Computers: Why the party's over

The computer and semiconductor-manufacturing industries are central to any inquiry into the reason for the acceleration of U.S. productivity after 1995, since together they account for roughly one-quarter of the national productivity growth jump. Both sectors already had very high productivity levels and growth rates: Productivity growth in computer manufacturing accelerated from an annual rate of 27 percent in 1987?1995 to 60 percent in 1995?1999; in semiconductors productivity rose from 43 percent to 66 percent.

Nearly all of the productivity acceleration in computer manufacturing stems from the industry?s success in selling more powerful computers, a phenomenon reflected in labor productivity data, not higher prices, since powerful computers become cheaper as technology develops. More powerful computers resulted from technological advances in inputs, such as microprocessors, memory and storage devices, and from the integration of new components, such as CD-ROMs and modems, rather than from operational changes undertaken by computer manufacturers. Even so, they were credited with much of the gain in productivity.

Productivity growth rates, though very high when measured by units per employee, accelerated only slightly after 1995. The improvement was due mainly to the simplification of computer architectures and the standardization of components. In every generation of computers, the number of semiconductor chips per computer has decreased. That decline, together with the integration of many analog and digital components into a few standardized semiconductors, reduced the time required to build a computer and thus made labor more productive. So did the outsourcing of many of the industry?s lower-productivity manufacturing activities to overseas manufacturers.

Competition upstream
But the more fundamental question is what made it possible for the upstream component industries, and in particular the semiconductor industry (which includes microprocessors), to accelerate their performance improvement? In semiconductors, the bulk of the labor productivity jump resulted from an acceleration in the performance of the mix of chips sold. That acceleration could be attributed to a shortening in the time cycle of Moore?s Law, but our research indicates that the increasingly frequent release of new chips is the prime explanation: As the interval between generations of chips shortened, the average performance of a mixed basket of chips moved closer to that of the most cutting-edge ones. Reinforcing this trend, consumers increasingly favored cutting-edge chips.

The decision to increase the frequency of new-chip releases was a strategic response by Intel, primarily to stronger competition from Advanced Micro Devices. Before 1996, AMD had produced several Intel chip designs under a disputed contract. In January 1996, the dispute was settled, and each company?s strategy evolved in its own way. Intel concentrated on Pentium designs, which AMD had no legal right to produce, and AMD upgraded its microprocessor design capabilities by purchasing NexGen. By 1999, AMD had reduced?from over 18 months in 1995 to almost nothing?the lag time between Intel?s release of a new design and the release of AMD?s competing chip. That prompted Intel to schedule more frequent releases, bringing the market closer to the cutting edge.

Intel?s decision was partly facilitated by improvements in the economics of producing new chips. The production of microprocessors is highly sensitive to environmental disruptions (from dust, for example), flaws in chip design, and mistakes in the fabrication process. These factors can strongly affect the yield of usable chips. For a new design to be economically viable, 70 to 90 percent of the chips produced must be of marketable quality. After 1995, companies in the industry decided to improve their manufacturing processes by exploiting technological advances--in manufacturing equipment, simulation, and wafer inspection--that reduced the number of production runs needed to get to a marketable yield. This approach allows manufacturers to reach profitable scale sooner and slightly softens the impact on margins of shorter product life cycles.

As all this was happening, competition in the market for memory chips (dynamic random-access memories) was increasing, albeit in markets outside the United States. The result--rapid price declines and an acceleration in the amount of memory being offered on all computer platforms--benefited labor productivity and the value of computers sold.

Spike in demand
Extraordinarily high demand for more powerful computers was also an important part of the story. A total of $1.25 trillion was spent on new information technology systems from 1995 to 1999. Approximately $350 billion of this sum involved responses to extraordinary events, including Y2K investments, the growing penetration of personal computers in consumer and business markets (a phenomenon driven by a desire to access the Internet), and the creation of corporate networking infrastructures. Moreover, the growing memory and speed requirements of new application software and of Microsoft?s Windows operating systems raised the frequency of computer upgrades.

But the tide has since turned. Over the next three to five years, fewer software upgrades and the near saturation of the consumer and, especially, the business markets could push growth below what it was before 1995. The inevitable result is declining productivity growth rates in computer manufacturing. Productivity growth will also slow in the semiconductor industry--though not as much, because of sustained international demand for microprocessors and other chips and of continued performance improvements.

For more insight, go to the McKinsey Quarterly Web site.

Copyright © 1992-2002 McKinsey & Company, Inc.