The U.S. Software Industry:
An Analysis and Interpretive History
W. Edward Steinmueller
Professor of the Economics of Technical Change
Maastricht Economic Research Institute in Innovation and Technology
University of Limburg
March 14, 1995
Prepared for University of California, Berkeley
International Computer Software Industry Project
Directed by David C. Mowery
Forthcoming in David C. Mowery (ed.), "The International Computer Software Industry," Oxford
University Press, 1995.
Acknowledgements: The author is grateful to the Markle Foundation and the Center for Economic
Policy Research for financial support underlying this research, to David Mowery for his substantial
intellectual and editorial contributions to this chapter, and to others who have helped me develop
the ideas and expression of this chapter including Timothy Bresnahan, Franco Malerba, Tom
Cottrell, Salvatore Torrisi, Yasunori Baba, Peter Grindley, and Robert Merges.
I. Introduction and Economic Foundations
During the past fifty years innovations in semiconductors, data storage devices, computer
architecture, software, and data communications have revolutionized information collection,
storage, processing, and distribution, creating new industries and transforming industries inherited
from past industrial eras. Explanations of this revolution in information technology have focussed
on the extraordinary reductions in the cost of the hardware components,1largely to the neglect of
the role of computer software in these developments. Nonetheless, every application of
information technology has required complementary "software"--computer instructions that
transform the tabula rasa of computer hardware into machines that perform useful functions.2
This chapter offers answers for several basic questions about the historical development of
the U.S. software industry. First, what determines the division of labor in software production
among hardware producers, computer users, and the companies that produce software as their
primary business, the "independent software vendors?" Second, have key events in the history of
the U.S. software industry created a distinctly "American" system of software production that need
not or cannot serve as a model for the development of the software industries in other nations?
Third, what economic effects follow from the high fixed costs of initial software creation and the
low marginal costs of reproducing software? In answering these questions this chapter will draw
upon economic theory as well as descriptive material from industrial and technological
Several definitions and distinctions are useful in setting the stage for answering our main
questions. In this chapter, software is a general term of reference for instructions controlling the
operation of information technology hardware. Programmers are people that devise specific
1For example, Gordon  estimates that the annual rate of decline reduction in computer
hardware costs averaged 19% for the from 1954-84 period.
2There is no clear boundary between hardware and software. Any information processing
operation that can be achieved with "instructions" can also be achieved by an hardware subsystem.
For example, systems may be designed that use software instructions for finding the square root of
a number or, alternatively, an electronic component can be constructed that performs the same
function within the system. Similarly, many electronic systems employ "programmable"
components where a single set of instructions is permanently built in at the time of system
manufacture. The software in these systems is "embedded" in the electronic system. The
economics and industrial structure implications of embedded software, while of growing
importance, are not considered in this chapter.
collections of instructions called software programs or, simply, programs. The use of the word
"systems" is ubiquitous and unavoidable; it will be used here to refer to either complementary
combinations of hardware and software or to collections of software programs that are "inter-
operable," i.e. programs that operate together and exchange information. The technologies for
acquiring, storing, processing, and transmitting information are collectively referred to as
"information technology"and include both hardware and software components.
Software is classified as a "business service" in the U.S. income and product accounts
despite the fact that "packaged software" products more closely resemble personal computer
peripherals or books in their methods of distribution and reproduction. Software that is delivered
as part of a business service can and should be distinguished from software that is sold as a
product. Software services are performed within the programming services (SIC 7371) and
integrated computer systems (SIC 7373) industries, although the latter industry performs hardware
engineering design activities in addition to creating software. Sales of software as a distinct
product are recorded as output of the packaged software industry (SIC 7372). Although the
production of software is a labor-intensive activity, in principle software is reproducible at very
low costs relative to the costs of its creation, a characteristic that is unusual in the service
industries. This means that companies engaged in selling programming services will seek
opportunities to reuse part or all of the software previously created for previous clients in selling
service to new clients.3A particular software program that is only produced once should be
viewed as a service output, while a program that is reproduced dozens or millions of times has
development and marketing characteristics closer to those of manufactured goods. Understandably,
such distinctions are not made in assembling statistics in U.S. income and product accounts, and
the output of the packaged software (SIC 7372), programming services (SIC 7371) and integrated
computer system design (SIC 7373) sectors contain an unmeasured mixture of "one-off," "reused,"
and reproduced software in each sub-sector.4Of course, creation of software for extensive
reproduction is a primary aim in the packaged software (SIC 7372) sub-sector.
