May not be redistributed commercially without the author's permission.


Boxes without Windows

Arnold Kling, "Arguing in My Spare Time", No. 8

February 28, 1998

The previous two essays explained why I am skeptical about the ability of Java to overthrow the existing order in computing. Here I will examine another area where the hegemony of Microsoft might be threatened: hardware.

My impression is that Microsoft succeeds to the extent that the market values versatility in the operating system and in computer applications. What some people deride as "bloatware" is in fact a very successful approach to a market in which people with disparate needs all use a similar hardware platform.

Today's personal computers perform a tremendous variety of tasks. They communicate, via local networks and the Internet; they prepare documents; they can be used to develop and provide multimedia education and entertainment; they work with a variety of peripheral devices.

In theory, we could get along with a machine that only can communicate. The network could provide access to software that prepares documents, performs multimedia functions, and connects to peripherals. If all we want is a machine that can communicate, then its operating system can be simple, and the need for Microsoft software becomes less compelling. This is the idea of the "network computer."

A network computer might not necessarily be "dumb" in the sense of having very little processing power. The software that it uses to communicate with the network probably would be at least as sophisticated as today's Web browsers. Browsers are getting to be fairly complex applications, although not nearly as complex as an entire operating system. Moreover, if communication is its primary role, then the network computer might need to be able to perform speech recognition, which requires considerable intelligence.

The distinguishing feature of a network computer is not necessarily going to be its lack of either a powerful processor or a hard disk. Its distinguishing feature is going to be its lack of versatility. Instead, it will gain versatility from the network.

Some relatively standard economic concepts can be applied to the issue of network computers. In particular, the terms "economies of scale" and "economies of scope" come to mind.

There are economies of scale when the average cost falls as the number of units produced increases. Intel and Microsoft both face tremendous economies of scale. The cost of building the first chip of a new design may be as much as a billion dollars, if one includes the cost of building the manufacturing facility. However, the more chips that are produced of a given design, the lower the average cost.

Similarly, the cost to Microsoft of making the first copy of some new software may be tens of millions of dollars, taking into account development cost. However, the marginal cost of distributing additional copies is low, so that average cost falls sharply.

Economies of scope refer to the average cost of producing two different goods. For example, if it is less expensive to produce two different computer chips in the same facility than it is to build separate facilities, then there are economies of scope.

Here, we are adapting the concept of economies of scope to apply to the issue of providing products that can peform multiple functions. There are economies of scope (in this context) when the average cost of providing two functions in one product is less than the cost of providing those functions in separate products. For example, my printer/fax/copier/scanner would appear to take advantage of economies of scope.

Often, versatility is not economical. That is, there are dis-economies of scope. Other things equal, diseconomies of scope imply that it would be less expensive to perform two functions using two separate products than using a single product. Diseconomies of scope explain why we do not have a combination of a toaster and a washing machine.

I would venture a conjecture that the computer industry provides products with more scope than is economical, other things equal. Companies add to scope beyond the point where pure economies of scope exist. They continue to add features to products even though they have reached the point where those features could be provided more efficiently in separate products.

The reason that companies absorb diseconomies of scope is that this allows them to benefit from economies of scale. Greater scope increases the number of consumers who can be satisfied with a product, and therefore it increases the volume of the product that can be sold.

Imagine that there were no economies of scale in producing computers or computer software. Would we all use PC's that have the same operating system and similar chips? Probably not. Would we see a single word processing program that tries to be all things to all possible users, rather than different programs catering to different classes of users? Probably not.

The network computer appears to be likely to fail because the economies of scale overwhelm the diseconomies of scope. As long as there are enough people who want full-featured PC's, then PC's start out with a scale advantage over network computers. It will be impossible for network computer makers to achieve sufficient economies of scale to undercut the PC market. The cheapest network computer will be a PC where somebody does not bother to use its stand-alone functions.

Another case for network computers is that supposedly they reduce the long term "cost of ownership." However, it seems to me that the cost of ownership of PC's vs. network computers depends more on software pricing policy than on hardware design.

The long-term cost of ownership is related to the cost of sofware upgrades. Are network computer advocates saying that it necessarily will be the case that software upgrades will be free in the network world? If so, then what will be the incentive for software companies to improve their products?

In a sense, thinking of software upgrades as a cost of ownership may be misguided. The option to upgrade software might just as well be thought of as a benefit of ownership.

While the installed base of personal computers may not be threatened by the "thin client" concept of the network computer, on the server side it seems to me that the diseconomies of scope may be significant.

Those of us running Web servers need products with much narrower capabilities than what is embedded in complex operating systems, such as Unix or NT. Because the Web server market is so large, it seems plausible that a company will develop a system that is dedicated to supporting Web servers. This system will support remote operation, communication via internet protocols, and connections to other servers such as databases. But it will not need to administer a company network, interface with printers and other peripherals, support legacy applications, and carry all the other baggage that burdens all-purpose servers.

The advantage of a "thin server" would be speed, reliability, and scalability. By focusing the operating system on supporting the functions needed for a Web server, I would expect that designers could focus on these key features and build a better product.

What I am suggesting is that servers have much more prominent diseconomies of scope than client PC's. If this is the case, then versatility may not be a virtue with servers. That in turn suggests that Microsoft may be vulnerable in this arena. When the best solution is a versatile product, Microsoft has an advantage, because of its experience and skill at developing multi-featured products. However, when the best solution is a simple product, other companies may be able to compete more successfully.