Over the past several years, there has been growing interest in developing ARM Holdings (NASDAQ:ARMH)-based server processors from a variety of vendors including Advanced Micro Devices (NASDAQ:AMD), Applied Micro (NASDAQ:AMCC), and others in the hopes of grabbing a piece of Intel's (NASDAQ:INTC) very lucrative data-center pie.
So far, however, the ARM-based vendors have yet to deliver, with many of these players still a way out from getting competitive products built on modern semiconductor manufacturing technology to market. On the other hand, Intel has executed well, particularly with its 22-nanometer FinFET lineup of products that has been in the market since late 2013.
Why doesn't Intel get any respect in micro-servers?
The simple fact is that many investors believe that Intel has one or more of the following fundamental disadvantages relative to the various ARM startups:
- Claim: Intel is at an inherent power disadvantage thanks to X86. While this is an easy marketing point to use on investors, the reality is that these days the overhead associated with some of the key "talking points" that ARM (and other RISC vendors) like to bring up is negligible. Some will cite recent comments from chip guru Jim Keller that ARM is inherently more efficient, but that's actually quite perplexing given that it was AMD's own Fred Webe, who helped defined the X86-64 extensions to Intel's X86, that pointed out that the overhead was negligible years ago, when transistor budgets were much smaller -- meaning that any overhead is even less important today than it was all those years ago.
- Claim: Intel is at a cost and pricing disadvantage. According to some, ARM chips will somehow be "cheaper" than Intel chips. The flaw in that reasoning is that, given that the unit volumes in this market are relatively small compared with, say, mobile chips, the gross margins per unit need to be quite high to offset the substantial (and growing) R&D required to develop competitive products. Further, given that higher one-time costs for significant power and energy savings over the lifetime of the hardware is a trade-off most data center operators will make all day, it's unlikely that a company can win simply by engaging in a price war here.
- Claim: There are lots of competitors, so Intel's share must go down. While it seems that a lot of companies are taking out ARM licenses in a bid to try to cash in on this trend, look at what is happening in mobile systems on a chip -- the vendors are dropping one by one, the most recent of which was Broadcom. R&D barriers are tough to overcome even if a small startup can "theoretically" be a player in a market with a large total addressable market. The same pattern is likely to repeat in data centers, and Intel's scale, manufacturing lead, and design prowess and IP leverage will serve it well here.
Intel is the only one shipping
Intel has already rolled out its highly integrated 22-nanometer FinFET parts known as Avoton and Rangley for micro-servers and communications, respectively. Applied Micro is still trying to get its first-generation 40-nanometer X-Gene to market, and AMD is set to launch its first 28-nanometer products here in Q4 2014.
Intel's competitive positioning is quite favorable, and what's even more interesting is that Intel has had these parts in the market since Q4 2013. AMD's upcoming "Seattle" product based on ARM's Cortex A57 and built on a 28-nanometer process should be more competitive than X-Gene, but on a performance/power perspective for general purpose computing, it is unlikely to do all that well relative to the Intel products.
To AMD's credit, it does integrate more accelerators and networking capability than today's Avoton does, but recent details about Intel's upcoming Broadwell-DE, which is a highly integrated system on a chip based on Intel's "big" cores and built on its 14-nanometer FinFET process, suggest that Intel will neutralize AMD's edge here while offering more efficient computing performance in roughly the same timeframe of late 2014 to early 2015.
Foolish bottom line
Intel's position of leadership today in the data center is very hard won and is the product of immense R&D leverage, world-class manufacturing, and economies of scale that few could hope to match. The hype around ARM servers is likely to turn out to be just that, with real-world market share gains against Intel mirroring ARM's share gains against Intel in Windows PCs as a result of the launch of Windows RT (i.e., negligible), even though expected that opening up Windows to ARM would drive significant share loss for Intel.
That said, it'll be important to watch for further developments in this space. While I do believe this thesis is sound, surprises and unexpected twists and turns do happen. ARM servers are unlikely to gain the share that some think they will, but as always, only time will tell.