3The ability to reuse may, on occasion, be contractually limited.
4The term "reused" refers to the use of large blocks of instructions in the delivery of multiple
outputs. Reuse allows software producers to "customize" software for clients more economically
than designing from scratch. While reuse is assumed to be quite frequent, there is little evidence
about its actual extent.
A second important distinction that is relevant in examining the software industry is the
division of output between intermediate and final goods. Intermediate goods are used to produce
other goods and services, while final goods are sold to consumers. With the recent, and nearly
simultaneous, arrival of home computers, information services, and programmable consumer
electronics systems (e.g. video game systems), independent software producers and system
producers’ sale of software as a final good have become a major industry. Indeed, the conceptual
and technological distinctions between such "consumer" software products as games, and those
historically considered to be "entertainment" such as music or video products, are rapidly fading as
suggested by the increasing frequency of the term "multimedia" to refer to hybrids of video, sound,
and software. Nonetheless, most of our attention in this chapter, as in the other chapters in this
volume, is devoted to software that is an intermediate good, employed by businesses in the
production of other goods and services or sold for the same purposes to other enterprises. We will
ignore many of the problems of producing and marketing consumer-oriented software.
A third important distinction is between software and the production of other economic
commodities. On the one hand, software, like other commodities, requires inputs that have
alternative uses and, once produced, has economic value as an intermediate or final good. But,
software is also an unusual economic commodity because its marginal costs of reproduction are
very low or negligible. The low costs of software reproduction imply that society must grant
businesses some right to control reproduction (and charge higher prices than the cost of
reproduction) if investments are to be made in software creation, especially packaged software.
Otherwise, third parties would make a business of reproducing software, and competition would
drive the costs of software to the low marginal cost of its reproduction.5As a result, intellectual
property protection is a key policy influence on firm strategy and evolution in this industry, as
Chapter 10 points out, and the respective political influence of hardware manufacturers, custom
software and service providers, and producers of packaged software may influence the structure of
software-related intellectual property protection.
Computer producers, users, and independent software vendors each have distinct
incentives to produce software. Computer system producers have incentives to produce software
because software is an economic complement to the sale of computer systems, i.e. the availability
5This appears has happened in nations with no effective intellectual property right protection
for software. At the same time, the rents available from protection are limited by the competition
of "non-infringing" substitutes.
of software will increase the sale of computers. Ideally, computer producers can receive revenues
from both the sale of computers and software. In practice, hardware producers in the U.S., with
the notable exception of IBM, have received a diminishing share of their revenues from software
production.6U.S. hardware producers involvement in software creation has followed a pattern of
vertical "dis-integration’," favoring user-produced software and the entry of independent software
Despite the incentives to combine hardware and software production, the ability of
computer producers to understand and solve specific user problems is limited. The presence in the
U.S. of an enormous variety of industries that use information technology has been a stimulus to
the creation of a correspondingly large volume of user-created software. However, as in other
nations, users have little incentive to sell the software they develop to rivals.
