Intel’s Challenge in the Post PC World

Let’s start with some comments on “Post PC.” If we take a step back and look at the global computing landscape, we realize two things. First, there is a “PC plus” market (people who use a PC and other devices like tablets and smartphones together) and a non-PC market where the only computer people use are smartphones. Understanding the PC plus segment is how I predicted 2014 would see some rebound of the PC market. Microsoft says 1.5 billion people use Windows every day. This does not mean there are 1.5 billion unique PC owners though. That number is much lower. Between corporate, small business, consumer, internet cafes, and point of service (retail) use of PCs, there is a healthy installed base and many of those are not going away. However, the market for the PC is also not growing and, in fact, may be on the decline. Servers remains a bright spot and a overall growth opportunity for Intel. However, a great deal of the volume in computing chipset demand is in smartphones. Intel’s lack of relevance in the massive market of smartphones, and even tablets to a degree, lies at the root of their greatest struggle.

Understanding Their Business

Intel’s business is to manufacture semiconductors. They are one of a handful of companies who own the physical space and equipment to make the millions upon millions of microprocessors necessary to bring about our wonderful computing future. The rub is Intel currently only manufactures chips it designs for the x86 architecture. Right now, those chips are only positioned well in the market for PCs and servers. The challenge as a manufacturer of silicon is you need to keep your semiconductor manufacturing facilities full in order to profit on the initial investment in psychical space and equipment. By only being relevant in PCs and servers, keeping semiconductor fabs full has posed a challenge. A question I like to pose is this: every single major semiconductor manufacturer is at or near 100% capacity except for one. Guess who it is?

The answer is Intel.

Betting on Moore’s Law

Intel has a lot riding on Moore’s Law. Intel is committed to x86 and, for server and most PC use cases, this is the right architecture. x86 has always had an advantage over competing architectures in performance. It struggles to have an advantage in power consumption. Thanks to Moore’s Law, x86 has been making strides in bringing the performance of x86 to lower powered solutions. Intel’s current processor technology is 14nm and it is the first process where we can truly have desktop class x86 “core I” chipsets that don’t require a fan. Even with the benefits of 14nm, Intel will be challenged to fill their fabs, making it tough to monetize the eleven plus billion dollars they invested in the 14nm process technology. Yet they will still need to spend double digit billions of dollars to invest in the next process node which will be 10nm.

There is good reason to keep Moore’s Law alive. But to bet so heavily on it means the economics need to be there. In fact, it is possible following Moore’s Law for Intel is more of an economic challenge than a technical one.

Investing in Moore’s Law requires investing in new process technology every few years. Currently, Intel is shipping semiconductors at a range of process technologies, but a couple of the main ones are 28nm and 22nm. Intel has also recently begun mass producing chips at 14nm. Intel’s next step in pursuit of Moore’s Law is to invest in and start mass producing 10nm chipsets. After that, they will go to 7nm. The cost to do this in CapEx is massive. To visualize this, here is a chart Intel showed regarding how they outspend competitors to keep a lead in process technology.

Screen Shot 2014-11-20 at 10.04.07 AM

This is Intel’s process technology advantage but the economics to keep it going is also their challenge. Intel mush keep those fabs filled, recoup their investment, and continue investing going forward to maintain a lead. This is what being absent in the high growth and high scale area of mobile phones, tablets, and, to a degree, the Internet of Things is costing Intel. Intel needs the scale of markets like smartphones to keep Moore’s Law working for them.

Why I’m Optimistic

Among the many benefits of advancing process technology is an increase or an approximate doubling of the number of transistors you can pack onto the silicon. Which means designers of said chips have an increased number of transistors at their disposal they can spend their “transistor budget” on in valuable ways.

I believe the future of the semiconductor industry is in the hands of those who have in-house capabilities to design semiconductors. A handful of companies have this capability. Among them are Apple, AMD, Qualcomm, Broadcomm, Nvidia, a few others and, of course, Intel. Intel sets themselves apart from the pack because they design and manufacture. While it is debatable whether they are the best chipset designers out there, they are at least in a position to make a run competing on chipset design. I feel this is exactly what they need to do if they are to compete with ARM chipset designers as well.

