A Bloomberg article hinted that Google (NASDAQ:GOOG) is seriously thinking of creating its own custom server processor. The author said Google plans to use the technology of ARM Holdings PLC (AMRH) to create an ARM-based processor for its global server infrastructure.
Another writer here at Seeking Alpha already dismissed this as baseless rumor, but a little investigation on my part convinced me that there's substance to the Bloomberg article. I believe Google has been planning to create its own processors for a long time now. I dug deep into Google's history for the past 10 years and came up with two big acquisitions that confirmed my initial suspicions.
Google bought a startup called PeakStream in 2007. PeakStream specializes in creating programming tools for multi-core processor software development. After which, Google made another big acquisition when it bought ARM-based chip designer Agnilux in 2010. These two events are enough evidence for me to give credence to the Bloomberg article.
Why Google Needs To Create Its Own Custom Processor
Google's core business is advertising through its search engine. Maintaining its lead in web and mobile search requires the company to build a massive hardware infrastructure of servers. Maintaining and upgrading the servers in Google's global data centers is one of its biggest overhead expenses.
Google is reportedly Intel's fifth biggest customer, buying a total of 4.3% Intel's (NASDAQ:INTC) processor production to power Google's data center servers. Google spends billions of dollars to buy processors but the purchase cost of processors is only a part of the overhead. The bigger spending outlay goes to the electricity consumption in operating millions of processors.
Google also has to pay for the power or electricity used to cool down millions of servers and their processors. The company therefore is keen to reduce its power consumption overhead. One optimal way to do that is to shift the servers to more power-efficient processors.
ARM is Fit for Heavy Server Duty
Bloomberg's article that said Google will use ARM in its rumored processor design has merit. ARM-based server chips, like their smartphone CPU counterparts, are very power-efficient. The smallest reduction in energy consumption can still mean millions of dollars in savings for Google's server infrastructure.
Google's decision to use ARM in its proposed custom processor design also makes sense because it costs less to make a custom chip based on the ARM architecture. Designing it will also be much faster than the 3 or 4 years it takes to make custom x86 server processors.
Intel's Xeon has ruled the server market for years now and some people are still doubtful about ARM's fitness for the server market. However, other companies are already rolling out powerful but power-efficient ARM-equipped servers, signifying that there's a market for them now.
Intel's rival, AMD (NYSE:AMD) is scheduled to ship out 64-bit ARM server processors next year. AMD states that its 64-bit processor will be four times faster and more energy-efficient than its existing quad-core x86 Opteron X-series of server processors.
Design & Manufacture or Design & Outsource
The only thing I'm not yet sure is how Google will go about its custom processor. It already has a great design team in Agniilux. The engineers who formed Agnilux were employees of PA Semi (Palo Alto Semiconductor) which Apple (NASDAQ:AAPL) bought in 2008. They were unhappy with Apple's pricing of the stock options it granted them so they left and formed Agnilux.
Google bought Agnilux because of its engineering expertise in making low-power processors. The search giant also owns PeakStream where its core strength is in making software tools to optimize the computing power of multi-core processors.
The in-house talent and technology are already present. Google can design its custom processor any time now. What it lacks is the facility to manufacture these processors. Will it buy a chip maker or just sub-contract the production to a third party? Since these chips will be for its own in-house use, I think Google will prefer to have 100% control on the design and production so it can implement everything that it needs in a custom processor.
There's merit to the rumor that Google is considering designing its own processor. It already spent millions of dollars to acquire the talent and technology to do so. There's a very compelling reason for Google to have its custom processor - it wants to lessen the energy-consumption expenses it incurs for the vast network of data center servers it maintains.
Buy more GOOG. Savings from less power consuming processors can significantly lower the annual overhead expenses of Google. The money saved can be used instead for more R & D on monetizing ads for mobile gadget users.