technology and business challenges going forward. It's all about collaborating in an open way. At the whole, what we do is design technology and deliver it to our customers who build chips. The cost of building a chip is pretty scary these days.
If you wanted to build yourself an advanced fab on 20, 22 nanometers, something like that, the bill for the equipment alone is going to cost you about GBP 6 billion. That's on top of the GBP 1 billion or so you will have spent, maybe GBP 2 billion, on the process R&D. Very expensive proposition. The cost of all of that has to get amortized across the chips that are built on it. That means that the cost of the silicon, unless you get the scaling out of it, goes up. That's not something, fortunately, that's happened so far. We're going to have to keep working away to make sure it doesn't happen in future. Utilizing all the transistors that you put down leads to increased design costs.
Verifying that when you put 1 billion transistors down on a chip, they're all connected up in the right way is a non-trivial problem. The cost of design, the cost of verification goes up. The cost of manufacturing has gone up. The masks that you need to run through the fab, very expensive, maybe a couple of million GBP for each design. You better not get that wrong. People spend a long time on verification. The cost of design going up. Then when you look at the software that's going to run, that is also getting very complex and hence very expensive to develop. All of these costs are pretty scary when you look at it.
Again, it comes back to the only way to approach this is to take a modular approach to integrate highly optimized and verified building blocks that you can put together to leave yourself with a tractable problem. You just cannot afford to do everything from scratch yourself. You have to leverage work that is done around the industry to have any hope of building a device of the kind of complexity that we and our partners do actually put together. To do that requires collaboration between the key partners in the ecosystem, between designers actually building chips, between people manufacturing them, and everything in between. Between the IP companies, the EDA companies, the software designers.
It's really important that we work together around open standards, around APIs that everybody can get access to ensure there's a high degree of reuse from one design to the other and a high degree of knowledge sharing to make this a problem that is actually you've got time to solve. I think that the industry's only going to survive against this increased complexity, increased cost, if we operate in that way. Openness, collaboration, absolutely the key to our success and to this industry's success going forward. When we look at the kind of chips that we anticipate our customers to build to solve a lot of these problems going forward, we're seeing an increased number of multi-core designs, gigahertz microprocessors integrated into a chip with maybe one or two specialized processors, things like graphics and video, running very low power at very low cost.
That GBP 10-GBP 20 price range, really important to enable the wide range of end device that we're getting used to dealing with every day. For our part, it's important that we build those building blocks to prevent wheel reinvention every time in our customer base, that those building blocks work really efficiently together so that jointly we can unleash the creativity of thousands of designers around the world. If we go back to the integrated model where only a few people do each step, that is not going to lead to a vast proliferation in the type of devices that we use. It's not going to lead to an awful lot of choice for consumers at the end of the day, and we think that is a bad thing. Which is what we're trying to prevent.
When we look at advanced process, again, there are some pretty hairy issues that you have to deal with. Annoyingly, physics does get in the way. As you're trying to build transistors as small as 20 or 14 nanometers, you've only got a few atoms of silicon to play with, and they tend to behave a bit erratically, and so you have to work out how to deal with them. Manufacturing is very hard. There's lots written about everybody having issues with manufacturing right now. That is not surprising. It is very difficult. It always has been hard, but it is really hard right now. Fortunately, the industry's got a fantastic track record of solving these problems, and whatever issues exist today will get solved over time. Of that, I am pretty sure.
As we look at how Arm designs get deployed on these advanced processes, as I was saying earlier, it's no good if we just say, "Well, I've designed my CPU. Implementing it is your problem. Off you go. Have a nice day." We long recognized that we have to worry a lot about that. We have to worry about how the CPU and the GPU connect together through our system components and ensure that we can deliver that in a way that can get implemented on these advanced processes. When we look at the 20/14 nanometer space, and we look at all of that technology that people want to put on chip, one of the other challenges that we have is the power scaling that we've enjoyed for the last 30 odd years has kind of run out of steam.
We're having to spend more time on design to really get the power out of these very complicated SoCs. We work in anticipation of these problems. We've engaged very early with the foundries on advanced process node. 20 nanometer, we've been running silicon for a number of years now to look at the challenges that come from advanced structures and to preempt the challenges that our customers are going to face when they come to do it in a real product. In this way, we can solve a lot of those problems up front and de-risk and increase the time to market for our customers when they come to, say, Cortex-A15 or our future 64-bit cores and put them on one of these advanced processes. We've been doing that with 20 nanometer for a number of years. Work on 14 nanometers is underway.
We believe we're in a strong position to provide that complete solution to the problem that we are solving. We aren't trying to solve every problem on the planet. That is what our customers do. They're going to take our building blocks, integrate them, implement them in a very power efficient, cost-effective way, and put their own creativity around the outside of that. We are gearing ourselves up to continue to be ready for when our customers move to these advanced process nodes. Around all of this, as I said earlier, is the ecosystem. What Arm is doing is very interesting to us. It is nothing without the work that we engage with across the entire industry. Our Arm Connected Community, with a membership of over 900 companies, is there to make sure that no matter what you're doing, help is on hand.
Whether you are building a chip, working with a foundry, getting taped out, whether you're designing software tools, you're designing software that's going to run on the chip, whether you're utilizing the operating system or running applications, that Arm Connected Community is helping you solve that problem. We view that as a really important thing. Nobody can do everything. We've never thought we could do everything, and that's certainly not going to change over time. Partnership is the way in which we're going to address the next round of technology challenges and do it in a business-efficient way so that everybody here can make money along the way. In summary, this disaggregated model, we believe, is the way forward.
Reaggregation, I don't think is going to help solve problems going forward, and I don't think is going to help lead to a high degree of diversity in the end products that consumers get access to as they walk down the high street. The way we've been working thus far has been a very successful way of structurally lowering the cost of doing design. When you can lower the cost, the volumes go up, more and more people can do it, and better designs are delivered at the end of the day. We think that's been an important thing for the last 21 years, and I think it's going to be an important thing for the next 21 years as well.
We've seen a vast array of products built around Arm, all of which have this key or leverage this key attributes of our products, low power, low cost, and utilizing the ecosystem. Whether you're building a chip for a dishwasher or a smartphone or DVD player, anti-lock brake system in cars, all of that is helping you do that in a very cost-effective way, and we believe that this is the way that we're going to address the next round of challenges. With that, I want to thank you. Thank you for coming today. Thank you for your attention. I'm going to hand over to Tom, who's going to talk about some of our roadmap for how we're delivering the technology into this next generation of devices. Thank you.
Thank you, Simon, and good morning, everyone. As Simon said, my name is Tom Cronk. I'm the Deputy General Manager of the Processor Division. As such, I'm responsible for the development of our processor technology and the development of that business in general. Not surprisingly, I'm going to talk about processors. Simon has talked about how we develop processors in a very symbiotic ecosystem, how we work together closely with people, and how effective that ecosystem is at both rapidly developing and perhaps more importantly, even rapidly deploying great technology. He also showed at a very high level how processors sit in modern-day system on chips. More and more of them embodying more and more of the functionality of the device comes from the processors. I'm actually going to speak, as he said, about another attribute of Arm and of our processors, which is energy-efficient design.
Ecosystems aside, low-power processor design and energy efficiency design has always been and will remain a prerequisite for the success of these systems. If you can't do that, you can't build successful systems. I'm going to use an analogy of DNA. We all know that DNA contains our genes. Our genes are contained in every single cell that we have. Our genes completely define the way we grow and the way we function. In that sense, it really is a good analogy, because I'm going to put forward a proposition that Arm has a low-power gene, an energy-efficient gene, and I'm hopefully going to show you how that manifests itself in the products that we build day to day and the difference that makes to the partnership and to the end products that are produced. Why do I say we have a low-power, energy-efficient gene?
We go back to the beginning. As Simon said, the company was formed 21 years ago. We had 12 engineers in a barn. This is a twofold effect here. First of all, the end market that the guys were designing the first processor cores for was PDAs, which was a new concept in those days, personal digital assistants. Those were battery-powered, and therefore, by definition, the processor had to be designed to work in an energy-constrained environment. The second effect was quite significant as well. It's more of an evolutionary effect or a pragmatic effect. 12 guys operating in a world which even then was a field full of giants. They basically had to produce a core that was small just to be able to do it. They were also constrained as a startup with budgets and things like that.
They had to find a way of doing it a different way. They had to find a different way of doing it. That led them to risk. Really, those two effects, part by design and part by necessity, really gave birth to the low-power gene. Time went on, as Simon said, phones became feature phones, became smartphones, became back to PDAs, and functionality got richer and richer. The low-power gene still prevails, and I'll explain how. The list at the bottom really is a hierarchical step to producing an end product on the Arm architecture. It starts at the top with a specification. Last year, we'll talk a bit more about this later, we defined the next generation of the Arm architecture, a specification we call the version 8 specification of the architecture, which happens to be a 64-bit implementation.
That is quite literally at the top level a paper specification. It describes the instructions that are included in the architecture. This is the instructions that the machines execute. It describes the way they interact, the way they interact with software, and the way you program the machine. Every single instruction that gets put into this architecture goes in with the conscious and subconscious thought, "What's the impact to power? Does this save power? Will this make efficient systems? Will it make it easy to build efficient systems?" That goes in right at the beginning, at the top level. Beyond that, we get to what we call the microarchitecture. These are the products that you know, the Cortex-A8, the Cortex-A9. That's a representation, a design based on the architecture. Again, at that level, the gene prevails. What's the impact of putting a register there?
What's the impact of putting a register here? Is this driving to an end low power system? On and on. The next part is worrying about integrating that design in a system. How's the processor talking to memory? Did it really need to move that piece of memory? Because if it did, that's going to cost power. These kind of trade-offs all the way through, low power. Implementation is how you actually take that design and produce it in silicon. Transistor structures, libraries, memory structures, all of these things, again, the low power gene prevails, and we go full circle. If you've done all of that on the way through, the guys that really bring these systems to life, the software developers, have the absolute best possible chance of producing low power and energy efficient systems. Why does it matter?
