Sometimes markets change through the appearance of a thoroughly disruptive new technology, but more often the change comes about because of a collection of events and market forces. Just such a collection looks as though it is in play for the semiconductor industry.  

And what has this to do with servers and infrastructure? Well, it looks like the events and market forces are being generated in just those areas and the result is growing pressure on the makers of processors and support devices for the server marketplace. 

It would appear that at least some server vendors are a little miffed at the likes of Intel about the amount of business it is now generating out of selling direct to what should be, in their eyes at least, `their customers’. There have been, historically, a large number of sensible, economic arguments against `customers’ building their own systems rather than buying them in from a specialist vendor or two. But those arguments are now getting close to running on empty and for some of the biggest customers, such as Google, the tank is already dry and they have moved on. 

Google is already making its own low cost servers to its own spec and, given the volume that it consumes them there is an obvious economic justification for it. But if other users and other server vendors base their judgements on just that `volume requirement’ justification, and so dismiss Google as a statistical freak rather than a true market dynamic in the process of change, they will be missing an important point. 

The issue now is not just that some big users are starting to build their own servers rather than buy – just because they have large datacentres and it therefore makes economic sense. It is that large datacentres – very large ones – are starting to make important economic sense in their own right. That means the economic arguments are starting to point at every user having the largest datacentre possible. But that in itself is all-round economic nonsense unless the next obvious step is taken – not actually have their own, but share the services available from a few extremely large datacentres. 

The reason is straight up and down economics of the technology. Datacentres are now not only avaricious and expensive consumers of electricity, they are being identified as such. And despite much hand-wringing there is in fact not a huge amount yet being done to solve the real and perceived problems. But there are solutions available.  

Shared, virtualised servers have better utilisation and make for reduced resources – less bucks for each bang. Each server works better under more pressure – power supplies, for example, are at their most efficient when nearly busting a blood vessel to keep going – so more bangs for each buck. As management tools continue to improve, more systems can be managed by the same number of staff, and the bigger the datacentre the more real hotshot specialist staff it can justify employing, improving service provision and reducing downtime.  

SaaS as an alternative to onsite datacentres centres and `owned’ applications may as yet be a small part of the market but is showing the way in how to combine the reduction of operating costs and the increase in operational flexibility. Scale that up and there does come a point where pure economics is likely to be the lever that fundamentally changes the buy/build argument about servers and datacentres. The answer, for all but the big service providers, will be neither – just sign up for a service instead.