The other benefit of Moore’s Law is the decrease in the cost of the transistors. This is not always true but, in theory, this is the goal. Which means Intel should be in a position to offer powerful, very low energy consumption and low cost chipsets for the applications for which they are designed.

In theory, Moore’s Law IS an advantage for Intel. The problem is Intel has always told us “just wait for the next node.” After a while we get tired of waiting. Those of us who follow Intel know it is not their current 14nm chip where we will know if they are truly performance, power, and cost competitive with their ARM rivals. It is at 10nm we will know if this is true or not. For the time being, I’ll buy the pitch that 10nm is the big difference maker for them when they actually have a solution contesting the high scale/high demand mobile market. But if Moore’s Law doesn’t deliver for Intel by then, it is time to execute plan B,C, or D.

Published by

Ben Bajarin

Ben Bajarin is a Principal Analyst and the head of primary research at Creative Strategies, Inc - An industry analysis, market intelligence and research firm located in Silicon Valley. His primary focus is consumer technology and market trend research and he is responsible for studying over 30 countries. Full Bio

100 thoughts on “Intel’s Challenge in the Post PC World”

  1. Too often people assume there is a big x86 tax, and that is why Intel wasn’t competetive in mobile. There is an x86 tax but it is relatively inconsequential on modern CPUs.

    The reality is more that Intel ignored small mobile (phones/tablets) to concentrate on desktop/Server/notebook CPUs. They were chasing high ASPs and with it high profit margins.

    Atom languished with the same core design from 2008-2013 and with a hand me down process. Intel parts as late as 2013 were a 5 year core design, that was 45nm die married to a 65 nm die in the same package. A half-hearted effort at best.

    Only in 2013 did the first serious 22nm SoC with new core emerge. Have we seen 14nm Atoms yet? Even the 22nm part was quite competetive in perf/power.

    On the process side Intel has no issues. They lead.

    The x86 “tax” is overblown.

    The reason for optimism is that Intel is now getting serious about SoCs and has disclosed plans to tic-toc iterations.

    There is no question they will produce competetive SoCs.

    The issue they face is installed base. Phones are nearly entirely ARM.

    Intel needs a killer part to sway people in its direction, more than 10% better.

    1. That high profit margins you speak of, that’s the Intel Tax. They can’t use the business model they have for laptop, desktop, server chips in the mobile market place. On the “PC” side of things, Intel is basically the monopolist (they have small competitors who don’t amount to much), while on the mobile side, they are an entrant into the market. They aren’t going to disrupt the market from the low end because their prices aren’t cheap enough. They aren’t going to disrupt the market from the high end because they are trying to sell to business who make smart phones and tablets and not consumers. High end disruption as a supplier to the low end of the phone/tablet market just isn’t going to work.

      1. Bingo. I would add that CPU was the big feature in PC’s for a while. In smartphones we talk about battery size, screen resolution, manufacturing materials, cameras, etc. The CPU is just not the killer feature it used to be, hard to see the CapEX of their best process being recouped at mobile prices.

  2. Intel could be riding the wave of smartphone market had Paul Otellini not declined Apple’s approach to have them make that chip for the iPhone. Otellini regretted that decision on his way out. Poor decisions cost dearly. Paul was a marketing man and it is surprising that he did not have the vision of what was ahead beyond the borderline. Unlike Microsoft, Intel at least truly innovates and creates. I hope and I am sure Intel will succeed in breaking through the mobile market with their newer nodes. However, overhead costs and profits might weigh that down considerably. Intel can give up its obsession with X-86 and adopt the competing technologies. It can start foundry business on a big scale. Apple is looking for a reliable manufacturer for its ARM processors. Intel can be a very reliable foundry in that regard. Intel is one company that is capable of coming back into the business. They truly invest in research and have developed a number of innovations. They have what it takes to come out of the doldrum.