Why does this energy efficiency gene really matter? Of course, in the early days with the phones, as Simon said, it was simply about talk time or on time in the case of PDA. As time has moved on, it's been about increasing higher levels of functionality. Doing more for less because battery technology is not really evolved at the same rate as. In fact, it hasn't evolved at the same rate as semiconductor technology. All designs are still very energy constrained. Even designs that are plugged into the mains are energy constrained because when you design a low power system, it means it's about not having to put vents in the system, not having to put fans in the system, reducing the carbon footprint, or just basically, it's more simply about building cool products. The TV outside is a fantastic example.
The Samsung TV, if you get a chance to see it later. It's a beautiful thing. It's a 46-inch plasma TV. Sorry, plasma. It's an LED TV. No frame. It just looks great. It's better than that. It has textual control, so you can move your hand around and navigate through the menus. You can even talk to it. All of that is really only possible because of the very low power design in the system. You'll see the thing is only an inch and a half thick. Where are the electronics? They're hidden. They don't have to be cooled. It's all made possible because of the low power design. Another example on the Internet of Things, and one I find quite amusing as an engineer, is an example of one of our microcontrollers being used in construction.
You take a microcontroller, you take a little battery, you take an RF transmitter, and you package it into a little package the size of pea. You take a shovel, and you throw these into cement when it's being mixed. You will not find out. We're just checking if this is real or not. Okay. Two minutes? Okay. Okay, yeah. Carry on. Back to these peas. You throw them into cement, and then they basically detect the temperature of where they are, and they form a mesh network. When you're building big structures like dams, apparently, it's super critical to the strength of these things that the layers of cement, subsequent layers of cement, all get added at the right point. Through this mesh network of temperature sensors, you know exactly when the right point to add the next layer is.
All of these amazing ideas that are spawning out of the Internet of Things. Let's look quickly at a couple of the products. At the high-end today, and these products, we have the Cortex-A15, which really defines what we call low power computing. The A15 was publicly announced by us back in 2011, September timeframe. It delivers a really significant uplift in peak performance. If I use the A9 as the ARM9 processor, Cortex-A9, as a reference point. The ARM9, actually, there's two Cortex-A9s in the plasma TV I mentioned just before. Sorry, not plasma, LED TV I mentioned just before. ARM9 systems really define the best user experience today on tablets. The best tablets out there today are ARM9. It's a superb web browsing experience. A15 will give a significant uplift again on that performance point with tablets.
In fact, A15 will deliver a surplus of computing relative to the bandwidth that you can actually get through connecting the device today. You'll see new experiences and new apps and things coming on the back of that one. At the other end, the always-on, ultra-low power mobile category, we also recently announced the Cortex-A7. The Cortex-A7 is one-fifth of the size and delivers five times the energy efficiency of processors that are in most of the phones today. A massive step forward. Now, you can either use that to maintain the performance point and realize a cost saving. You can imagine entry-level phones tomorrow, based on this processor, delivering high-end phone performance from the high-end phones today, but at entry-level price points.
You can use it in another combination, which I'll show you in a moment, where you can get the best of massive battery life and peak performance. The other story that this slide shows is in the pictures on the right. The pictures on the right show two generations of process technology over three to four years. If you were just to rely on Moore's Law, you would expect the area of the device to shrink to 25%. Halving every 18 months, 25%. Actually, through smart design in the processor cores, we're actually achieving better than that with A7. With A7, we're down to one-tenth of the size that you were two generations ago in process technology.
Another good example of the low-power gene, the energy efficiency gene, coming to the fore was another announcement we made last year, which is this concept which we call big.LITTLE. big.LITTLE allows you to mix different sizes of processor. The Cortex-A15 and the Cortex-A7 both deliver fundamental different performance points and operate at different efficiency levels. From an architectural perspective, they are identical. Physically, they're quite different, but architecturally, they're identical. This means that software can run on either of these cores without having to be aware of the difference in the cores. You're also probably familiar, lots of end equipment manufacturers are today selling products on it's got two cores, or it's got four cores, this concept of MP. Today, those cores are the same. You'll have two Cortex-A9s, or you might have two Cortex-A8s or four Cortex-A8s, whatever.
This concept of big.LITTLE allows you to mix. I can have two Cortex-A7s, and I can have two Cortex-A15s, for example. You might wonder why. The reason is it allows you to really stretch the performance envelope. I can have a very high peak performance delivered by the A15, and yet I can have a very, very long battery life sustained on background tasks by having the software run on the A7. The software will migrate completely transparently between the two cores, depending on the need of that particular task. Clearly, if the Cortex-A7 has enough performance, it makes sense for the software to sit there, because you get this energy efficiency benefit. The chart shows how each of these cores save energy relative to a dual-core A9 implementation.
The blue shows you A15 with some very high-performance dependent tasks, and the green shows you A7 with some lower performance dependent tasks. It's interesting, you can see, I think this is a good example, actually that low performance point isn't actually that low. The A7 is more than capable of running the platform OS and a game like Angry Birds, for example, without you even needing to bring in an A15. Another great illustration of smart energy efficient design. Next two slides I'll just touch on briefly. The processor is part of the problem. It's a big part of the problem, but you mustn't forget the rest of the system. As I mentioned at the beginning, we think about the way the processor sits in the system, how it physically connects to the system, how it talks to other processors.
I gave you an illustration of that just now with big.LITTLE. We think about energy efficiency on all of these things. Consequently, we also, within the Processor Division, produce system, we call it system IP products. Memory controllers, interconnect technology. We actually produce and sell alongside the processors because we need to develop it to keep evolving the processors in the right direction. We also produce graphics processors. The same ethos prevails, and people talk about those in a moment. Back to the ecosystem message. Of course, it isn't just about Arm in this context. Actually, going back to the DNA analogy, the best of breed in the end comes from mixing DNA and mixing the best DNA, and that's the feature of this symbiotic ecosystem. The processors are important, but in the end, everything matters.
The implementation, the EDA, the operating systems, the software, it all needs to work together in a symbiotic way. One slide to the future. In March this year, we launched a product which we call the Cortex-M0+. This is the latest of our microcontroller offerings. This is the type of processor you would find in the cement example I used. This processor delivers 40% more energy than its predecessor, so that's Arm to Arm comparison, and it delivers greater than 2. In fact, it's 2 to 6 across the spectrum, so a greater than 2 times energy efficiency gain over any of the competitors in the field. As I said before, it's really fueling people's imagination in terms of applications. Another great example, you can just let your imagination go with this. Another great example I like is the umbrella, which I call the Take Me umbrella.
You leave this umbrella by your door. The umbrella's connected to the internet, and it has a proximity sensor. You go out the door. It knows it's going to rain, so it calls to you, "Take me, take me," as you go out. It can just go on and on with the things that you might want to invent. Moving on. Less of that. The Armv8, I mentioned that we released this architecture publicly last year. Ian, I think you're going to talk in more detail about it, aren't you, in the high-end context? This will drive us at a rapid pace into new market segments. We announced the spec last year. To date, we have four architecture licensees, so that's spec-level licensees, and four of our processor implementation licensees. Four architecture, four implementation licensees already on this.
We'll talk more about our products, Cortex-A, whatever we end up calling it, later this year. A quick recap. Arm was formed 21 years ago, with the ethos of energy-efficient design, low-power design processors. That was part by design, i.e., the market, and part through necessity. Small team, limited resource, had to think small, had to design efficient processors. That was then. The need in the market for low-power and efficient designs still prevails. In fact, it's an absolute necessity if you want to make successful end products. I really like the DNA of energy, as you've probably picked up, and I think big.LITTLE is just another fantastic example of how that low-power gene permeates through everything we do. Another important attribute is once you have that basic capability, you can achieve massive scaling.
I think we're in a unique position with the Arm architecture in being able to deliver these tiny microcontrollers right the way through to server-type machines, all based off fundamentally the same instruction set, same architecture. That, right on time, concludes the talk. Logistics, I think we now have a break. Is it 20 minutes? If you'd come back at 20 past, that would be much appreciated. Thank you.
Welcome back. My name's Pete Hutton. I'm the General Manager of our Media Processing division. What I'm going to do in the next 20 minutes is take you through how we've laid the foundations for graphics leadership over the last six years. Six years ago, we acquired a small company in Norway. Since then, we have built on that. I'm going to take you through what we've done. There we go. Firstly, why is graphics important? It's very important to consumers in terms of interactions with devices. You've seen some of the devices outside. Clearly, you have your tablets and your smartphones. Graphics is increasingly one of the things that drives consumer purchase of those devices. It's very important to our end customers. Graphics is a priority for anything with a screen. Smartphones, tablets, digital television, set-top box, personal navigation devices.
In the future, washing machines, printers, anything with a screen is a target for us. By 2016, there will be four billion internet-connected screens, all of which are a potential target for our graphics. It's a very compelling market for consumers. It's a very compelling market for customers. Clearly, it's a very compelling market for Arm. Four billion devices is a fairly attractive market to go after. This slide is probably the most important in the entire deck. I think My part of the deck, guys. Okay? Sorry. No, it's actually the most important. No. Anyway. I think we've been a bit reticent about saying where we are in terms of graphics. This slide does it. It's fairly busy, but let me walk you through it. We are the most widely licensed graphics processor available. We have 60 active licenses.
What I mean by an active license is a license that we think will generate units and will generate royalties. We have more licenses than this, but my IR colleagues regularly prune them if they think they're dead. We have 60 active licenses, and you can tell that 46 of those have been closed since 2009. If you're familiar with the Arm business model, you know that it takes a long time from license to get into royalties. The vast majority of those licenses that we have right now are not yet in products. They are not yet generating units. The wins we have are across some very high volume markets, and they're spread. We're not concentrated on one single market. We have smartphones, we have digital TVs, we have mobile computers. Very widely spread, very stable business.
Last quarter, we talked to you. We said we were number one in digital TVs. We're still number one in digital TVs. You can see some examples outside. We can confirm that this year we will be number one in Android tablets. Mali, in terms of graphics, will be number one in terms of Android tablets. I should clarify this. This is not just Android tablets that you will see reported in the West. This includes all China gray market tablets. I have an example in my hand. It's not a phone. It's my colleague's phone. Big phone. It's actually an Android tablet. This is a tablet designed in China, manufactured in China, largely sold in China. You won't see it. Well, actually, this one you will see now. It's available on Amazon, and it costs GBP 58. A very large market in China.
