Inside the Windows Bureaucracy

The Seattle Times has a fascinating article that nicely illustrates the inefficiencies of central planning:

It’s worth noting just how complex Vista became. BusinessWeek estimates it took 10,000 employees about five years to ship Vista.

In an interview with Microsoft Chief Executive Steve Ballmer a few weeks ago, I asked if he had added up how much money it cost to develop Vista. He laughed, “I can’t say I have. It would be impossible to count up. … I’m sure it’s a lot.”

If we assume Microsoft’s costs per employee are about $200,000 a year, the estimated payroll costs alone for Vista hover around $10 billion. That has to be close to the costs of some of the biggest engineering projects ever undertaken, such as the Manhattan Project that created the atomic bomb during World War II. And while Microsoft toiled on Vista, its stock price stayed flat.

So many things went wrong with the building of Vista that it’s hard to know where it all started. The original code name was Longhorn, kicked off in 2001 after Windows XP shipped.

The company tried to pioneer on a lot of fronts, trying to change the code language used to write the operating system and fiddling with the basic file system the software uses as its foundation.

It pondered many ideas for 3-D interfaces that would help users navigate the computer more easily. Not everything worked. After a few years, the company aimed lower.

Ballmer says Microsoft tried to innovate too much. So the company reorganized and tried to placate impatient consumers by shipping Service Pack 2 for Windows XP then rebooted the whole Vista effort in mid-2004. It’s hard to imagine exactly how much Microsoft flushed down the toilet.

Two billion dollars per year. Compare that to Ubuntu, which is reportedly subsidized to the tune of about $10 million per year by Mark Shuttlesworth. Microsoft spends about two hundred times as much on its operating system as Shuttleworth spends on his. Shuttleworth’s company, Canonical, will only have to earn about 1/200th of Microsoft’s revenue in order to turn a profit.

In many ways, Vista is still a much better operating system than Ubuntu. It’s got a more polished user interface, it’s supports a wider range of hardware, and there are certainly a good many features that are supported “out of the box” in Vista, while Ubuntu requires you to fiddle with apt to get equivalent functionality–if it’s available at all. But Vista is nowhere near 200 times as good as Ubuntu, and in recent years, and Linux distros have been improving more rapidly than Windows has.

What’s going on here, fundamentally, is that organizing software development in a top-down fashion doesn’t scale terribly well. An operating system is too complex a beast for any one person to understand it in all of its gory detail. So when you organize the software development process in a huge, 10,000-programmer bureaucracy, it behaves just like any other bureaucracy. The people at the top are out of touch with what people at the bottom are actually doing. Resources are wasted on duplicated effort and poor coordination. Deadlines get missed. Code has to be tossed out when it becomes clear it won’t work as the managers expected.

So far, network effects have insulated Microsoft from these diseconomies of scale. The Windows development process might be 10 times as expensive as the alternatives, but as long as they enjoyed 20 times the market share as their nearest competitor, that wasn’t a big deal.

But that won’t last forever. Microsoft’s competitors are gradually learning how to organize ever more ambitious software projects in a decentralized manner. It’s only a matter of time before they figure out how to build an OS that’s as good as Windows at a tiny fraction of the price. And when that happens, Microsoft will no longer be able to get away with spending $10 billion to produce an operating system.

SHARE: 12 comments