The Four Hundred

Every time the economy slows, the companies that sell the cheapest servers see a surge in sales, while the stalwart established vendors see a decline. These economic cycles have driven the acceptance of server platforms as much as technological advances--plug compatible mainframes in the late 1960s, minicomputers in the mid-1970s, Unix servers in the late 1980s and early 1990s. And now Wintel and Lintel iron, in the early 2000s, is getting a push from bad times.

According to research by IDC, spending on Intel-based server platforms grew to $5 billion in the third quarter of 2002, the first time since early 2001, when this sector of the server market saw revenues increase. Price competition is heated in the Wintel and Lintel server markets, so achieving any revenue growth at all is something of an accomplishment. Sales of all servers will be flat at best for calendar 2002, IDC predicts, and that is if the fourth quarter holds up. In 2001, the server market contracted by 20 percent, to $54.5 billion, and the third quarter was the deepest part of the trough, with only $12 billion in sales. While seeing improvement in the third quarter is heartening, it is hard to imagine a worse time in the server market than the third quarter of 2001. If 2001 server sales levels had contracted to levels not seen since around 1994, the third quarter of last year was pushed back to the early 1990s, when mainframes and minicomputers still ruled the world.

The analysts at IDC say they expect spending on servers in the United States to have increased by 4.7 percent in the third quarter of 2002, and up sequentially from the second quarter by 8.6 percent, with sales of entry servers being more robust than for other platforms. IDC says that most vendors of so-called volume servers--which is shorthand for Intel platforms--were strong in July and August, and the market researchers say further that requests for proposals and quotations across several vertical markets were up in the third quarter. However, because the first half of 2002 was pretty weak in the U.S., server sales across all platforms are expected to decline by 9 percent in the U.S. for all of 2002. We're turning the clock back to the last recession, it seems. Vernon Turner, group vice president of global enterprise server solutions for IDC, cautions that the server market has not broken free of the economic freeze and that companies are buying the minimal amount of capacity that they need, but this is an indication that the market is improving. Turner's team of analysts reckon that conditions in the U.S. and in some emerging markets in the Asia/Pacific region are improving, but that conditions in Latin America, Western Europe, and Japan are not doing so hot. IDC says further that the fact that customers are buying smaller increments of processing capacity is causing aggressive pricing across all server tiers.

Earlier this year, IDC predicted that server revenues worldwide would hit $70 billion in 2006, but its latest forecasts call for the server market to grow at an annual compound growth rate between 2001 and 2006 to reach $63.4 billion. (I think that even this is optimistic, to be honest.) Under its current server market forecast, IDC expects sales of Linux servers (mostly on Intel platforms) to triple from their current levels to $6.5 billion in 2006. Over the same time, sales of Windows servers (exclusively on Intel platforms) will increase by 36 percent, to $19 billion, an increase in sales of $5 billion. Unix server sales will be crowded out of the entry and midrange sectors of the server market by 32-bit Linux and Windows servers and to a certain extent by 64-bit servers based on Intel Itanium and Advanced Micro Devices Opteron chips. But Unix servers will still comprise $27.7 billion in server sales in 2006, according to IDC's estimates, or about 44 percent of the total worldwide market. That leaves another $10.2 billion in sales for IBM proprietary mainframe and midrange machines, Hewlett-Packard Tandem Himalaya and some OpenVMS servers, and a dwindling number of other exotic HPC and parallel machines.

It remains to be seen what share IBM's iSeries platform can get of this future server market, but it seems likely that IBM's Power-based servers, whether they bear the iSeries or pSeries (and maybe even the zSeries) labels, will be one of the two or three dominant platforms in the world, even in 2006 and beyond. If HP can successfully consolidate its various Intel-based server lines with its HP 9000 and AlphaServer lines down onto a single 64-bit Itanium-based platform, this will be the other dominant platform in the world. Sun Microsystems' UltraSparc-based servers will be the third platform, and possibly quite a bit behind IBM and HP, in terms of total sales but not in terms of ardent supporters. That Power and UltraSparc will be around is a foregone conclusion, while HP's Itanium future is still a work in progress. What iSeries shops need to remember is that the OS/400 platform will be supported so long as IBM has iron that runs it. There is no question about this.

Whether or not IBM will aggressively push iSeries sales to try to get a bigger share of the server pie is a real question, and one that many of us have been asking IBM for a decade. IBM's strategies in the second half of 2002 seem to be inward-looking--the company is trying to reconnect to its 250,000 customers, many of whom have been ignored by IBM and its partner channel for five or 10 years. Attacking its own installed base of customers with vigor is to be commended, but this is not the same thing as getting new customers. Intel iron gets new customers every day, and there are good and logical reasons for that. Intel iron may not be the best solution in the world, but it is cheap. This is what seems to matter most to customers these days, and IBM seems to be coming to this realization, however slowly, as evidenced by price cuts on iSeries processors and features in recent months. I've got a few ideas about how an iSeries turnaround might be engineered, which I will be sharing with you in the coming weeks.