I think there's about 30 million units in the China gray market. We're number one in that space. We're not yet number one in Android smartphones. This year, we'll get to about 20%. Right now, I think we're about 15%, but by the end of the year, we'll probably be 20%. That'll put us second or third in the market behind, in this case, a proprietary GPU. For me, one of my major opportunities is actually proprietary GPUs. The great thing about all these licenses is you can see that they start to build to volume. 2010, we had about 3.5 million units shipping. Not that impressive. 2011, we had 12 partners shipping, and they shipped 48 million units. A very high growth.
This year, we'll get about 25 partners shipping, and we'll be in the triple figures, so over 100 million units. One of the things I said is we're leaders in DTV, and why are we leaders in DTV? Largely because a lot of the main OEMs have chosen our technology. Samsung and LG design their own chips in-house, put it into their digital televisions. A lot of the Chinese OEMs are also designing around our solutions. We have some very significant silicon partners in this space. We have MediaTek, MStar, ST, and Amlogic, who are building silicon, which is going into entry-level, mid-level, and high-end digital TVs and also set-top boxes. The reason we're very successful in this market is echoing one of the points that Tom talked about. We have the performance density leadership in graphics.
That means we have the best performance per millimeter or best performance per dollar or best performance per watt of any graphics solution out there. That's why we lead in digital TV. The other reason we lead is because on the graphics side, a lot of the success and a lot of the product is actually software. You have to have very mature software, you have to have very good integration and support around the software, and we do. We regularly get feedback from our customers that our software is very high quality and our support is excellent. I'm particularly proud of the support that we do, not just for our silicon partners, but for our end OEMs and their customers.
I'll show you later, we have a next generation product coming out this year, which will address larger DTV screen resolutions, so the 4K by 2K resolution. That's coming out later this year. In terms of mobile, I say we have a roadmap to leadership. We're not there yet. We are number one, or we will be number one this year in Android tablets, which is great. We do need to build momentum in smartphones. We do have some very nice flagship wins. The Samsung Galaxy S III that was launched earlier this month is a fantastic example. It's well ahead in terms of graphics performance of any other smartphone on the market. There's already 9 million pre-orders, including one of mine. Go buy it now. That would be excellent. That's a nice flagship product.
We also have quite a few semiconductor partners in this space as well. MediaTek, MStar, Spreadtrum, particularly in China, doing very nice low cost and high performance smartphone chips based around our technology. To echo one of Simon's points, we think we have the right technology for the right market. In all of the spaces, in the tablets, in the super phones, in the entry level, in the mid-level smartphones, we have a CPU and GPU combination which fits well in those spaces. One of the nice things about our technology and one of the nice things about the CPU technology is it's scalable. You can take the same basic graphics core, they scale from 1 to 4 cores, for example, or in our latest generation, 1 to 8 cores, and you can have a single core instantiation. It runs the same software.
It gives you all the same features, it gives you all the same OS support, or you can scale all the way up to eight cores and have the ultimate in performance. One of the changes we have made in our graphics roadmap is really we've identified over the last year that one size doesn't fit all. In the graphics space, the demands are bifurcating. It's very similar in concept to the way the processors went. On the CPU side, they actually trifurcated. Thank God that didn't happen to us. You now get the application, the real time, and the microcontroller CPUs. Exactly the same thing has happened on the graphics side. We now have two completely separate roadmaps, which are addressing two completely separate areas.
Just as you wouldn't take an A profile CPU and shove it into an M profile slot, you can't do the same with the GPUs. The two roadmaps are really graphics, so fairly easy to understand. Pure graphics and these guys, the customers there, they just want the ultimate in performance for the smallest cost and the smallest power. They don't want fantastic or OS coverage, and they don't want all the latest embraces that GPU compute complexity. The other roadmap is graphics and GPU computing. I'll talk to that in a minute. That's where you're using the GPU effectively as a processor. It's a parallel processor, but those are fairly complex, and they do take up a little more area. They do take up a little more power. The market has actually bifurcated. We do have two completely separate roadmaps now.
As I said, the easiest one to talk about is graphics. This is where you're using the GPU as a graphics processing unit. You're running games on it, you're doing user interfaces here. We have the best performance density in the market on this. That's really led by our Mali-200, Mali-400 ranges. The example I showed you earlier is a Mali-400 base device at single core, Mali-400 base device with Mali-400 scales up to 4 cores. Later on this year, as I said, we'll be launching Tyr. Tyr is aimed at higher resolution DTV displays and more complex smartphones. There are products on that roadmap beyond Tyr. I'm not going to talk about them today, but we have a continuation of the roadmap beyond the product we're releasing this year.
On the graphics and GPU compute side, we have spent the last 3 years developing a completely new ground-up architecture. What GPU compute does is it blends the parallelism, the multi-thread capability of GPUs with the control paths of CPUs. We've actually taken engineers from our CPU side or the Processor Division, put them into our graphics division, and come up with effectively a blended architecture. I think it's very difficult to do this if you don't actually have both capabilities in-house. It has taken us a long time. It was planned to take 3 years. It did take us 3 years, and it's one of the largest investments we've ever made as a company. What this now does is it enables completely new use cases to be run on the GPU.
The kind of things you can do on this are image recognition, you can do gesture recognition. People are using this for new and innovative use cases. Again, these products are out there. We released them the end of last year. I've had silicon in-house for a long time. We have optimized all the software on that, so we've optimized all the drivers, all the APIs, and all the operating systems. We will see the Mali-T604 and Mali-T658 shipping in consumer products second half of this year. I'm very excited about that. We actually have a second-generation set of products coming out later on this year, and then we have Skrymir, which is our third-generation product. That we'll release next year. We're already onto the second generation of this. Third generation is next year.
As I said, the kind of things you use GPU compute for are new use cases, new ways to interact with those devices. People are also using the fact that you have a complex GPU on board the chip to reduce cost in systems. You can reduce bill of materials by taking DSPs out, taking DSPs off the system, and putting that kind of complexity on the GPU. You can also use it for lowering power. Basically, the GPU compute is a wide parallel processing machine, so you can put the algorithms on that, run them across all the GPUs, take the voltage down, and you can reduce the power significantly. People are using it for that as well. Now, as Tom said, we have the low-power gene. We also have it in my division.
We have moved people from Processor Division across, so I'm sure there's an analogy with genes and passing. Anyway, you would expect our GPUs are industry-leading in balancing performance and power. That's great. Well, I would take that as read. What we've also done, though, is because as Arm, we're supplying the CPU, we're supplying the interconnect, we're supplying the memory controller, as well as the GPU, we've also optimized the system power. There's a number of techniques built into the GPUs which significantly optimize the system power. This is accessing external memory, which does cost an awful lot of performance, but it costs an awful lot of power. We've done an awful lot within the GPU to optimize system power and work very closely with all the Arm components. We've also started to work on what we're calling Graphics POPs.
If you're familiar with our physical IP side, you know we have Processor Optimization Packs, which are really aimed at getting the best performance out of processors. On the graphics side, we don't need the performance boost. What we are looking for is really an area in power optimization. It's optimizing a different path. We're bringing those out fairly soon. Clearly, we also have tooling with inside Arm. Our system division have tools which allow developers to see how their applications are running on real silicon, look at the power being consumed by the CPU, look at the power being consumed by the GPU, look at the power being consumed in the system, and then optimize around that. It's not just the GPU itself, it's the wider system, which actually is where a lot of the available power reductions are.
Ecosystem, we spent a lot of time on ecosystem. I got 30 engineers dedicated just to ecosystem support. That's working with gaming developers, user interface developers, apps developers, giving them free tools, giving them content, making sure that their applications and software are optimized on our graphics. It's a big investment for us. Clearly, you could have 300 engineers. You could have 3,000 engineers. It's like painting the Forth Road Bridge, but we think we have the balance just about right. Where we do have large numbers of engineers, or around about 250 engineers, is on the software side. We have a very big investment in terms of operating system support, in terms of APIs, and in terms of enabling external partners. You can see some of the main operating systems up there. You can see Android, Windows, pure Linux, Chrome.
We have others, Python, Nucleus, too many to fit on the chart itself. We have teams of engineers who are working on all of those areas and who are porting those operating systems to customer silicon and our own test silicon. Then the API side, you can see there's a myriad of APIs you have to support. DirectX, if you're supporting Microsoft Windows. GPU compute I talked about. It's a great thing. It enables fairly experienced programmers, but they don't have to be black belts to program the GPUs. There are a whole host of languages which support that, which we then have to support. There's OpenCL, there's RenderScript, there's DirectCompute. There's so many standards, it's really good.
We have a lot of engineers focused on the operating system side, a lot of engineers focusing on the API side, and then further out into the community. Those engineers will also work directly with our silicon partners and their end customers. We will send people out directly to our partners and customers to optimize the software on their platforms or to fix any issues they find, even in some cases if they're not our issues. I think we've done a pretty good job in laying the foundations for graphics leadership. We have very strong foundations in place, which we're going to build on. As I said, any product with a screen represents an opportunity for us. We are already in leadership positions in terms of DTV and Android devices. We'll ship about 100 million units this year.
Smartphones we continue to focus on will be about 20% of Android smartphones this year. We think we have the right technology for the right markets. We're getting that confirmation from our end customers. We have the best graphics performance density on the market, and we have a completely uncompromised support of GPU compute. One of the things we've done on the GPU compute side is made sure that it supports all the features that customers want, which has been fairly complex to get through, but we've managed to do it. I'm particularly proud that the next-generation technology ships in some very exciting products end of this year. Thank you. Ian.
Good morning. My name's Ian Ferguson. For the last four years, I've been running the server initiative for Arm. I've been maniacally focused on it for the last two years. Simon mentioned it earlier that there was a demonstration of the Calxeda technology next door. Public demonstrations of the Arm technology in the server space are starting to happen, and we felt this was an appropriate milestone to start sharing more of our vision and our strategies around what we're doing in this space with you. Simon mentioned earlier, but I really want to emphasize it, that I think for me, the key takeaway that I want to share with you is that like the mobile phone market previously, what we see happening in the server market is the emergence of highly integrated system-on-chip devices that are going to be very optimized for a specific set of server applications.