The potential profits from widespread sales of particular software products7have
encouraged the entry of a third group of producers, "independent software vendors," (ISVs). The
boundary in software producing activities among ISVs, computer producers, and users is
determined by limits in the abilities of computer producers’ ability to increase their "span of
control" of joint hardware and software creation and by limits in the capabilities and incentives of
users to produce software for sale to other firms. These factors, along with the design and
marketing abilities of ISV firms, constitute the "infrastructure" of software creation in any
particular nation and differ across nations with different patterns of industrial development. For
example, the diversity of user-industries in the U.S. has made it very difficult for computer
manufacturers to pursue vertical market strategies similar to those pursued by some European
6Moreover, although all software is complementary in demand with hardware, some software
may raise the level of hardware demand more than others (e.g. software that demands intensive use
of the computational or mass storage hardware), and we expect to find hardware producers more
active in these areas than in other areas (e.g. in areas that permit users to conserve on their use of
7Information goods that can be protected from copying, can command prices in excess of the
marginal costs of reproduction. If a given product’s price provides more than sufficient revenues
to recover the costs of developing, marketing, and maintaining the product, the producer will earn
economic profits. Entry of independent producers is therefore likely if such entrants can overcome
advantages of computer producers in the simultaneous design of hardware and software and
computer users are disinclined or relatively inefficient at offering their internally produced software
to others. As we will see, all of these conditions eventually were fulfilled.
computer manufacturers.8This was not because such strategies were unavailable, but because
computer manufacturers’ gains from a more exclusive pursuit of hardware improvement (combined
with the vertical "dis-integration" of software production) were greater than their gains from
controlling integrated software and hardware development in a large number of vertical markets.
For example, Digital Equipment Corporation, abandoned its early position as a leading producer of
integrated hardware and software systems for newspaper publishing, ceding the hardware
integration and software development of such systems to other companies. In the U.S. market,
most hardware vendors have retreated from software production, and as described below, recent
entrants into computer production are minor participants in software production. The most likely
explanation for their behavior is that U.S. computer producers have found that ISV entry results in
a greater supply of software than would otherwise be available. The U.S. software industry has
also benefitted from the prior existence of an enormous number of small contract programming
companies that originated from the needs of larger companies and governments for custom
software production. This infrastructure has played a major role in the growth of the U.S.
packaged software industry, through its development of human capital (noted as a problem of the
European industry in Chapter 7 of this volume) and its pioneering development of particular
applications.9In the U.S. market, ISV participation appears to have fostered a faster sales growth
than computer producers could have realized from joint production of hardware and software. For
users, the presence of ISVs offers an alternative to internal production.
Change in the solutions to the "make or buy" problems of computer producers and users,
solutions that are unique to different stages of the of evolution of the U.S. computer market, along
with some key legal and policy decisions, explain the changing division of labor among producers
of software in the United States. In other words, our working hypothesis is that there is no
"natural" industrial structure for software production; the structure and evolution of a nation’s
software industry depends upon particular historical and institutional events. This chapter
examines this hypothesis by exploring the development of the U.S. software industry, beginning
with the period before its birth. During this period, the design and construction of computers as
scientific instruments were co-mingled with writing computer instructions. I bring this discussion
to the present day when independent software companies produce and sell packaged software
for large installed bases of particular computer "platforms." Although it is tempting to simplify
8See Chapter 7 in this volume.
9For example, one of the pioneering relational data base products, dBase II, III, and IV,
originated in a program developed at the Jet Propulsion Laboratory.
this account by discussing only the independent companies that produce software "products," that
approach would obscure the enormous economic significance of software production by other
organizations, including producers of hardware and the users of information technology. The
specifics of history, institutions, and participants are important, and mean that the appropriate unit
of analysis for a study of the software industry’s evolution must begin at the national or regional
level, rather than the "global market."
The outcome of these evolving make or buy decisions of computer users and producers in
the U.S. market in recent years can be briefly summarized. Virtually all of the people involved in
the information technology industry, from electronic component producers to users, write software.
