Intel (NASDAQ:INTC) is dominant in server chips to the same extent as ARM Holdings (NASDAQ:ARMH) dominates smart phones chips, with Intel having over 95% of server market share while ARM has 95% of the smart phone market.
Intel bulls hope that it will start to take share in mobile, but what if instead ARM takes market share in servers?
How seriously should we take these reports and what are the implications of these developments for Intel?
Servers are a big deal for Intel
This table shows how important servers are for Intel:
Source: Intel 10Q
The Data Center Group - what Intel calls its server division - makes up 21% of revenue but 40% of operating income. Mobile is just one part of "Other Intel architecture operating segments", and that group only accounts for 7% of revenue and currently costs about 5% of Intel's operating income.
Moreover, if ARM based chips can be competitive in servers then they can certainly be competitive for PC use, at least from the point of view of the chip's performance and power.
Still, is it really credible that companies will choose to go with ARM based servers instead of Intel's servers? One way of looking at this question is to compare and contrast the difficulties Intel has in penetrating the mobile market with any potential advance ARM might make in servers.
The benefits of switching to ARM servers seem small
First of all the cost of server chips are much less of a factor for Facebook, Google and other tech companies, than the cost of a chip is for a mobile device vendor.
Google, for instance, is reported to spend $500 million annually on Intel chips, which is only 1% of Google's revenue. In contrast, a smart phone chip - that's costs $30 say, makes up a double digit percentage of the total smart phone cost.
Another thing to consider is that power consumption for server chips is something tech companies want to minimize but electricity costs are still fairly low. For mobile devices, on the other hand, the opposite is true: power consumption can make or break the device in the minds of consumers.
This is why ARM has done so well with mobile device vendors: its chips have historically been designed for low power and low performance applications, as well as being low cost.
This would seem to suggest that Intel has nothing to fear. However, while the benefits of switching over to ARM in servers is small compared to the benefits of switching from Intel's chips to ARM's chips in mobile devices, so too are the costs of such a move.
But the costs of switching to ARM servers are relatively small too
What I mean by that is switching over to using chips that have a different Instruction Set Architecture, ISA, (x86 chips to ARM's RISC or vice versa) is a lot more of a hassle in mobile devices than doing the same in a server farm.
This is because mobile devices are used directly by end consumers and those mobile devices come with of large number of legacy applications. Therefore switching to a different ISA, will either involve rewriting those apps for that new standard or changing the OS so that it can run on both kinds of chips, just like Android.
Let's assume Apple (NASDAQ:AAPL) wanted to switch from ARM to Intel for its iOS devices. It would then either have to persuade its app developers to rewrite all their apps - which would be a complete nightmare - or reconfigure iOS so that it operated like Android. But there's a performance penalty involved with having the capability to run on both ARM and x86 chips, which is one of the reasons why Android devices are so often sluggish compared to the smoothness of Apple's iDevices. As loyal as Apple's customers are, it simply wouldn't be acceptable for Apple to degrade the performance of all their customers' mobile devices to that extent.
In contrast, integrating both ARM and x86 servers together won't affect the end consumer one bit: no one using Google's search engine, or logging in to Facebook will know the difference.
That isn't to say that there aren't significant software issues with both designing a competitive ARM based server chip as well as running those servers alongside ones from Intel. The whole ecosystem of tools and software programs necessary for developing these servers will have to be advanced. However, ultimately it's just a question of throwing lots of engineering resources at the problem, something both Facebook and Google have plenty of.
Don't call my bluff
How serious are Google and Facebook about ARM servers? My belief is that they don't care that much - the economics of the thing really aren't that compelling. However, and this is important, I do think Google and Facebook are making noises about using non-Intel servers as part of a negotiating ploy with Intel - they obviously want lower prices. If Intel doesn't oblige them, then Facebook and Google will say to Intel we'll create our own servers. Now, if all these companies do is talk, then Intel will call their bluff, therefore at some point Facebook and Google will need to go beyond just talking about making their own servers.
The actual loss of business from these two companies to Intel won't be that large. Google, for example buys $500 million worth of servers from Intel, which is only 1% of Intel's total revenues - the real downside for Intel is that by developing ARM based servers Facebook and Google will have helped advance that supporting software ecosystem for ARM servers.
In turn this will make it much easier for other, smaller companies, who don't the resources that a Google or Facebook does, to switch to ARM based servers. Therefore what might start out as a negotiating ploy on the part of some of Intel's biggest customers could end up unravelling Intel's business entirely.