I'm going to talk to you for a few more minutes about why we see that happening, where we see that happening, and when we see that happening. Warren said this earlier that the data center area, which is really the focus of my presentation, is really the entry point for Arm technology into the server space. I personally believe, based on discussions with the ecosystem and end customers that I spend all my days talking to, that actually the opportunity for Arm is significantly broader than that, especially as we get to 64-bit technology that Tom talked about earlier. I will touch on some other areas later on in the presentation. Okay. On one of Simon's foils earlier, he talked about cloud computing, and he talked about really what we see there is companies where information technology or IT is the business. Okay?
What I mean by that is that the server infrastructure itself is the profit and loss generator for that business. There's a few examples up there. Some of them you might have heard of, Facebook, Google, yes. Tencent, Baidu, Alibaba, companies in China that are delivering similar social media technologies out there. These companies, because that server is the profit and loss generator in the business, they are very motivated to look at new technologies that will help them make more money. Okay? Why is that important for us? Well, the way they've set up their businesses is to be very mobile to look at and be adept at evaluating new technology. What I mean by that is they write their software in high-level languages. I had a question in the break about legacy code, for example, compared to incumbent architectures.
In the data center area, very little of that. High-level code, as Simon said, people are specifically looking at one or two workloads. It is not a server where you have to run 50 different things. They write high-level code, whether it is Java, C++, whatever. Very portable. The other thing that these guys look at is that software is either residing in themselves, Facebook have their own specific libraries, or Google have their own libraries, or they use open source. If they see a TCO benefit to migrate to a new technology, it is inside their own control on how they get there. They are not waiting for a database from a third-party software company to go port it. It is under their own control. I have brought up a picture there of Facebook's data center in North Carolina. There is a parking lot.
Sorry, I have been in the U.S. for 13 years. Car park in the lower end there. You can see some small vehicles there. You know how big vehicles are in the U.S., so that has got to be a really big building, right? As we go forward, what we see and what our customers are seeing is these things are energy constraint. As Simon said, as Tom said, we have spent 21 years understanding energy constraint systems. This is a battery base, but when you have 10 megawatts and that is what your business is run around, it is energy constrained, it forces you to think of different ways on how you are going to solve that problem. You cannot just look at a pure performance vector. A couple of examples about what Facebook has been doing. They build these buildings. They have got some in Luleå.
They have got one in Oregon. As I say, this one is in North Carolina. The way they cool these systems, radically different from how these things have been done in the past. They are also looking at the structure of these boxes and saying, "Do I need everything on this board?" The old traditional servers that have these, what they call vanity cases, the plastics that go around it actually blocks airflow. Facebook have said, "We do not want that stuff." It just is an unnecessary cost, unnecessary recycling thing when we roll it out. They are looking at how they drive down cost. They are looking at how you can replace these systems more easily in the field when they fail. They are looking at how you cable this thing in a cheaper way. They are driving a standard called the Open Compute Project.
It is for hardware that Facebook will use, but they are also broadening it out. Again, looking at driving server hardware to a standard to drive volume, to drive down cost, to Simon's point earlier. Some of Arm partners, Applied Micro, were announced a few weeks ago as getting involved in that. It is a processor-agnostic standard. We will see where it goes. Again, an opportunity where people are looking at doing things differently because they are in an energy-constrained system. Let us look a little bit more about some of these workloads. Walk through this fairly complicated chart on the right-hand side. This came from HP when they did their announcement with Calxeda back in November last year, a project they called Project Moonshot.
What they have basically done is done some analysis on different types of workloads and compared how do those workloads run on incumbent server architectures as compared to running on microservers. They look at those workloads on three metrics, cost, power, and space. Okay. The way you read this is if it's to the left of the line, it's a win for your incumbent architectures. For example, if you look at that in compute-intensive area, if you have an application that needs a lot of performance, Simon was mentioning earlier about the weather forecasting over Tokyo, that's a very mathematically intensive platform, excuse me. Now, you could build that out of microservers, but you're going to need a lot of microservers down there.
What you see there is that the cost of that is going to be something where it's advantageous to use your traditional way of building servers. A couple of takeaways from this. We do see a different set of tasks and different variability in terms of usefulness of microservers versus incumbents. Indeed, where HP is spending most of its time is around those sort of what they call light scale-out applications. What do we mean by light scale-out? Modest CPU compute, a task that is largely parallelizable. For example, on web servers, if a node of a server is working on somebody's website, it doesn't really need to know anything about the website next to it. Actually, you'd rather that it doesn't know anything about the website that's running on the same thing. Largely parallelizable.
Again, here, as Simon mentioned, it's just running one or two things. If you were to go into Facebook's structure, they have servers that just do one thing, whether it's the front end reading of a query and working out where it's going to go and directing a query, whether it's what I call Memcache, or the industry calls Memcache, where you just have these big servers filled with memory because you don't want the delays going to storage. They just do one thing. To Simon's point, when you know it just does one thing, there's the opportunity to go and integrate the appropriate level of compute, networking, and storage around that particular application. It's a space where 32-bit is an entry point. Depending on where you go, some people will need 64-bit. In this space, it's largely about addressing.
This isn't, again, a place where you need massive compute for these types of tasks. Some people have written their applications where they need a lot of memory space. Okay. Let's just talk a little bit about some of the light scale-out applications. I view them as mixing of compute, networking, and storage. Really the magic is how do you find the right balance between those three? If you're using cheap disk, spinning media in Facebook, you don't need a massive CPU. Put it another way. You need a CPU that utilizes that hard disk to a high level of efficiency. You don't need something that has way more horsepower than can actually read the data off the disk. Okay. How do you balance those three things inside a system? Compute can be pure CPUs.
It can be GPGPU compute from sort of Pete's Media Processing division there as MPD start to drive that technology into some broader areas beyond mobile. Or it could be hardware accelerators. People look at certain algorithms and can accelerate those more efficiently in hardware than doing it in software. Okay. One of the interesting areas that I have seen recently is around this whole Hadoop term. Not sure how many of you know about Hadoop, but it's basically a search algorithm. Originally came out, the MapReduce technology came out of Yahoo. Yes, it's used for search queries in your Google search or Bing or whatever. The underpinnings of how it goes and works out what you need is using those algorithms. It's also used by financial organizations as they start to do queries into databases.
Maybe not the transaction itself when you buy or sell a stock, but as you build up data over days, months, years, or big data, the analysts are looking for searching for patterns, right? They're looking for, am I going to buy this stock based on this? That type of offline data analytics, we feel one of the main places where the initial deployments will occur. One of the interesting trends in the financial area is this desire towards real-time Hadoop. People can't wait 20 minutes for a thing or wait for a batch job overnight. This is an area where hardware accelerators or potentially V8 hardware can actually go in because it will provide more performance and reduce the latency to getting those queries back. The other area of differentiation is around the rest of the SoC.
As Simon said, we feel very passionate about what we do, but we have to enable our partners to innovate and differentiate. Some people are using the cost advantage of the high volume tablet chips and driving that into this sort of server space. Other people are integrating far more server-specific functionality down onto the device to provide more robustness, reliability, and a more optimized solution. Again, you can talk with details at Calxeda's technology next door. Broad set of applications, different performance points, different IO, and what that leads to is diversity of devices. I'm just showing three here, from Calxeda, TI, and Applied Micro. To the far right, you have a device that Applied Micro started to talk about, which is 64-bit, up to 32 processors on an SoC, 3 gigahertz per processor, very high-end, 10 Gigabyte internet integrated on, SATA integrated on.
A lot of integration. At the other end, you have the device that you see next door, and in 5 watts, you have a quad-core 32-bit processor. Very modest, and they've taken this completely different performance and power and integration point. Over time, I could have added Marvell to this. As Tom talked about, we have architecture licensees on V8. We have partners looking at our Apollo cores and Atlas cores that you'll hear more about later this year, which are our first 64-bit cores, looking at getting into this space with Arm-based technology. James McNiven showed this fall last year, for those of you who were here last year, and talked about how we go about building software ecosystems, or let's say ecosystems in general. Talked about how it takes a long time to build these ecosystems. It starts and then evolves over many years.
If we go forward to today, I think we've made some pretty good progress. There are now all of the critical software pieces you need in place to be able to ship a 32-bit server platform. We will see server shipments this year, albeit in limited quantities, but boxes will start to ship this year for the sorts of applications I described earlier. Web serving, Memcached, content delivery networks, Hadoop types of areas. Things like a server-grade Linux technology. Things like a good performance optimized Java. Why is that important? If you remember, I was saying that these data center guys write their software in high-level languages, so it's more portable. Hadoop has a very large Java component on it, so it's important for us to get a performance optimized Java compiler there.
Calxeda, the boxes next door, they've been running their website for the last few weeks on that technology. You're going to see the demo next door. They also showed a number of applications at the developer conference a few weeks ago. They were showing a set of software called OpenStack. That is actually, if you think back to what I said about the Open Compute Project, that is the software stack that Facebook is looking to drive as an open set of technology to go and drive critical mass around these data center solutions. The specific applications, there was something called WordPress and Node.js. I'm happy to have discussions about what they are. It's less important about what they are. Really what was important was that they just worked out of the box.
They took the technology in that open source, put it on their OpenStack box, and it just ran on Arm. Again, in this space, relatively little ties to legacy incumbent code. Now, as we go forward, as I mentioned earlier, 64-bit is a place where we actually see the opportunity for Arm broadening. Our focus remains in the data center area for the beachhead, but as we get to 64-bit technology, it allows us to go and address other markets that are concerned with energy efficiency or space constraints. Again, energy constraint systems starting in data centers, but areas of high performance computing. Some areas of enterprises. If you think about New York, where they're less about power constraints. It's a little bit about power constraint, but it's about space.
People that are wanting a private cloud, not wanting to necessarily put all of their information out into public clouds like Amazon, but you've only got a small space in your New York Stock Exchange area or in Japan. How do you go and cram more density around that particular problem? We see opportunities there. Emerging markets. Simon talked earlier, I think, about how we see smartphones being adopted, not just in the U.S., but very broadly out into emerging markets. How does that technology go out there in spaces? How do you serve that technology at the other end of the wire in places where there's limited power or very unreliable power? Again, real cost constraints. We see some massive opportunities there where people can rethink servers based on solving these problems. We start the software ecosystem now.