Software written for internal use within companies has historically been the largest single source of
investment in software creation; this type (which is not separately captured in the U.S. national
income accounts) is followed by software produced by design service and software firms, and
software "embedded" in electronic systems.10 In 1987, the receipts of U.S. software programming
service companies (SIC 7371) were $14.2 billion, the receipts for computer integrated systems
design (SIC 7373) were $7.1 billion, and the receipts from prepackaged software (SIC 7372) sales
were $5.9 billion. The preponderance of revenues from programming services and a large share of
integrated systems design revenues are derived from the production of "specialized" software
solutions, that are unique to individual companies. By contrast, packaged software is an
intermediate good or "tool" that providing applications solutions or entertainment for large
numbers of users.
In assaying the size of the software industry, it is important to understand that the sale of
software services and packaged software does not include the investment in software creating
activities within organizations, the first of the three categories mentioned in the preceding
paragraph. Some indication of the magnitude of these activities is available by analyzing the
patterns of employment of software engineers, which are available for broad industry groups. For
example, "business services" (SIC code 73) includes prepackaged software, software programming
services, and computer integrated systems design, as well as other information processing-intensive
sub-industries such as credit reporting, advertising, and mailing services. In 1990, the "business
10 Unfortunately, embedded software cannot be covered in the present paper. Of the features
discussed here, the only one of particular relevance to embedded software is the role of feedback
in accelerating the advance of user interfaces, discussed in the section on the personal computer
revolution below. Chapter 5 in this volume attempts to estimate the size of the "user-
produced"software sector in Japan.
service" employment of system analysts and programmers was 74,000 and 130,940 respectively.
In 1988 and 1989, by comparison, 217,310 system analysts and 264,110 programmers were
employed in manufacturing and the other service industries (including state and local government).
In other words, establishments outside the business service sector employed many more computer-
related professionals (individuals who bear a major responsible for specifying and creating
software) than did firms within the "software industry." Obviously, programmers and system
analysts in SIC 73, manufacturing and all service sector activities (excluding SIC 73) are not
engaged solely or even primarily in writing software. Nevertheless, the "in-house"employment of
these occupational classes is so large that it raises the possibility that in-house software may
account for a larger share of total U.S. software production than the output of firms producing for
the (custom or packaged) market.
This chapter’s sections treat the development of the software industry chronologically.
Section two considers the origins of software creation and the early history of the software industry
from 1950-1965, a period when computer producers dominated software production. The third
section examines the critical 1965-1970 period when rapid growth in the utilization of computers,
antitrust litigation, and delays in some key technological improvements fostered enormous growth
in user-produced software and the birth of an independent software vendor sector. The fourth
section examines two themes that shaped developments in the 1970s: growing dissatisfaction of
users with the productivity and effectiveness of their internal software development efforts and the
impact of smaller-scale "mini" computers, which altered both the distribution of software-creating
activities and the use of computers within organizations. The fifth section discusses the role of the
personal computer and the rise of "software publishing" and introduces the final element of the
story to date, the role of consulting companies and information system integrators in addressing the
problems of internal software development. The sixth and final section of this chapter summarizes
these historical developments and identifies some of the forces that are shaping developments
during the 1990s. This history addresses two elementary questions noted earlier--(1) What
determines the division of labor in software production among hardware producers, computer users,
and the companies that produce software as their primary business and the "independent software
vendors" and (2) What key events in the history of the U.S. software industry have created a
distinctly "American system" of software production?
The third and final question this chapter addresses is why has software persistently been
identified as a "bottleneck" in the growth of information technology markets and as a drag on the
realization of productivity gains from utilization of information technology. This question has
both a general answer dealt with briefly here in the introduction and specific answers for the U.S.