  3. “Intel needs the scale of markets like smartphones to keep Moore’s Law working for them.”

    Intel has been trying for years to get manufacturers to adopt their low power atom designs for use in phones and tablets, and it mostly hasn’t been working. At this point I think it’s safe to say that dog isn’t going to hunt, and they need to switch away from their x86 comfort zone and start selling customized ARM SOCs. Competing with the established players will be tough, but what choice do they have.

    There’s also the question of whether Moore’s Law will continue working for anybody at this point. The driver to Moore’s law has always been lowering the cost per transistor (increased computing power wouldn’t have gotten off the ground if it wasn’t affordable). And it’s looking like at 20nm and below, manufacturers are not seeing the traditional reduction in cost per transistor anymore.

    1. seems the scaling continues to at least 10nm TSMC , INTEL, and Samsung/GF i would think. 7nm is certainly then on intel’s roadmap. But what after that? I suspect at some point they get to 5nm but we’ll know before then if intel’s one trick pony show is up. The problem for intel is they can’t design diverse chips fast enough to fill gaps that arm ecosystem can and switching to arm would be an admission of defeat for X86.

  4. Excellent analysis Ben, I think you hit the nail on the head with your thoughts on Intel. I have a few questions that maybe you can expand on for me but first I need to frame it in the right way. Intel has process leadership, everybody knows this. They can design very powerful cores, OoO, IPC, power efficient etc, their one fault has been not to have able to include LTE on the SOC which they have been working on. I have always thought that Intel’s major problem, and it really showed with Atom is that they tried to maximize their returns on the high end, core series, while not worrying or ignoring the low end, Atom. Apple must have had the same dilemma when they introduced Ipad in the sense that would a lower priced product cannibalize their Air series of laptops. To their credit Apple has never been afraid to cannibalize their own products. So a) in market share dynamics other than running counter to basic business logic of trying to maximize profit while not cannibalizing your own business divisions If Intel brought core level performance and process node down to SOC chips would volume make up for price per unit drop b)Do they need to view this as either get into mobile or become irrelevant in the near future. The cost of the fabs and the returns from limited production runs forcing them to change business directions completely. Or c) does the actual cost of the fabs make them powerless to compete at the level of ARM designs.

    1. I think their goal is core level perf, at ARM low-power and cost dynamics. 10nm is hopefully when they achieve this. In this brave new world Intel can’t live on chipset margins but should hope to make them up on volume where they are cost competitive with ARM in broader applications like mobile. There is only one premium player who would or could pay a little more for x86 if that was the goal and that is Apple. Who does not need them in mobile.

      I think they do need to get the core level performance and new designs around it to compete in mobile in markets like China and India which are huge. Those are both very spec driven markets even though the specs don’t make a heap of difference really they like the perception or idea of specs at low prices. This is why we only see octa-core chipsets in China and India. They are un-neccsary but the market seems to think they want or need it.

      Should they not achieve low-power and price parity with ARM designers at 10nm then yes the costs of fabs could become an issue. The economics will be hard to justify. They made a big deal at their investor day yesterday about how much better the transistor quality is on x86 and in particular the efficiencies they achieved at 14nm. They also pointed how difficult it was to get yields up, and leakage down. I believe they did this to suggest that this task of moving from 20nm to 16nm will be very tough for the ARM fabs. Perhaps taking them even more time and giving Intel more advantage in this area because they are there first in areas where it matters.

      Again, it all hinges on the promise of 10nm delivering. Yes Sofia is also a key, it is to them the same as the necessity to be relevant in graphics. If they can’t embed the connectivity they are hosed in mobile also. So lots has to come together. But there are some reasons to be optimistic as I point out.

        1. Actually many are better than Atom level performance, but that is besides the point. We need more perf in more mobile devices as we move to things like deeper encryption, 4k video and beyond, much much more visual compute to come to our pocket computers. Start adding sensors and we need even more.

          ARM knows this and wants to get to core level so Intel needs to get core level into ARM power/size ranges.