We've been working with Applied Micro, who's one of the pioneers in this space with us, one of our silicon partners. This is one of their FPGA boards down in the low-end corner there. This is a board that's software compatible with what they have coming down the road in terms of real silicon chips, but it allows us to get that stuff out into software partners and get that software ready for when they have devices. To be frank, when other of our V8 partners have their devices too. We intersect their chips with software, and we move forward from there. Just make sure I've got everything. Yeah, I've got everything. Really to sum up, we see servers as increasingly becoming regarded as an energy constraint problem. It's starting in the data center.
I think Arm's applicability into the server market over time will be tied with the fact that more markets will start to view their server challenges as being energy constrained. Why is that important? That's a fundamental thing where people then have to rethink how they build that server technology. Okay? They have to start thinking about what they put in the hardware. They have to think about how they go and integrate to piece point. The more you put onto the device, the less off-chip accesses, you're saving power. Really this will be the rise of highly integrated SoCs for server markets. Like I said earlier, we have a strong track record of playing in energy constraint systems.
We want to put a little bit of guidelines onto here, and I think Tim will talk to you a little bit more about the financial modeling to help you with your different things here. The market size, there's a typo here. The market in terms of deployed servers out there's about 50 million servers out in the marketplace today. Okay? Less than 10% of that is currently in the data center area. That is the area experiencing the most explosive growth. We expect that to be somewhere in the 20%-25% size of the overall server market by the year 2050. 32-bit Arm-based servers are gonna ship this year. We've set the expectation that it's gonna take time, right? We're at the early stages. Hardware will go out there.
We will find some beachheads, it's really gonna be several years from now before you see meaningful shipments. What we're lining up to do is see the first 64-bit servers shipping in the 2014 timeframe. As I said, we're building a software ecosystem right now to intersect those platforms that are gonna be coming out in that timeframe. In terms of ASP, a theme that you probably heard through all of these presentations, I think, is one size does not fit all. We see some people looking to use tablet technology, and those are gonna be devices that are relatively cheap. We're seeing other people, if you look at that Applied Micro device, very high set of functionality in terms of cores, in terms of IO. You're gonna see a range of ASP points. We are expecting a range in the $50-$200.
If you need some more granularity on that, work with myself or Ian Thornton. With that, I'll hand over to Tim. Thank you very much.
It's almost morning. Finishing, it's almost afternoon. I'm on the cusp, and we should be about 15, 20 minutes away from Q&A. Long-term growth opportunity. I think pretty much everyone in this room sort of lives in a world of fluctuating macroeconomic news every day where we're trying to make sense of it, and the world tends to look a different place every morning we wake up. We also operate in an industry, you guys invest in and analyze an industry that's characterized by 24-hour news flow. Again, it can lead to wild extrapolations.
I think it's very good for us, once a year, away from the glare of the quarterly or half-year or full-year results, to be able to focus on Arm, our business model, what that business model drives in terms of industry economics, our low power technology, our progress in graphics, and some of our longer term product opportunities. What I'm gonna spend a few minutes doing is try and draw that together as we think about the financial modeling of Arm out into the future. I probably haven't done this in this forum since 2007, when I did talk about some sort of longer term shapes of the Arm business model. Really, that's what we're gonna do.
In a bit more detail, just remind ourselves of what we've been building for the last 20 odd years in terms of installed license base, and where today's royalties and how today's royalties relate to that installed license base. We're gonna look at the acceleration, if you like, of the pace at which that installed base has been built in recent quarters and recent couple of years, and what that might mean for future royalty and for future share gains for Arm. Most of you who look at our segmentation slides will be aware that we currently have about a 30% share of the total embedded processor market. Maybe we can think today about how that might develop over the next few years. As we increase that share, which is a unit share, what is happening to Arm's royalty % per chip?
Not average royalty rates across the blended space. That's interesting, but it's just a mathematical artifact. What really matters is how much value is Arm bringing to each of the chips where we're designed in. Drawing that all together, what does it mean for the P&L? What does it mean for license revenue growth, royalty growth? Talk a little bit about our cost base so we can understand how the operating margin develops and the earnings develop. I don't usually get into detail about one particular line on the P&L, but I think with the tax changes that are going on in the U.K. jurisdiction, I think it's important to understand as we look out at our modeling five or 10 years, what's actually happening to the tax rate. That's changing quite significantly over the next year or two.
As I say, here's a reminder of where we are today. We've signed 870 licenses, just over 300 companies. Most of those companies are going to end up paying our royalties in due course. About half of them do today. You will see there in the box that we've signed 320 of those 870 in the last three and a quarter years, they're not really moving the dial. In fact, in 2011, moving the dial in terms of royalty. 2011, 99.5% of royalties that we reported were generated from the licenses signed before 2009. There's a lot of pent-up royalty already, if you like, with the building blocks put in place. Interestingly, of those 320 licenses signed in the last three and a quarter years, 80% of them are Cortex and Mali.
Of those 320, 25% of them are either Cortex-A or Mali, and therefore typically characterized by a higher percentage royalty per chip than we have traditionally been used to with Arm at the sort of 1% plus. We're now moving, as we discussed in recent quarterly presentations, beyond that, and we'll look at that. What this license base has driven so far is, or in the last 10 years at least, a royalty CAGR of 24%. Obviously, well ahead of the industry rate, and we'll look at that in a bit more detail as well.
Again, followers of Arm who track quarter to quarter and try and develop direction about how this investment proposition unfolds will, I think, have become used to this slide, which is our attempt to share with you what designs, what licenses, and therefore what designs into how many semiconductor companies are important for us to increase our market share. What that chart is basically showing you, those round blobs represent semiconductor companies. We believe that we need to turn all of those blobs blue to have an 80% plus share in each of our target markets. You can see from there, as you would expect in something like smartphone application processors, it's mainly a blue picture. Some of them are a combination of blue and green. Green is these companies are shipping some Arm-based chips.
Yellow is companies that have announced designs on Arm that may not yet be shipping. The reds are the things we need to go after to get them Arm shaped. On the right there, the 2011 share shipments, that is by segment analysis of the 30% that I referred to earlier. Just a little bit of a color work here for those who are bored with reading words and looking at numbers. If you take away the segments and just look at the blobs, you've got 108 major designs there that we need to turn ultimately blue. From different shades of red towards blue. You can see that half of them, almost, are pretty much there. Again, putting that in a easy-to-understand grid, that's how it looks at the moment.
Our goal is to get that nice blue shade moving from left to right. We believe that looking at and talking to our customers and our customer's customers, talking to our in-house teams and our commercial folk, we believe that in 2016, that grid looks something like that. 49 blues have turned into 62. There are only six reds hanging out there that we need to go and turn. Clearly, this is a bit of crystal ball work, and it's directional, but this is actually based on our best estimate of how the market is thinking about deploying Arm technology over this period. Licensing is obviously a precursor to share gain.
It's actually a little bit better than that because, as most of you will know, many of our licenses are perpetual licenses, which means that you take a license to an Arm design, and you can design it into your chip for as long as you are prepared to pay royalties. Therefore, for Arm to be generating a new royalty opportunity, you don't necessarily need to see a new license. You just need to see semiconductor companies getting more leverage and deploying existing licenses more. You will see more licensing. 2016, we put in their market share 40%-50%. Sounds quite hand-wavy. If you look at 2011, I said 30%. Five years before, it was 17%. It's been growing at, on average, around about 3% per annum. A little bit quicker, actually, in the last year or so.
I think with the opportunity, the very high volume opportunity in microcontrollers, there is a reasonable argument to say that our increase in penetration is going to grow at a rate that is higher than we have seen historically on a volume basis because of microcontrollers. We see a world where, let's say 3%. 3%-4% over the next five years, we can see our 30% going into that 40%-50% range. This, of course, is an increasing share of markets which in themselves are growing and will continue to grow in this period. Actually, if you look out five years beyond that, which I think you need to do when thinking about Arm, we see a world where those markets actually continue to grow.
Some of the areas where we have a very, very high penetration to date, given competitive environment, et cetera, we expect the market share to be flatter. Basically across most of those end markets, we see Arm continue to grow looking way out. I didn't want to leave you with the impression that once everything was turned blue or green, it was game over. Actually, the world is a very dynamic place, and you can see that some of the items on the right there, some of them have been touched on this morning, others not in much detail. The things that we currently include in the segment chart that we show you, that is not the end game by any means. Every time Ian and Jonathan update that, new things get included.
Not just because of their own personal decision, but because there are new products out there that are capable of deploying our sort of technology. I think there's going to be much more opportunity when we get out to 2016. As we increase this unit market share, what is happening to the value per chip? I mean, fundamentally, this is an outsourcing business. We are substituting for our customers fixed costs in their business by way of engineering headcount, largely, with variable cost in the form of licensing and royalty. Basically, the more sophisticated the processors, the more work needs to be done by our customers, would be needed to be done by our customers to do it. Therefore, the more value we are saving them by providing an outsourced option.
In that environment, our customers understand that if there's more cost for us and more value to them, that they will pay more royalties. This is why that graph is the shape it is. As we've moved into a world of complex processors, A-class processors, moved into a world of multiple processors like the big.LITTLE concept that Simon talked through, these are attracting higher royalty rates. Five years ago I was sitting here, we'd be thinking 1% plus for Arm. Now we're thinking for the general purpose processor itself, 2% plus or minus. We're talking about a world of graphics where Pete's explained the trajectory we were on. That typically brings another 1% on top. We've talked about attach rates of physical IP and optimization packages. That brings a further royalty again.
The next generation V8 will be for the same value reasons, looking to continue the trajectory. As we look further out, you are seeing a world of Arm getting much more value per chip. Before I look at the overall picture, just a reminder on what is actually going on with the tax environment in the U.K. We're forecasting about a 25% normalized effective tax rate in 2012. The U.K. corporation tax rate, as I'm sure most of you know, is already being legislated down. It dropped down in April 2012. It's going down again in April 2013. It's going down again in April 2014 to 22%. More importantly for Arm, the Patent Box tax regime is being introduced from April 2013.
Fundamentally what that means is that for profits arising from qualifying patents or qualifying profits arising from patents, they will be taxed at 10% rate. Clearly, there's a lot of devil in the detail about what specifically constitutes a relevant patent and a qualifying profit. Suffice to say, this legislation is aimed at companies like Arm, who obviously the U.K. government wants to encourage to invest in this country rather than all of the other places around the world that Arm could invest in given the talent pools that are around the world. It comes in in April 2013. It is being implemented on a transitional basis. The total benefit that a company like Arm will get in the end, 60% of that benefit happens in year one, and the remaining 40% happens 10% per annum for the next four years.