that are main themes in the fourth and fifth section of this chapter. The basis for this question is
the observation of rapid rates of growth in the level and share of software costs for information
technology systems.11 The growth of the cost share of software has been linked to the "craft
production" techniques in the software industry that allegedly cannot match the pace of hardware
performance improvement.12 The purported result is rapidly rising costs of information
technology due to the "bottleneck" of increasing software costs, a consequence may help explain
the low measured productivity gains from investments in information technology.13
Do recent data on hardware and software expenditures support the "bottleneck"
characterization? The sales of packaged computer software and services as a share of computer
sales from 1970-1993 are depicted in Figure 1 using current dollar revenues.14 During the
11 See, for example, OECD .
12 See Baumol, Blackman, and Wolff .
13 See Roach  and Strassmann  for reviews of this issue.
14 In addition to the information provided in Figure 1, other sources provide some evidence
about earlier periods and suggest some qualifications to the empirical conclusions suggested by
Figure 1. For information on the pre-1970 relative expenditures on hardware and software, see
Phister . Phister reports results of a series of Datamation facility surveys that indicate that
purchased software costs only became significant in 1967 when they were (0.1%) and gradually
increased to 1.4% by 1974, the ending date of his report. More significantly, costs of system
analysts and programmers accounted for 20.9% of system operating costs in 1955 and gradually
rose to 32.3% by 1970 and remained near that level for the remaining four years of data he
reports. Two snapshots are available in OECD , pp. 82, 1979 and 1985. Between these two
years, total software expenditures, more than doubled (from $51.9 billion to $113.3 billion
(thousand million)) and the share of software expenditures increase from 15 to 19%.
There has, however, been some controversy over the extent of software cost increase.
Gurbaxani and Mendelson  note that the software costs reported to IDC (a leading
consulting company) by data processing managers are a stable fraction of total EDP (electronic
data processing) budgets. There are, however, three problems with this analysis. The first is that
the growing deployment of computer hardware should engage economies of scale in software use
for the shared systems under DP manager control, easing cost increases in this portion of the
company’s computer budget. Second, EDP budgets are often only a portion of the software
investments since many companies have active departmental computing programs and the share of
software in departmental budgets may differ from department data processing expenditures. Third,
it is unclear how EDP department personnel costs should be assigned since many of these
personnel are involved in programming activities. Since personnel costs are the most rapidly
increasing portion of costs, accounting for internal software production would raise the share of
software in corporate EDP costs for the years reported.
1970s, Figure 1 suggests that expenditures on packaged software and software services grew at
the same rate as hardware expenditures. Thus, unless it can be established that the amount of
purchased software utilized during the 1970s was decreasing (which is unlikely), price increases do
not provide a plausible explanation for that decade’s trends in software expenditure. In other
words, for the bottleneck hypothesis to be valid, software use would have had to decrease. The
microcomputer revolution of the 1980s accelerated growth in software sales, pushing software’s
share above 50% by 1988 and close to 75% by 1993. During the 1980s, the enormous growth of
low-priced "packaged software" for personal computers mean that growth again cannot be
explained with a cost-push or price-increase model that is consistent with the bottleneck view.15
Outward shifts in demand and relatively elastic demand appear to perform better in explaining
increasing expenditures on software than does a tightening bottleneck. This analysis, however,
does not address the costs of internally produced software.
Although an inaccurate general characterization of U.S. industrial experience, the
"bottleneck" problem has shaped the interactions among hardware producers, computer users, and
independent software vendors, and has motivated some of the key institutional and policy events in
the U.S. industry’s development. Institutional reforms have been directed at sources of cost
growth, particularly the development costs and maintenance expenses of internally-produced
software. Thus, specific responses to the software bottleneck in the U.S. included changes in the
organization of the delivery of information processing services such as the growth of system
integration and outsourcing services as well as the deployment of specific types of hardware such
as minicomputers. In both cases, efforts were directed at widening a bottleneck, although this
bottleneck may be more accurately characterized as organizational rather than stemming from
software per se.
15 Even with very high costs in developing such programs as Lotus 1-2-3 or WordPerfect, the
number of purchasers for these programs were unprecedented and increasing returns must surely
apply to these products. Moreover, this is not an industry where the costs of failure can be passed
on to consumers by increasing the price of some other product, instead they are borne by investors.
