Over the years there have been obituaries for celebrities which have clearly been premature and really pissed off those celebrities (Bob Hope clearly was not amused).
The same thing has happened in technology. The most obvious was the mainframe, which was pronounced dead in the 1980s and remains IBM’s most profitable platform today.
We just had another occurrence last week: a story was released in Wired with the provocative title “Intel Confirms Decline of Server Giants HP, Dell, and IBM.”
While HP is clearly having some issues connected to their turn around (why don’t people get that these things take 5 to 7 years?) both Dell and IBM reported enough growth to make up for HP’s slight decline. But in looking at the numbers, IDC shows a slight decline in volume from 2008 to 2010 and no decline in revenue. The decline is 2 percent and IDC’s confidence interval is typically above 3 percent, which even calls that one decline into question.
So the idea of a decline is false. However, there are real changes going on in servers and you’d likely do well to understand them because they will impact volume numbers.
Google/Facebook Non-Trend: Build Your Own Servers
The nice, or sad, thing about following tech for a few decades is that bad ideas have a habit of coming back. One of them is “build your own hardware.” I kind of thought we’d put this behind us in the 1980s but here we are again, talking about young companies relearning the lesson that was taught by the Industrial Revolution: specialization always trumps encapsulation at scale.
You see, before the Industrial Revolution people pretty much had to do everything themselves. You wanted a house, you built it, you wanted to sell carriages, you built the entire thing from scratch, etc. But around 200 years ago we learned that companies that focused on a particular thing got something called economies of scale and “experience.” This means they not only can make a thing cheaper they can make it better than you can. This is why we don’t build our own cars and why firms can’t profitably build their own servers.
Sure, you can build a car, but you’ll just find that you can buy a better one cheaper. Same with servers but every decade or so a company comes up thinking, gosh, look how easy it is to build your own and bypass an OEM.
Now with PCs you could do that, and what you get is a firm like Acer or Samsung, which are both an ODM and an OEM because they then build out that business to scale. The issue with servers is that with the exception of the very low end, the OEMs are the ODMs; you are basically buying direct.
This simply means that Facebook and Google haven’t either implemented activity-based cost measurement or hired an experienced CIO, which is why they are repeating a mistake common with young companies or inexperienced CIOs. Eventually someone will point out the waste and they’ll bid out servers with the rest of us.
In the meantime, research firms measure servers that are sold through channel and since these servers don’t go that way they shouldn’t be impacting the numbers (except as an invisible drag) anyway.
Real Server Changes
I kind of think the folks at Wired were perhaps looking too hard for bad news at IDF. I just don’t see the support for this declining position that they do.
This doesn’t mean there aren’t server changes going on. You’d have to be blind to miss that we both have a massive consolidation/virtualization effort that is accelerating and an offsetting massive desktop virtualization effort going on.
The first forces volumes down but revenue per enhanced server up. The second is driving straight new server growth.
Desktop virtualization appears to be lagging consolidation, as the IDC numbers may be pointing out (remember the 3%+ confidence interval). But this should begin to reverse itself next year as BYOD (Bring Your Own Device) efforts accelerate and virtualization becomes the only cost effective and secure way to address this trend.
In addition there are services like Azure and Office 365 that are accelerating (we switched to Office 365 ourselves this year). And that too will drive different buying behavior (service providers like Microsoft buy servers from the big three server vendors, but much larger servers).
So market makeup will change. But given that most of all this points to a move to large consolidated virtualized servers, it tends to play into the strengths of the big three – not against them.
So announcing the death or decline of the big three server vendors, as a group, is really premature.
However, there are clear trends to be aware of toward virtualization (servers and desktops), consolidation, cloud services, and even new server types and vendors (Oracle/Cisco) playing in the market.
So Wired was right: there is change you should be aware of, just not a decline.
In short, Intel couldn’t confirm a decline that isn’t actually happening. Servers aren’t declining, they are evolving, and that’s good thing. The real trend is that young companies tend to have to relearn lessons and Google and Facebook will probably not be the last.