The little graph on the bottom is not a very sophisticated way of showing that the tax rate is going down without putting specific year-on-year numbers, because we all know there are lots more things that go into a tax rate on an annual basis than just this one. If I was to put one more level of science on that graph, I would probably have the rate going down a bit more in 2013 and 2014 for the reasons I just said about 60%, and then flattening thereafter. Essentially, as I said on the Q1 earnings call, in five years' time, Arm's tax rate should be, based on what we know today, sub 20%. Long-term growth opportunity. 2007, I stood up here and painted a picture of license revenue growth that would be mid to high single digits.
In fact, in the five years prior to the downturn, CAGR of license revenue was 9%. I also painted a picture of royalties growing at broadly 2x the rate of licensing and typically well ahead of the industry. Mid-teens or a little bit more. What's actually happened since we talked through that is that licensing has actually grown. Looking back over eight years and at a 14% compound growth rate, obviously quite a lot of that pickup has happened in the last two and a quarter years, where partly due to bounce back out of the downturn, but partly because of the increasing utilization of Arm technology across these broadening end markets we've been talking about. Licensing has been growing faster than that. Royalty has been growing, as we say there. I mean, the industry over that period has been growing at about 7% per annum.
We've grown at 22% on the royalty, at the top end of our 10%-15% range. From an operating margin standpoint, over that period, we've gone from 20%-45%. Leading up to the downturn, Arm had a margin in the very early 30s for three or four years. We've now seen a significant injection. Partly, of course, because of the higher than trend run rate license growth we've seen in recent periods, but also because of this royalty growth. That was all driven, an earnings CAGR in the last eight years of 27%. Where do we go over the next eight years? We see a world, again We've been consistently guiding license revenue growth to still be in that sort of 5%-10% on top of this higher base that's been built rapidly in the last two and a quarter years.
I think when we started growing licensing at 30%-40%, the main question that I was getting was, is this sustainable? Is it pent-up demand out of the downturn? Is it everything coming together in a glorious fashion, and it's actually going to ease down? The answer is no. We expect trend license revenue growth on top of the base that we have built. We also expect to see our royalty growth, our royalty revenue grow in a similar way to the way it has in the past, 10%-15% higher than industry growth. You can see in the last eight years, it's been at the 15% end of that range. That's what we see looking out longer term.
From a margin standpoint, we have this conversation obviously a lot with investors about where is this 2007, I stood up here and said, when our margin was 31%-32%, that this business is capable of sustainably supporting margins of 40% and above. Of course, everyone said, "When?" I said, "In the medium term." Turned out I was right. In 2010, we went through 40%. Maybe we would've gone through a little bit quicker if it hadn't been a downturn. Don't know. 2011, 45%. Of course, everyone asks us now, what is it going to be in five years' time? I think one thing we can be fairly sure of is that it goes up into the right. I think the question is it 50%? Is it 55%? Is it 60%?
Will really depend on partly, of course, the rate of penetration of these markets that we've been outlining, but also what are our investment opportunities in the out years to develop more technology to generate more licensing and more royalty. We're not managing this business for the highest possible margin in the short term. We're managing this business to be able to grow our overall profits and cash flow optimally, and that may well be a more likely shape of a 50%-plus margin rather than a 60% margin. We'll have to see. There's nothing that we can see in terms of the trajectory of the cost base, and the need to invest that is different from the model that we've painted before. R&D costs are going to go up in absolute terms.
You've seen us recruiting about 10% per annum in headcount in the last two and a bit years. We're going to continue to invest as necessary to access the opportunity. That is all within the overall shape. You're going to see periods of Arm where we are quite flat on our headcount. For three years, 2007, 2008, and 2009, our headcount was flat. We had a big year of investment in 2006. We've had investment in 2010, 2011, and the start of 2012. You're going to see it go a little bit in periods. Generally speaking, there's no change to the relationship between our cost base and these revenue projections. That means that our margin continues to grow over time.
From an earnings standpoint, that revenue growth, that margin enhancement, and of course, the boost that we're going to be getting from the tax rate, I think sets a good foundation for our earnings going forward to be consistent with what we've achieved in the previous eight years. That's kind of how I would update what I said in 2007. In summary, before we move to Q&A, the reach of Arm technology is broadening rapidly. We know that. It's been driving licensing in recent quarters. The guys have outlined some of the markets that are underpinning that. The installed base has been growing steadily for a long time. That growth has accelerated in recent years. We're now reporting license revenue at 2x the level that we did two years ago.
The order backlog, which is a contractual order backlog, it's not a discretionary drawdown item, is more than 2x what it was two years ago. That underpins continuation of that trend. That is driving long-term royalty opportunity. We're increasing the value because we're bringing more to the party. The government is being helpful. All of those things are driving a very promising outlook for our earnings growth over the next few years. Thank you.
Thanks, Tim. We are going to do some Q&A in a moment. This is slide 69 in the pack. Obviously, 69 slides is quite a lot to remember. If you just had to take away a few slides from the pack, then we'd say these are the four key messages to take away. The first is the importance of Arm's partnership business model. That's a key differentiator. That's what delivers the innovation and the economic benefits. We have to partner with a huge range of different companies, right from manufacturing technology to high-level operating systems. The next thing to take away is that the growth of this business comes from, amongst other things, engaging with new partners, new customers, new markets.
There's a huge range there, whether it's changing the way people mix their concrete, and we have smart, intelligent concrete, right through to changing the way people design their servers going forwards. Whichever end of that spectrum it is, we're bringing Arm's low power DNA to the party, and that delivers greater benefits to our partners and their customers. The example on the slide there, it's a big.LITTLE slide, an example of the system design, but we could have also chosen other slides out of the pack where we're bringing our low power DNA. Start over in the front left corner there. We've got some microphones going round, and we'll try and keep up the pace.
Hi. If I go back to Simon's presentation, slide number six, where you describe the multi-decade disaggregation of the semiconductor industry. Obviously, a very long-term trend. If you look at the current, the two big successful handset OEMs, Apple and Samsung, there's an element of going back to vertical integration, both with application processors and in Samsung's case, of course, also memory, and I'm told they're even interested in baseband. How should we look at-
Start over in the front left corner there. We've got some microphones going round, we'll try and keep up the pace.
Hi. If I go back to Simon's presentation, slide number six, where you describe the multi-decade disaggregation of the semiconductor industry. Obviously, a very long-term trend. If you look at the current, the two big successful handset OEMs, Apple and Samsung, there's an element of going back to vertical integration, both with application processors and in Samsung's case, of course, also memory, and I'm told they're even interested in baseband. How should we look at those two features in the context of the long-term trend?
I think to every rule there are exceptions, of course. I think the thing to look at is to get the most out of this, those who are going to lead are going to understand the complete supply chain from top to bottom. I think that's the really important thing. It isn't just a case of taking what suppliers give you, bolting it together and shipping out a product. It's about understanding how the software works and how the transistors are made and how to get the most out of that all the way along. Sure, there are going to be some companies that do integrate. Some of those people that you mentioned manufacture, others don't. The key for me, I think, is understanding the whole supply chain top to bottom.
Whether you have to do it all yourself or not, I don't think you do to get the most out of the cost benefits of disaggregation. I think to get the best solution, you do have to understand every step along the way.
Last one.
Yeah, thanks. Francois from Morgan Stanley. The first question, and I would say all, the question is about production capacity. Will the Arm partners have enough production capacity to produce chips based on Arm at 28 nanometer and below? Is it the case that at some point they will have to beg Intel to get their chips at a much higher price? That's the first question. The second question is about Windows 8. Yeah, we are there again. Very simple question. Does it work? Does it work well? What's the incentive for OEMs to use Arm over Intel for Windows 8 tablets in particular? Is it just power consumption? Is it because they will generate profits? With Intel, they won't generate any? If you could elaborate on those two questions on Windows 8.
Okay. Let me talk about Windows 8 while Simon's coming up with an answer on capacity. Windows 8, this is important to say, this is a Microsoft product. Microsoft are controlling the launch of this product. From what we've seen of the product, we have played with the product, of course it works. It works very well. It's a nice operating system, I'm sure that they'll find many customers who want to use it. In terms of advantages of Arm versus incumbent designs that already use the Microsoft operating system, yes, low power is absolutely an advantage in these thin form factor products like, for instance, tablets, like thin clamshells. Keeping the electronics small is very important. Tom cited the example of the very thin television out there as well.
Another example where keeping the electronics small is important and low power is the key to that. I think we also talked about the business model delivering economic benefits, we believe that the innovative yet cost-competitive supply environment that comes from the Arm business model will certainly be advantageous for people who are building Windows-based products. Those people at the moment, of course, don't have the benefit of that supply environment. It's going to be a new thing for those manufacturers, and that's certainly a benefit that they'll be getting from it. Your other question was about manufacturing capacity. If you have any comments on that.
Yes, I think that's a simple answer to your question is yes, I believe the Arm partners are going to get enough capacity for 28 nanometers and beyond. Whatever issues exist today are going to get solved, there's a lot of capacity being put in place around the world by many different companies to fulfill demands for the future.
Yes. Thank you. Sandeep is next.
Hi. Good morning. Sandeep Deshpande, J.P. Morgan Cazenove. Firstly, a question to Pete on graphics. You've talked about these two different streams of graphics processors that you are getting into. You've had a key market share in the TV graphics market, that it's having these two different streams is going to help you in the handset smartphone market to gain share. Then a follow-on to that would be, how should we be modeling the royalty rate in graphics? Is it a different way from how you do it in the microprocessor? How should we be looking at if, given that those products will have Arm processors as well, will there be discounts on the graphics processors or graphics processor when an Arm-based processor or multiple Arms are on that same chip?
Let me try and answer both of those. In terms of smartphone gain share, you asked about the two different roadmaps we have. Yes. You can see smartphones are segmenting superphones. There's mid-level phones. There's entry-level smartphones. There are smartphones at all points, and having the widespread of products enables us to target all of those markets. You wouldn't take one of our top-end GPU compute cores and put it into an entry-level smartphone. You're not going to get the benefits out of that, and you're going to get some overhead. Yes, it'll help us gain share in smartphones. In terms of how you model it's just the same as the processor. It's exactly the same financial arrangement. It's exactly the same kind of licensing arrangement, and as Tim said, you get royalties for the processors, you get royalties for the graphics.
