IDC Forecasts Server Sales Declines Until 2011; Blame X64 Boxes
June 22, 2009 Timothy Prickett Morgan
In normal times, the server market is a leading indicator of sorts for the economy. But we do not live in normal times, either with respect to the global economy in general or to the server space in particular. The economy has been on the rocks since late 2007, and the economic meltdown last summer certainly has put a damper on IT spending, especially at large enterprises. But a far greater problem, in terms of server revenues, might be the mainstreaming of virtualization and the advent of cloud computing, both of which allow companies to support growing workloads with a lot fewer servers and considerably less money.
Market researcher IDC took some future wind out of the sales of the server market last week when it put out a forecast that shows revenues across all system types and sizes will decline not only over the next two quarters, but through the end of 2010. Many server makers had been hoping for some good news in early 2010 or maybe even the last quarter of 2009. But the continuing poor economy and the mainstreaming of virtualization in the X64 server space have combined to put a serious damper on server revenues this year. IDC is now projecting that worldwide server sales will fall by 22.1 percent to $44.5 billion this year, the worst decline in the history of the systems business. And it looks like the second quarter is shaping up to be the worst in history.
“IDC has lowered its 2Q09 forecast by 8.5 percent or nearly $1 billion from our previous forecast as the depth of the worldwide recession increased in the first half of 2009,” said Matt Eastwood, group vice president of Enterprise Platforms at IDC, who put together the forecast. “Although we are now forecasting a 22.1 percent year-over-year decline in server spending for 2009, the worst of the market contraction is behind us. In fact, by the end of the third quarter this year, nearly 90 percent of the cumulative market contraction will have been realized as the market begins exhibiting significant signs of stabilization. It’s important to note that IDC believes many IT users will begin making strategic compute platform decisions during the remainder of 2009 in advance of improving business conditions and server demand in 2010.”
I am not so sure that IDC’s own numbers warrant that optimism, and I think the situation could turn out to be far worse than even IDC expects a few years from now.
First, let’s just go over IDC’s projections. Here are the numbers:
First of all, the server market hit that low level of revenue in 1998, on the way up to the 2000 dot-com peak, and 2001 on downside after the dot-com bubble burst. IDC pegs the peak at $67.5 billion in sales in 2000, so look at how far we have fallen. (If you want some history of server sales in the late 1990s and early 2000s, I went through it all here back in 2003.) This year’s server sales will basically match the $44.3 billion in worldwide sales in 2001 (when the market only contracted by 11.6 percent). In some ways, the clock has really rolled back to the early 1990s. And as you can see from the table, it looks like we are going to be staying somewhere around 1996 or so from here on out. (In the IDC taxonomy, a volume server costs less than $25,000, a midrange server costs between $25,000 and $500,000, and an enterprise server costs more than $500,000.)
The systems space undergoes radical phase changes every now and then, and it has a dramatic impact on sales for specific platforms, which are reflected in the sales figures. The AS/400 was hit hard by data center consolidations in the early 1990s (driven by cheaper telecom) because the subsystem architecture of the OS/400 operating system allowed many different applications to run side-by-side on a single system; logical partitioning, introduced in 1998, caused further contraction. Add on the steady progress of Moore’s Law on all of the electronic components of the system, and revenues get crunched even further, and then add in the effect of the installed base bleeding off customers (somewhere around 75,000 since the peak in 1998, when there were around 275,000 unique AS/400 shops in the world), and it is no surprise that the AS/400 business that used to generate $5 billion in hardware sales in the early 1990s now is lucky to do over $1 billion as the Power Systems i.
If you believe in comeuppance–and I certainly do–I get a small sense of satisfaction that the Windows and Linux communities and reseller channels seem to have no idea of the crunch that is coming their way. AS/400 shops, as well as Unix shops which were also relatively early adopters of virtualization but are still a few years behind the AS/400, have lived through the Virtualization Crunch. They have been compressed, and when the economy does recover–perhaps at the end of 2009 we’ll see some signs of life and budgets for upgrades and new systems set for 2010. Life will return to the Power Systems i channel because across those 200,000 customers, people are going to start running out of computing oomph on their machines.
But the X64 server base, which is going to compress by a factor of maybe 5 to 1 or even 10 to 1, is still on the other side of that virtualization meat grinder. Not all X64 servers need to be virtualized and not all Windows and Linux workloads can be virtualized efficiently, of course. So the compression is probably going to be something on the order of 3 to 1. Now, do you honestly think that X64 server makers are going to be able to sell machines that cost three times as much to users to maintain their revenue streams and profits? I don’t. Eventually, all of the machines that can’t support modern virtualization will be replaced–maybe in two to three years.
And if you think cloud computing is somehow going to save the X64 server business, it will only make server sales dip in the long run as people decide to not only share their workloads on a single set of systems in their data centers, but to drive up efficiencies by interleaving workloads across different companies and pay for only the capacity they use. During the cloud buildout, to be sure, millions of servers will be sold to cloud providers. And then the phones will stop ringing once the infrastructure is built. At that point, X64 server sales will be tied to actual workload growth, and upgrade cycles both at companies and at cloud providers will get longer as each processor bump delivers much more capacity than most companies can use.
Any of this sound familiar?
I thought you AS/400 people might know what I am talking about. Well, now you get to point and laugh at the X64 market for a change. I don’t think X64 server makers and their chip partners, Intel and Advanced Micro Devices, will be laughing much in 2012 and 2013. Then again, I don’t think any server makers will be laughing much if cloud computing eats half of the server capacity in the world a few years later. What happens to the server ecosystem–and the profits that partners need to stay in business–when all servers are running at 95 percent efficiency and they have three times the performance of today’s machines at the same price?
The good news is this: the software running on virtual/logical machines and on clouds, both systems software like databases and development tools, will still be needed. Application developers will still be doing what they do for businesses, too. And a lot of those core applications running on internal and public clouds will be written in RPG or COBOL (albeit modernized) and smacking a DB2/400 database. We will have come full-circle to 1978 and time-sharing.
On more thing. Let’s talk for just one second about extrapolation and prediction. It’s a tricky business because conditions can and sometimes do change in ways that you can’t anticipate. And what I am about to say is not meant to be any kind of criticism, but a word of caution. Back in July 1999, when I was writing for Computergram, I covered an IDC server forecast that spanned from 1998 to 2003. It was just as the dot-com boom was raging, the ERP boom was roaring, and the Y2K bug was causing people to buy lots of server capacity. The server biz was hot, and getting hotter by the second. And IDC was projecting that global server sales would grow from $65 billion in 1998 to $89 billion in 2003. Java and interpreted languages were expected to play a big part in the server build-out. (Can you imagine if we had chosen to compile C++ instead? Yikes!) Anyway, we never got anywhere near that $89 billion in 2003 because the booms went bust and the world went crazy on 9/11 and spooked the global economy.
What can be honestly said is that economic hardship drives technology shifts, and if you don’t think the economy will improve, allowing us to be lazy about our systems again as we were in the 1990s, then you better brace for some changes. Because if $45 billion is the new “normal” in the server business, I am willing to bet there’s an error bar that says it can be $35 billion. Or $55 billion, for that matter.
Keep an open mind, keep your options open, too.