This article is more than 1 year old
Microsoft should buy Rackable instead of building custom computers
Build Versus Buy
The decision as to how far you're willing to go to mimic Google is a tough one.
The public has no idea how effective Google's data center operations really are at reducing power consumption. It could well be the case that the time and money Google spends fiddling with its own systems outweigh the gains from a reduced power bill.
In addition, Google's go-it-alone strategy is rubbing a number of infrastructure companies the wrong way. Only a handful of top providers exist for cooling, power and data center infrastructure systems. Our sources tell us that Google's lofty demands and penny pinching have irked many of these providers to the point that Google is finding it difficult to recruit partners for new data center build outs.
Google must also presumably eat the costs associated with component price fluctuations, while other service providers that buy gear from the Tier 1s will benefit from static prices, as the server vendor itself eats the higher component costs.
All that said, Microsoft must consider the notion that Google does achieve significant cost savings through the bespoke approach and those cost savings place Google in a better competitive position. If that's the case, then Google will only increase its lead over rivals in the coming years as service providers buy millions instead of thousands of servers.
The strongest argument against building homegrown systems is that it's not a service provider's strength. Companies usually fail when they stray too far from their core business.
By purchasing Rackable, Microsoft would turn building servers into a core business and immediately gain a leg up on all the other service providers except for Google.
Of course, there's room for the likes of Amazon, Yahoo! and others to eye Rackable as well. These service providers have all bought large amounts of systems from Rackable and need every competitive edge they can get. Even when you're pretty darn good at what you do, you can pay to get better through talent. Just ask Apple, which bought PA Semi's chip brains for $300m to give the iPhone an added kick.
And the need to remain vigilant against Google is crucial. The ad broker has demonstrated a willingness to overpay for companies where necessary just to move faster than the competition. Through the PeakStream acquisition, Google also showed that it will purchase a tool with potential for industry-wide application just so its engineers can play around with the technology in-house. That has to intimidate rivals.
Given that Chrapaty has already confirmed Microsoft's plans to require custom designs from hardware makers, it would appear that the bespoke play is well underway. But, rather than being a sloth-like follower, Microsoft has the chance to shift from dabbling with custom hardware to grabbing the idea with full force and actually putting pressure on competitors. All it would take is a large check - one Microsoft can write without even noticing the missing money. And, it could be the case that Microsoft saves $1bn in data center costs just by having the Rackable team in-house to aid with more efficient designs at the centers already being built.
A software company buying a hardware maker? Yes, that's what it has come to in this day and age. Those old lines of demarcation that defined the last major infrastructure boom are no longer relevant. ®
Many people will argue that Microsoft and other large service providers can just play the major vendors off each other and gets massive amounts of very low cost systems. Keep in mind, however, that this could leave the rivals one to two generations behind Google or a company that's willing to take a more radical approach.
Even with their more dramatic cloud/utility systems, the major server vendors make certain compromises and are not matching what Rackable or Google, presumably, can do today. Microsoft is talking about getting 1,200 watts per square foot out of its new data centers. We doubt that IBM's iDataPlex system, for example, will get Microsoft to that point. But a 96 per cent efficient custom power supply from Rackable and some super, super-dense systems might.
At some point, we're talking about cost versus innovation and competitiveness and where you make those trade-offs. If Microsoft wants to dominate the next wave of computing, it had better push the envelope rather than do what everyone else is doing.