Just clarifying that because on the processor, as we understand that subsequent processors get discounts on the processor. Would this be classified as a subsequent processor on the chip?
No.
Short answer, no. It would not be. It's an incremental royalty as presented on the slide in Tim's section. I think slide 64 in the pack. We're also sticking to the principle, and we're starting to illustrate that in slide 64 on the pack as well, that more functionality in the graphics processor means more value added by the graphics processor. If there's more value added, we're replacing a higher cost that would otherwise have to be done by an internal design or from an alternative external supplier. That means an opportunity for increases in the incremental royalty as well as the base incremental royalty from the graphics processor. Let's have the next one. We need to keep cracking through these questions because it's quite a lot towards the back of the room.
Gunnar Plagge from Citi. Could you talk a little bit more about your structural power advantages, particularly this idea of combining different cores? To what extent are there barriers to entry, and do you see any adoption of this in your competitive landscape? Secondly, from the moment you've developed processor architecture to putting it into silicon, and I think you talked about this, that you're working with foundries at 20 nanometer, 40 nanometer. You're developing a lot of know-how, and I was wondering, you're monetizing this at the moment mainly through sometimes hard macros, mainly through these Processor Optimization Packs. Do you really have the feeling that you efficiently monetize this know-how, or are there any better ways and new business ideas? Thank you.
Okay. Tom, do you want to do the first one, Simon have a crack at the second one?
The big.LITTLE concept and I think the question was really are there any technological barriers? The answer is honestly, no. That's one of the really smart things about it because today's phones embrace two things. They already embrace MP cores, multiple processors. The other thing they already embrace is they have the intelligence to do this thing called dynamic frequency and voltage scaling. As the software load goes down, the systems back off. That basic control capability is all you need to enable the big.LITTLE. That and the MP capability are the enablers. They're already there, and the software is over and above that, completely unconscious of the switching that's going on. There are no barriers. To the question of adoption, I think to date, 11 partners have both the A7 and the A15, product is in the pipe and coming through.
We have silicon in-house now. It's just a question of a few months, and things will start to emerge, I think.
Simon, you going to?
On the monetization of physical IP, I think there's two things to look at. The Processor Optimization Pack really does help a lot of our customers get to market sooner. That pulls through a royalty stream that we get on our processes earlier. If we can help more of our customers deliver higher performance and lower power, have more competitive products, that is just generally good for Arm's economics. In terms of the actual pricing, the way we do the Processor Optimization Pack, we've had these products in the market a little while now, and it feels like we've got the level of license fees, the way the royalty works about right to ensure that we do get good uptake and continue good relationships with the foundries. So happy with how that is working.
Yep.
Hi, it's Didier Scemama from Merrill Lynch. Couple of questions. Maybe first question is related to the above 90% market share by 2016 in smartphone processors. What I'm wondering is what makes you comfortable putting this target or even this prediction out there given the notoriously short cycle for handsets and obviously the entry of Intel in this marketplace. The second question is related to the PD licensing CAGR in the high single-digit. What are the reasons behind that number, and can you talk whether the potential for licensing with the analog and sensor companies are encapsulated in this number? Thank you.
Do you want to do that? Both Tim, or you do the second one? I'll do the first one then. 90% share that we talk about in 2016. Well, who knows exactly what that share is going to be. Today, we know that it is, to all intents and purposes, 100%. We all understand that Intel have introduced some much more power-efficient chips, and these chips are getting design-ins into smartphones. We believe that they will continue to get design-ins in some smartphones. We've played with the Xolo phone. It's a perfectly adequate phone. It does the job.
However, as and when Intel chips get even more competitive with Arm chips, and they're still a couple of generations behind Arm chips at the moment, but as they get more competitive, then they will become one of the about 20 suppliers who supply chips into the phone space at the moment. On a purely arithmetic basis, they will get some share. By 2016, we don't expect it to be any bigger than arithmetic suggests, which is in the order of 5%-10%, and hence our greater than 90% number. Sorry, Tim.
Yeah, I mean, on licensing, the way we do it is we look at our existing licensees. We look at what we think those existing licensees will want from our technology. If you're an existing licensee operating in one vertical, you will be coming back to Arm every two or three years to upgrade that technology. That drives additional licensing. If you're an Arm licensee that has increasingly and will probably more take Arm into other end markets, which is actually most of our semiconductor licensees, you will be coming back for different flavors of Arm technology over time, and we model that out too. Of course, to your last point, we're trying to identify, at least in terms of scale, new companies that are likely to be able to take advantage of Arm technology as these new markets open up.
When we model that out, as we did in the 2007 timeframe, we envisage a period over time of high single-digit growth. Now, in reality, you're going to get periods which may be related somewhat to our product cycle, or our engineering delivery work, or even the macro environment where it's going to deviate from that. Sometimes it's going to be much stronger than you've seen recently. Others, like in 2009, it's going to be negative. When we look through it, that's a pretty detailed exercise which will obviously be wrong in the granularity of it. That's how we do it. It is actually based on a fair degree of science and understanding of how the semiconductor industries will want to deploy our technology.
Okay. We've got one from Brendan Smith.
Thank you. I'm Ambrish from BMO. Two very quick ones. One for you, Pete, a clarification. Is the roadmap on the GPU compute points us to going after the discrete graphic business as well?
No. We do not intend going after the discrete graphic business. No.
That's a quick one. Then one for you, Ian, again, a blocking and tackling one, which is if on the server roadmap is virtualization, and I believe VMware is a partner, is virtualization the support there, or what's the timing for that?
Yeah, good question. From the A15 onwards, we have hardware virtualization support. The core that Tom talked about as we move forward into V8 will have hardware hooks. We're traditionally seeing in that data center area other virtualization things like Xen, KVM initially. Certainly, we have a partnership with VMware today, and we'd look to extend that hopefully into the server space too.
When AMD comes out with their silicon, it'll have virtualization support?
Yes. They've announced it has hardware virtualization support.
Okay. I'll go with the trend. I'll go with a hard one with Simon. FinFET transistor cost on a nice little linear curve that Intel showed, obviously you guys are not standing still, you have the architectural innovation. What's the right way to think about it in terms of a framework, whether it's at the same power, X performance, just help us out there. The first question is, do you subscribe to that the foundry camp is going to have a difficult time, you yourself said that it is getting harder as you go down the node. A, do you subscribe to that? If the answer is no, then what is Arm doing with the architectural innovations that kind of helps to offset that so-called advantage that Intel has, thanks.
I think the roadmap for process technology in the foundries is pretty clear, that they're all going to introduce FinFETs at some point. I mean, they've all come out and said that. The question is about when. The question is about how a particular process technology is suited for the type of product that you are going to wrap it around. If you're building an SoC today, people typically mix up transistors. There's a lot of analog integrated onto a digital chip. Doing analog with FinFETs has some new challenges. How you can get all the IT in place to support an SoC industry, it's going to take a bit of time. There's the question of getting the mixture of performance and power consumption at the right point as well.
I think FinFETs are going to come along from the foundry players, it's about putting together the complete package of technologies that you need to enable an SoC to go be built using those technologies. If it's going to happen, there are clearly new sets of manufacturing challenges associated with that. It's not as though the foundries haven't been looking at this problem for a long time. I mean, Intel did not invent FinFETs. They came out of University of California, Berkeley a long time ago. The guy that pioneered the research on that used to be TSMC's CTO. It's not as though this technology has just come out of Intel everyone's been scrambling to catch up.
Been around for quite a long time now. Lots of work's been done across the industry on how you tame them, and how you get them usable for the right type of products.
Good afternoon.
Thanks. Nick Hyslop from RBC. You used the chart a couple of times where you're segmenting the smartphone market into super phone, mass market, and entry level. Could you just give us a bit more of an explanation of what you mean by those, and in particular, the amount of Arm content that you're expecting in each of those 3 segmentations, either by numbers of chips or preferably by value? I know it's not an exact science, but it's interesting to hear what you think.
Yeah. I think this is something which is going to change as smartphones evolve. A little while ago, we had feature phones and we had smartphones. Then we talked about really high-end smartphones and simpler, more basic smartphones. Now we've sort of got 3 levels. I think what we mean by the entry-level smartphone is, in a few years' time, that's going to have things like the A7 processor in it that Tom talked about in his section of the presentation. It's going to be targeted at low price. This is enabling the next billion people to connect to the internet with an entry-level smartphone. The super phone is going to have GPU compute in it. It's going to have big.LITTLE, and it'll probably be quite sophisticated big.LITTLE implementations. Maybe two big cores and four little cores, those sorts of things showing up.
May even have multiple graphics engines as well. This is going to be the highest price type smartphones that are bought in the developing world by the people that have to have the latest and greatest. These chips will be interchangeable with the chips used in tablets and computers. They will be effectively the same chip. You've got something in between, which is for the slightly more cost-conscious but still quite sophisticated user, and that's what we're defining by our middle range. From a financial point of view, the entry-level phone is going to have a modem and an applications processor in it. An applications processor is going to be a low-end apps processor in the GBP 10-GBP 15 range. At the high end, you've got an expensive applications processor, lots of other chips around it.
You're probably getting towards a multiple of three to four times the first one. The in the middle one is where most of the volume will end up being, and that's your GBP 20 chip.
This is your GBP 0.20-GBP 0.65 range across those.
This is what I was quoted as saying on Monday. Tuesday, yeah.
Okay, great. Thank you. I have one more question. You mentioned two things on the graphics processor. You talked about close integration with the CPU. You talked about low power as being Mali's key characteristics. Where you've been successful and won against your competition, what would you say it was that made them select Mali?
It depends on the customer, really. Some customers have chosen us because we have the best power and the best performance density per power. Other customers have chosen us
Again, the roadmaps are different. Customers on the GPU compute side have chosen us because we've made no compromises at all. You can have full profile GPU compute running. You can get every single OS supported. It depends on the customer. Some, as I say, it's low power. Some, it's functionality combined with low power.
Hold that back.
