A better part of my early days contributing to Seeking Alpha involved formal responses to works from other authors. While these days, I haven't really had a good reason to do so, I believe that the time has come again. Fellow contributor(s) Trefis put out an article in which they predicted that impending competition in the "sever market" would rob Intel (INTC) of its market share in server processors. I submit to you that this is completely unfounded and that there is absolutely no evidence at this time that Intel will do anything but gain market share in this space.
The X86 Power Myth...Do I Have To Write This Again?
There is something that I write about often but really wish I didn't have to - the Intel Architecture "power myth". See, ARM's (ARMH) marketing department has managed to brainwash the investing public into believing that chips based on their instruction set architecture and/or processor core designs were inherently superior to designs from Intel. They, and some of their partners, misled investors by pointing out that because Intel's high end "Xeon" products used more power than cell-phone chips powered by ARM cores that Intel's chips were inefficient and that it was time to put ARM chips into the data-center.
This is marketing nonsense. ARM chips in phones (and Intel's own Atom, too) consume much less power than Intel's high end "Core" and "Xeon" parts because - surprise! - they're much lower performing! As you scale performance upward, and as you add a lot of the system architecture features required to support that high performance (new instruction sets, fatter caches, more memory bandwidth, beefier execution units, higher clock speeds, etc.), the performance/watt curve starts to look very non-linear as you scale up performance.
However, this is a moot argument, as Intel will be first out of the gate with 64-bit, high performance, very low power solutions for micro-servers, comms, and other applications with its "Avoton" and "Rangeley" SoCs based on the "Silvermont" micro-architecture...8 cores worth of chips with performance/watt that will very likely dwarf that of the ARM chips. The nearest competitor - Applied Micro (AMCC) - just started sampling 40nm chips (from a design team with zero track record and substantially smaller budget), and will probably be later to market. Is this even a contest?
So, with Intel adopting a very flexibile, "SoC" methodology, it will not only have the best processor cores, the best reputation, and the smallest, lowest power die, but it will be able to quickly develop different SoCs for different target markets based on the base IP, as illustrated here:
The Intel Cost Myth
Let's talk basic economics. When it comes to profitably providing the cheapest chip, what does it take? Well, logic would tell you that you need to have a lower cost structure, right? The winner here is Intel. Not only can Intel amortize the R&D costs of its low power cores/IP across many different product families from micro-servers to smartphones and everything in between, but it has the scale to further lower per unit costs.
Oh, and don't forget that Intel's transistors are much smaller than the competitions', so it makes very little sense to expect that for a given performance/feature level that Intel would be at a disadvantage - quite the opposite. Intel can pack in more transistors in a given area, meaning that it can either provide the same functionality with a smaller area, or provide much more functionality in the same area.
People are often confused and think that because Intel sells high end chips for high ASPs that they can't/won't sell cheaper, smaller chips. It's absolutely preposterous. The die sizes for an Intel Atom or an ARM based SoC for mobile use are much smaller than those of the high end "Core" products. So Intel could charge less and maintain 60%+ gross margin without a problem. Remember that Intel doesn't have to pay an external foundry's margin (it keeps it all), while all of its competitors have to make sure TSMC (TSM) gets its 40% gross margin. Furthermore, ARM demands its royalties on its IP - more ARM IP used means less gross margin for the fabless companies.
See, companies like Qualcomm (QCOM), Nvidia (NVDA) and others don't run a charity - they're out to make money, too. Do you really think that any sane management team would stay in a business in a chip which R&D and manufacturing costs get more expensive if there weren't >50% gross margins to be had? If a fabless company like the two aforementioned names can command 50%+ margins on mobile processors even with the foundry overhead as well as the ARM royalty overhead, why can't Intel maintain 60%+?
The Server Market Share Loss Myth
Finally we get to the server market share loss myth. Everyone is getting excited because ARM has finally developed a 64-bit instruction set architecture. Now, because they think "64 bit" is the only "requirement" to make a dent in the server space - especially the high end, high value space that Intel currently dominates - all of a sudden, we start seeing claims that Intel will lose market share.
No, that's not how it works. First of all, Intel's high end "Xeon" processors are safe. IBM and Oracle (ORCL), the only two powerhouses left with competing parts in the high end, are rapidly losing share to big-iron X86 processors. "Nehalem-EX" started the ball rolling, and I believe the 2013 chip, "Ivy Bridge EX" should be the final nail in the "classical RISC architecture" vendors in the high end space. The market share gain trend for Intel has been very real. Compared to IBM's POWER and Oracle's SPARC, anything that ARM makes is the farthest thing for competitive.
The 2-socket/4-socket market? That's also on lockdown by Intel. However, the low end Web 2.0 market - that is micro-servers - is still up for grabs. Intel currently owns this market with its Atom and Xeon E3 products, but of course some of these startups are going to gun for a piece of the pie. It's not likely that they succeed. Why?
- Inferior Cost Structure: As I said, Intel not only has better cost per transistor, but it can also amortize the costs of development over a much wider swath of products than any of the other competitors can
- Worse Performance/Watt: Being one process node behind in a bleeding edge computing market hurts you both economically and when it comes to performance in a given power envelope. Who will want to purchase chips that are potentially more expensive and that will cost money in the form of a bigger power bill each month?
- TSMC's FinFETs Coming Soon? Nope: By the time TSMC moves to its first FinFETs on "16nm" in 2015 at the earliest, Intel will be ramping its 10nm production. Oh, and TSMC's "16nm" is really 16nm FinFETs on the front end but recycles the 20nm back end, negating all of the scaling benefits that normally come with a "node shrink". TSMC will pass the higher cost/transistor on to its customers, so they will either pay dearly for bleeding edge TSMC nodes or they will be stuck a node back. In either case, this is 1-3 generations behind Intel's leading edge
Then, of course, there's the notion that AMD will gain back server share. Not likely. AMD has been bleeding server share for years now, and there is nothing on the horizon that will stop that. AMD is (rightfully) putting the development effort into the low power cores for the client space, and is trying to whip together a me-too ARM SoC for the server space. All of the process disadvantages are still present, but now AMD is giving up its biggest asset - X86 compatibility. AMD is very unlikely to gain any meaningful share in servers, unless you could actually selling Intel-powered boxes.
Intel isn't in too much danger of losing material server share thanks to its strong cost structure, aggressive roadmaps/products for both the low power and high performance spaces, and a reputation as a reliable server chip supplier that is second to none in the chip industry.
If running a profitable server chip business were as easy as taking an ARM A57 license and hiring a team of engineers to build an SoC around that core, then it would have been done many years ago with the MIPS64 instruction set architecture. I still don't see a MIPS64 compatible competitor to Intel's processors in the general purpose server space...the smart companies like Broadcom (BRCM) are busy focusing on profitable, high margin, low volume niches where Intel's chips aren't really targeted...yet.
Additional disclosure: I am short ARMH