  5. My own personal thinking with regard to “post-PC” is that it is a somewhat hyperbolic aphorism that has never been seriously intended to be taken literally. What is really meant is that the focus of personal computing is shifting away from the desktop end of the PC spectrum and towards the mobile end.

    The impression that the traditional PC market is itself actually in decline is, I believe, false. PCs will continue to be used in business, academia, for scientific purposes and by regular people any time the convenience and capability of a full computing platform is helpful. While the growth of mobile makes this a smaller proportion of overall use, this is more than offset by continued worldwide growth in computing generally. Ten and twenty years from now there will be more PCs (including Macs) in use, and more being sold, than there are today.

    If I am correct then Intel’s x86 architecture does not face a challenge from dwindling applications, but instead from competing architectures in PCs, such as from ARM, which sensible people already believe will soon power at least some Macs. This seems to be a battle over pricing more than technology, with the market suggesting that Intel’s immense profits no longer brings its customers suitable returns. If until now it has been mainly Intel on whose impressive profits the continuation of Moore’s Law has rested, perhaps in the future it will be the ARM manufacturers’ turn to carry that torch.

    1. “The impression that the traditional PC market is itself actually in decline is, I believe, false. PCs will continue to be used in business, academia, for scientific purposes and by regular people any time the convenience and capability of a full computing platform is helpful.”

      I beilieve when you are referring to “a full computing platform” you are actually referring simply to the desktop GUI. Can you name a single critical thing that is tied to the underlying x86 architecture that wouldn’t be possible on an ARM powered device, especially ten years from now as you mention?

  6. Hi Ben,

    I’ve read of another challenge not included here: Intel is missing the efficient shared low cost IP creation going on within the ARM industry. Two glaring omissions in the Intel chart you show are Apple and Samsung. Where is their R&D? Apple’s R&D probably surpasses Qualcom’s and Samsung’s is probably in the neighborhood of TSMC.

    Within tablets today, the A8x (Apple + TSMC or Samsung) has //very profitably// defeated the Intel chip that has only $7B of losses to speak for it; in profit view, Apple is the Death Star of Intel mobile profits today.

    Apple + TSMC created two chips in this ONE year: the A8 with 2B transistors and the A8x with 3B. Seems pretty spectacular to me. Now Intel has accomplished something like this in the past, but Intel’s economics are different; they would need to milk a new chip architecture for several years of beneficial profits before rolling out their “A9”. Apple’s appetite for //profitable// processing power and efficiency “tick tocks” every single year!

    How does Intel compete with Apple and friends? I believe they can catch up only by becoming friendly with Apple, Qualcom, Broadcom et al. As a fab within the ARM industry Intel can become a very profitable team player. The question is, will the likes of chip designers Apple, Qualcom, and Broadcom trust Intel and schedule Intel’s fabs for their own needs?

    I’ll leave it at this. I believe Intel has to solve the challenge of the players in the top chart coupled with Apple’s dominant cash pile and obvious design chops and the excellent pace of Moore’s Law advances coming out of both Samsung and TSMC.

    1. About the Intel R&D chart: It would be interesting to compare Intel’s R&D directly to all the R&D summed within the major ARM chip designers and fabs. I would justify doing this because most of the IP created within the ARM system is reusable and sharable. In fact, I wouldn’t be surprised if TSMC and Samsung are sharing fab IP between themselves so each can stay competitive on a Moore’s Law basis with Intel. (Apple has the $$$$ and motivation to write big checks to both TSMC and Samsung, intentionally making sure each is thriving.)

    2. I have a lot of thoughts on this. And obviously what Apple is up to could benefit Intel as well if Intel would play their cards right.

      The first is we have to remember Apple doesn’t compete with Intel here because Apple does not sell chips to competitors. It certainly doesn’t help Intel because Apple is not a customer in mobile and I can confidently say with 99% assurance, an Intel designed chip will never be in anything Apple makes other than a Mac.

      But the point I want to make clear is that the future of SoCs is in the hands of those who can design chips. And those win among them will be those who spend their transistor budgets wisely.

