Intel (INTC 0.84%) didn't start off as the undisputed leader of server processors, but once the company set its sights on servers, it quickly tightened its iron grip on that market segment and squeezed out all competition. Today, Intel holds a seemingly insurmountable 96% share of the server processor market, but if recent reports are true, this might be the chip king's high-water mark. With the PC market already in decline and no clear path to mobile dominance yet laid down, this could be very bad news for Intel indeed.

Rebellion of the online kingdoms
By Intel's own reports, search giant Google (GOOGL -2.73%) is already its fifth-largest customer, with annual spending estimated at half a billion dollars by FBR Capital Markets analyst Christopher Rolland. However, rumblings are now spreading through the tech world that Google itself -- which operates more than 1 million servers in its quest to be the gatekeeper of the world's information -- wants to build chips of its own to more efficiently power its massive server operations.

Source: Darren McKeeman via Flickr.

If Rolland's calculations are close to the mark, that would still only mean that Google accounted for about 5% of Intel's server-segment revenue in 2012, and about 1% of Intel's total annual revenue. But with servers gaining in importance in the post-PC era, and with Google always hungry for more capacity in its world-encircling operations, it stands to reason that Big G would become a more important part of Intel's operations. Unless, of course, it starts building its own chips.

By licensing ARM Holdings (ARMH) designs, Google might take three years to whip up chips custom-built for serving up searches and YouTube videos and Gmails. ARM chips, known for low-power operation, could become a superior option for power-hungry data centers. Google is one of the power-hungriest companies in the world -- its worldwide operations are internally estimated to consume more than 220 megawatts of power -- so shaving even a small percentage off its per-server energy usage across those million servers could result in major cost savings.

Google isn't the only online powerhouse that's reportedly investigating switching to ARM chips. Facebook appears to be adding ARM processor support to its code base, and has also posted a job to its hiring board in search of an ARM server software engineer. This isn't quite as drastic a step as Google's custom chip-making, but it could be enough to signal the start of a sea change away from Intel's chips. The timing is also fortuitous for distant second-place server-chipmaker Advanced Micro Devices (AMD 0.41%), which has pledged to begin shipping its own ARM-based server chips in 2014's first quarter. According to PC World, AMD's betting its server-based future on the ARM designs -- it announced 64-bit ARM designs in 2011 and will be one of the first to ship ARM server chips.

The gates are still strong on Fortress Intel
The same factors holding Intel back from readily claiming market share in mobile devices also work in its favor when it comes to maintaining server dominance. Since ARM designs caught on quickly in mobile, hardware makers were used to optimizing their devices for ARM chips -- a hardware setup that works with one chip design won't work with a different design on a very fundamental level. And since all those early mobile devices were built for ARM designs, every app maker that flooded the market with the latest and greatest Angry Birds knockoff optimized its code for ARM as well. Making the app work on a different processor design takes a lot of work, and there's little incentive to put in the time when a very slim minority of users have devices running that hardware.

Source: user langleyo via Flickr.

This insulates ARM chips somewhat well from Intel's increasingly determined efforts to establish a mobile beachhead, but it might not be as great a barrier in the server-chip market, which has increasingly come to be dominated by large online companies running proprietary software. Tweaking their code to run on ARM chips would still be onerous, but it would be far more feasible than 100,000 small-time app developers shifting wholesale to a new chipset.

But there are reasons to suspect that, while Intel might eventually see its overwhelming market share decline somewhat, it probably isn't headed for the exits anytime soon. Intel's chip designs, while perhaps not as explicitly optimized for low power consumption as ARM's, are by any measure well ahead of ARM's in technical specifications. Intel recently released a 22-nanometer Xeon server processor. AMD's ARM server chips, which won't be out for months, use 28-nanometer designs. Intel is already demonstrating 14-nanometer chips (they won't be available until next year), whereas ARM designs are only just beginning to move toward 16 nanometers, but these probably won't be available until 2015 at the earliest.

Shrinking the transistor size not only helps increase the processing power and speed of a chip, but it also reduces power use by reducing the distance electrons must travel while the chip performs its operations. That's why Intel has been so keen to claim that its 22-nanometer Xeons boast a 45% improvement in energy use over earlier 32-nanometer chips. As Intel moves down to 14-nanometer designs, it's likely to enjoy further improvements in energy efficiency. Google might eventually be able to design a low-power ARM chip for its massive server operations, but by the time it does, Intel is likely to be at least two chip generations ahead, if not more. Google might not necessarily like having to pay $500 million to Intel each year to keep its server farms humming at optimum efficiency, but it does so for a reason -- Intel's chips work better. That might change in the future, but Intel's betting at least $10 billion a year in dedicated chip R&D funding that it won't.