A report on Bloomberg report claims search giant Google is looking to ditch Intel as a chip supplier and make its own CPUs, using a custom processor developed with an ARM license. Assuming that even is true, don't hold your breath on that one happening any time soon.
The amount of work to be done would be massive, expensive, time consuming and probably have little to no return for years if not decades. In the process, Google might just weaken itself, although it has no real competitors to be much of a threat.
Google has massive data centers around the world, many of them larger than a football stadium. To cut out costs, the company stopped buying servers from the big three vendors – IBM, HP and Dell – and began making its own stripped down, bare bones server designs. It redesigned the standard rack motherboard, eliminating things like the video port, excessive USB ports or other ports and certain chips on the motherboard.
Because it outsources motherboards, it has to buy CPUs directly from Intel. Bloomberg said Google is Intel's fifth-largest customer, accounting for 4.3% of Intel's total sales.
Now, if Google went the ARM route, it would save a whole lot of money. ARM licensees pay a fee on every chip sold. Well, Google won't be selling the chips, it would be for its own consumption. Google might even negotiate for no fee, just a one-shot license purchase, since it's not selling a product. I don't know how the license fee would work out but it would have to be a lot better than the hundreds of dollars for a Xeon E3000 series processor.
Second, it would save Google probably in the millions of dollars in electricity fees, since ARM processors typically consume under 10 watts of electricity, while the Xeon is around 75 to 90, unless you get the low-power version, which is 35 watts but is also underpowered when compared to the mainstream Xeon. Multiply that over the millions of chips in data centers and the savings accumulates rapidly.
Now for the down sides. Google needs a 64-bit chip, and there isn't one available. Oh, some have tried, like Apple and Qualcomm, but those are not for sale nor are they fully 64-bit, and they don't have many of the extras that a server processor demands, like error correcting code (ECC) memory or reliability and availability features such as failover and load balancing. A server processor is a lot more than just an instruction processing device, it has to keep the server up and running no matter what.
ARM is working on a 64-bit design of its own but won't ship it until late 2014 or early 2015. Then comes the qualifying it with motherboards, and all of the internal gear used in data centers, and then the software, and then the services… you get the idea. It will be a very, very long and expensive process.
Then there's the issue of who will make it in the volumes Google demands. TSMC is overloaded as it is. Globalfoundries might not have the capacity. And Intel certainly won't make them.
There are already ARM-based servers in the works, like HP's Moonshot and Dell's Copper. These are in the early stages and are used primarily as front-end systems to back-end services. You don't need a powerful Xeon processor just to present an HTML page and take some input. However, we don't have any sales figures yet for these servers, so there's no way of knowing how well they are doing.
I don't doubt that Google has the engineering skill to pull something like this off, and it will undoubtedly do its own custom silicon for the motherboards, perhaps to improve parallelism and inter-server communication. And since Google owns the Motorola smartphone business, I can see it doing chip design there as well. But making its own CPU that's a worthy alternative to Intel? The investment is just too great.