      Process technology will be interesting simply because Intel is on pace to have a density lead at 10nm. But their challenge as I point out is that they are only manufacturing chips they design. There fore they live and die by their ability to design chips to keep fabs full, or enough to monetize and invest in the next process technology.

      I do think the ARM foundries will have some challenges getting to smaller and smaller nodes. We will see what ARM does over the next few years in terms of architecture advances so the foundries can move forward.

  7. Apologies but this article seems a little light on substance and is very topical. At what point is the cost a barrier for intel ? What are the drivers of that cost? At given nodes it is already clear they don’t make the best mobile soc’s and their mobile soc’s are 1 mode behind, why is that not addressed . I am curious for more insights and i can read intel’s powerpoint slides as well.

    1. One can only go so long and so deep when writing for a general audience. But I’ll address a few of your questions here.

      I’ll start by stating you point out in your comment above to Glaurung, that Intel can’t design diverse enough chips to fill gaps. That is absolutely true and, while we have to wait to see, that was my point about them having to become better designers. It was also the basis of my point that those who design the best chips will be in the best position going forward. On THAT note, I’m not sold that even the ARM architecture licensees are the best designers of general purpose SoCs. Given where I see mobile going, I think we need the designers to step up their game on mobile SoC design. On that point, I think Apple is one of the best chipset designers out there and what Apple is doing with their chipset designs is ultimately point out huge gaps in the designs of other players. SO at a base level I think competition in chipset design of general mobile SoCs is still anyones game.

      Yes ARM has the favored architecture for low-power, but for the most part those chips are ATOM level processing power not core level from Intel. Intel is in a position to being core level processing to 14nm and then 10nm. If they do this and it delivers on the low-power performance and they are designing efficient chips for general and vertical use cases, they are in a good spot. Of course, they also have to solve their modem issues. I can state with absolute confidence, if Intel does not offer a competitive global modem/baseband integrated into their chipset that covers LTE, 3g, and even 2g, they will not be relevant in mobile. But there are positive signs they are working this out and getting their technology globally certified, which is a huge step forward.

      What interests me about Intel, is that I do believe CISC has some some advantages of RISC in terms of architecture. I also believe Intel has very good process technology and the quality of their transistors is top notch. Knowing some of the struggles the ARM guys may have getting to 16nm and then beyond, it leads me to have some confidence at 10nm that Intel will be competitive. But as I said if they are not by then, then it is time to look at other plans.

      For more on this, we chatted about it on our podcast as well this week.

      1. u guys have been putting out very good market color lately. Thx for the reply. I guess we will just have to wait and see how this all plays out. I think not enough thrift is given to good enough computing and cost per watt per dollar for a given workload. I am big on photography and i am shocked at how gopro (8.2b ev ) is bigger comapny player than nikon (7b ev ) let’s say.
        In fact the more i think of it, and as you’re writings allude to, what we probably care about more now are orders of magnitudes improvements in performance, and no node shrink will give that, save for perhaps graphics (which outside intel is stuck at older nodes ).
        One more thing, irrespective of cpu performance, apple has shown the value of vertical integration on the smartphone, hence i am not so sure intel has the right bogey if all they do is an soc. What about the screen or security.
        Moving to the server, with intel being able to subsidize server cost thru pc volumes, i am not sure being leading node won’t end up eventually be disastrous for them. The really need at least flat pc sales, and i have gotten some of that intel/thesis (no pun intended ) from speak to analysts that cover the company.
        But give them credit, they really know how to milk a monopoly. The question is for just how long will that work.

  8. The disaster scenario for Intel runs as follows: 1) x86 revenues/profits drop, 2) overcapacity and reorganisations drain resources, 3) capex/R&D budgets are reduced, 4) their production advantage/scale withers away over the next three/four generations of process development.

    Intel has their work cut out for them in trying to avoid this. Particularly, because Apple owns so much of the profit in the PC and mobile industries.

Leave a Reply

Your email address will not be published. Required fields are marked *