Thanks. Gabe O'Connor, Goldman. Question for Simon on the manufacturing side. Simon, in your discussions with foundries, how confident are you that they can get back onto the shrink roadmap once again? Or do you think there's a risk that going forward at the lower geometries, that the gap between foundries and Intel could extend that bit further, and maybe a potential risk there for market share? A second one for Pete on the graphics side. Pete, you spoke about a focus on proprietary graphics. Maybe you can give us some timeline when you might update us on that? Also, if you could just speak briefly on your DirectX support? Thanks.
Okay. I think in terms of the shrink roadmap from 28 to 20 to 14, I think that is going to happen. I don't think there's any notion that roadmap has stopped at all. There are some challenges facing everybody in exactly the same way who's trying to manufacture about the equipment that enables a shrink to the new node. Everybody's adopting double patterning at the 20 nanometer point because EUV hasn't come to industrial scale production and still isn't, and doesn't look like it's going to be a way off. That's going to impact everybody, whether you're a foundry or an IDM. It's going to impact everyone. As I said a moment ago, the foundries and Intel have different business models. The foundries are supporting hundreds of different customers. They're running many different processes simultaneously in their fabs.
They're able to do that in a very high yield way, and they're able to offer a very cost-effective, dense, high-performance solution. There's a whole load of trade-offs to support that business, and that has underpinned the SoC industry for the last 20-odd years. I don't see any of them stopping that. Nobody, as far as I can see, is throwing in the towel and saying, "You know what? It's all too hard. I give up." That is just not happening. If anything, I see the focus on R&D development going up and not going back.
Okay. I'll just take the graphics question quickly. We had one on proprietary GPUs and one on DirectX. I think, as I said, one of our main opportunities is proprietary GPUs. I think what we're seeing on the GPU side is very similar to what happened on the CPU side. Simon said there were a lot of proprietary CPUs over time. Gradually, people realized that outsourcing and doing a buy versus make decision, it was just better to buy. We're hoping we're going to see the same on the proprietary GPU side. In terms of DirectX support, we have full DirectX support. We have support for DX nine. We have support for DX 11.1. The hardware is available, the software is up and running, and we're ready for OEM launch.
Okay, I think that we're moving back there. Somebody has the microphone. Gareth?
Gareth. Yes, I've got a few if I could. I just wonder on the cost side, Tim, you provided some very good revenue signals. I just wondered, given the proliferation into a variety of end markets, given that the PC OEMs aren't used to maybe the same business model as they're now discovering, what you feel on the cost front going forwards, whether you feel you have to grow costs at slightly more than half your revenue growth, or whether those costs will be borne by others in the industry. Then secondly, I just wondered if you could talk about graphics. You mentioned Graphics POPs. You've been very clear in terms of the benefits of POPs historically, and I just wonder whether you could maybe give some of the benefits that you see on power consumption, et cetera, on Graphics POPs specifically.
Then finally, just on Intel pricing versus Intel, you put up the interesting slide earlier about $20 versus $100 for CPU. It's about a fifth. Can you tell us how you see that trending over the next five, six years? Thank you.
Let me do the first one and then the POPs. You do the cost.
You'll do the cost.
Yeah, I'll do.
Sure
On the costs, Gareth, it's what I said when I was standing there, which is I don't see any fundamental change in the shape of our R&D trajectory which is why I think the operating leverage still comes through. I did make the point that we will obviously be investing to optimize our opportunity. Therefore, if we see the need to invest in certain things that could have appropriate returns, this is why I said, maybe in five years' time, we're more valuable at 50% margin than we are at 60%. We don't see anything. Our model, the way we go to market, the way we operate with our customers in the ecosystem doesn't fundamentally change. Clearly, we need to grow. We need to grow our infrastructure within the business. We need to grow our commercial feet on the street.
This is what's been happening, and this is what we're taking into account in the guidance. I don't see that the model changes shape fundamentally.
Okay. On the graphics POPs, really good question. I think one of the questions earlier was, are we going to go into discrete graphics? No. The thing that's driving our increased complexity in the roadmap is there are new performance requirements out there, and the size of the GPUs are increasing constantly. What we've done with the G POPs, the aim is to take about 10%-15% of area and power out of the GPUs. The size of the GPUs, that's worth anywhere from $0.50 to $1 per chip.
Okay. You had a question about the pricing chart that we showed there. I think it's important to stress that Arm does not sell chips. Our semiconductor partners sell chips, and the prices at which they sell them is up to them and not up to us. What Arm provides, though, what we showed on Simon's slide was, that Arm provides a business model which delivers a whole ecosystem that encourages innovation and generates a competitive supply environment. When we look back, what that's done for the handset market is has enabled a huge increase in functionality and still produces chips that get sold in the sort of $15-$20 range. As we look forward, we can see much greater functionality going into smartphones.
We expect that same business model to deliver the competitive pricing environment so people will still be able to buy smartphone chips for around $20. When that smartphone chip is capable of doing everything that it takes to drive a PC, then it's very much up to the supply and the market dynamics between the people who are buying the chips and the people who are selling them to say, "Well, if I can buy a smartphone chip that does everything I need for a smartphone chip price, then that's what I'm going to pay for my smartphone chip that I'm going to stick into a mobile computer." We'll have to see what actually plays out. That's what the business model delivers. Yep, we're moving towards the back of the room now, and we have about five minutes to go. Okay?
Thanks. Taking a question.
No more.
Julian Yates from Investec. Just a quick one, Tim. In terms of the royalty outlook you put up, I'm quite interested why you didn't sort of move up the lower end of that 10 to 15 range in terms of growing success of the industry. Just looking at the licenses that you've signed in the previous few years and those royalties obviously yet to come through in a material way, taking market share gains. Wouldn't it be normal to sort of assume that maybe you're looking more towards the higher end of that range rather than the lower end of that range with those dynamics?
Well, as you know, I try and position myself as a cautious sort of guy. You saw from the last period that we did actually, the outturn was at the very top end of the range. I think when you're looking at crystal balls so far out, I think getting caught up in the excitement of these sorts of occasions can take you into areas where you create hostages to fortune. I think as long-term guidance, I think that's right. I did say that if you look at the rate at which we are gaining market share or likely to gain market share in those markets could well be at the upper end of what's been normal. Obviously, some of that comes from the very large microcontroller unit opportunity, which comes at a lower royalty per chip.
Okay. We have time for two or maybe three. Let's see how quickly we can go.
Yeah. Thanks. It's Brett Simpson at Arete. I had a question really on the Cortex-A unit shipments. They seem to be on a quarterly basis, at least. They're getting dominated by three players, principally Samsung, Qualcomm, and Apple. It's happening in a way we haven't seen before in the industry, and it doesn't look like many other chip makers have much prospects to make money in this environment given how much they're taking share right now. I had a couple of questions on that. How does Arm see this trend sort of building? Should we be expecting a sort of shakeout in wireless semiconductor? If so, how would that really impact your licensing business? The second question, are these three players that are doing so well right now, are they paying the same royalties to Arm that the rest of the market will be paying?
Thanks.
Shall I sort of start off on that one, I think, right? What people pay in royalties to Arm is we talk about a band, you talked about a very small number of semiconductor companies there. We're not going to get engaged in a discussion about how much individual companies pay. Everyone pays within the bands that we talk about when we talk about our royalty rates publicly. That means the Cortex-A processors, they pay more than they paid for processors that are less than Cortex-A. Typically, Cortex-A high-end processors now are commanding royalty rates getting towards 2%. I think in terms of do we see multiple players able to compete in that space? Yes, we do. In fact, I presented a slide yesterday. We were talking about the mobile ecosystem and picking out Cortex-A partners there.
The reason that I talk about roughly 20 players supplying into this space of mobile phones and mobile computers is that when we look at the people who actually license Cortex-A products and who are either shipping or whom we know have serious plans to ship, otherwise they're committing commercial suicide because they've invested quite a lot of money in their development programs. The number is about 20 players. I'm sorry, we don't quite buy the theory that just because Apple and Samsung have a very high share of the smartphone market, that translates into what happens with people who are shipping Cortex-A products. Cortex-A goes into a whole load of other products besides the high-end smartphones that come from Apple and Samsung. Sorry to shatter that illusion.
Okay. Thanks, Warren. Maybe just a quick follow-up for Simon. In your presentation, you talked about this world in design that's changing where 20 nanometer, there's a lot more verification, a lot more software that's going into these types of chips. Is there new opportunities for Arm to build fresh revenue streams on IP? Do you see any opportunities to get into the software business or adjacent areas where you can leverage that ecosystem you talk about perhaps in verification or in software going forward? Thanks.
Potentially, yes. We've looked at embedded software a few times over the history of Arm. At the same time, we like the fact that we have a broad ecosystem of partners that we're working with to do some of those things that we are not expert in. As I said, we can't do everything. We do some things really well, and there are some areas where we can leverage our historic strength and our business model, and repeat over again. There are some things that we choose not to do or choose not even to try and go and do, because there are already a bunch of other players in our ecosystem who are doing it quite well. We're going to carefully look at the opportunities around this. As you say, verification is a big problem.
Our historic approach to that has been to work with EDA companies who are really good at solving that sort of thing, and enable that for the Arm ecosystem. I think for now, that's the approach that we would take there.
Okay. Very last question.
Sure. Thanks. It's Lee Simpson from Jefferies. Maybe a couple of quick ones for Pete, if I could. Pete, I wonder if you could maybe give us some comps for Skrymir versus the next-generation architecture coming out of a rival, who's talking about a baseline of 100 Gigaflops for cores in the next 12 months. Maybe alongside that as well, we're hearing increasingly about HSAs. AMD making a big noise about that back in February. I wonder what that does for or how that sits alongside your philosophy of optimizing next to CPU, and looking to make some clip there on the space size too.
Okay. I'll be very quick. I'm not going to give any details on Skrymir just now. We'll be doing a launch on that specific product later on. It's a nice roadmap blob. That's all we're going to talk about just now. You would expect it has a lot more performance. That's what's driving us there. On HSA, we like the ideas that AMD and the HSA Foundation are coming up with. We have talked to them quite extensively, and it's very interesting, and it's completely in line with our ideas as well.
Okay. With that, I'm afraid it's 1:00, and we have to terminate the session. Thank you all very much for coming along, and we hope you enjoyed it.