II. Early Developments: Origins until 1965
The development of computers during and immediately after World War II was directed
toward scientific and technical rather than business objectives.16 Like their electromechanical
business machine precursors, early computers were programmed by rewiring and thus were highly
specialized to particular information processing tasks.17 After the war, Alan Turing and John von
Neumann’s ideas for stored program computer created the possibility of a general-purpose problem
solving device that could be "programmed" to emulate and replace more specialized data
processing machines.18 Programmability and the possibility of "reusable"software gave general
purpose computers an advantage over the large installed base of punch-card data processing
equipment.19 Maurice Wilkes, Director of Cambridge University’s Mathematical Library
expressed prescient views on the economic importance of reusable software: "There would be
almost as much capital sunk in the library of sub-routines as the machine itself and builders of new
machines in the future might wish to make use of the same order code as an existing machine in
order that the subroutines could be taken over without modification."20 In short, the problems of
software "lock in"and the incentive for creating machines that could emulate the operation of
earlier machines were present at the start of the industry.21
16 Histories of the origins of the computer industry include Flamm  on the government
role in the creation of the computer, Bashe et al  on IBM’s early computers, and Williams
 on the technological history of computing devices (including precursors to the electronic
17 See Austrian .
18 The Univac I, Remington Rand’s commercialization of a computer designed by Eckert and
Mauchly, and the IBM 701 launched the first generation of general purpose commercial computers.
19 In 1928, the five largest manufacturers of data processing equipment based on punched
cards recorded profits of $32 million on sales of $180 million and the same companies recorded
lower profits ($19 million) on an equal sales volume in 1937. The annual prewar revenues in this
market were larger than the value of the installed base of general purpose computers in 1956, the
era of the first generation (vacuum tube) computer. (Beniger ).
20 As quoted in Bashe et al , p. 322.
21 For an exploration of the role of lock-in as an influence on user choice see Greenstein
. An explanation of how "lock in" was partially overcome through the growth of "niche"
markets is developed in Bresnahan and Greenstein .
An important early demonstration that computers could be used for purposes other than
scientific computation was the development of the SAGE air defense system whose software
requirements led to the founding of the System Development Corporation in 1956. The early
commercial use of computers in the 1950s however, also gradually stimulated a market for
software services. Producers of computer systems such as IBM provided programming services
and software tools.22 Providing these complements accelerated adoption of new general-purpose
computers, reinforced links between computer producers and users, and laid a foundation for the
reuse of software in future machines. If software instructions could be made less machine-specific,
the costs of adopting new machines could be reduced.23 Computer system manufacturers
accordingly focussed on producing the tools for creating applications programs, rather than
developing application programs themselves.24 Corporations using computers thus needed to
develop software for their own information processing applications. As noted above, computer
producers have an incentive to stimulate the production of any and all software that will increase
the value of computers and enhance their sales of computers. Accordingly, IBM also supported
the formation of users groups such as SHARE, which, as the name implies, was devoted to the
exchange of software routines.25 Computer system producers that offered services and software to
stimulate the use of computers, users that developed applications for their own use, and users that
cooperated in the exchange of programming routines and methods formed the early economic
organization for software development activities. The structure of this organization heavily favored
the "make" rather than the "buy" choice in the acquisition of software.
Larger companies scaled up their in-house software development to utilize faster processing
capabilities and substantial improvements in peripheral devices such as tape drives, printers, and
22 For example, IBM supplied interpreters, programs that ease the problem of machine
language programming by translating "assembly code" instructions into machine readable computer
23 Of course, this process also increases the substitutability among all types of computers.
Programs such as operating systems (many of which are machine specific) therefore may play a
special role in computer system manufacturers’ strategies.
24 IBM and its competitors provided significant assistance in software through the sales
function. In IBM’s approach, an application with significant value was identified and IBM and
company engineers worked cooperatively to implement the software, often in cooperation with
25 Bashe et al , pp. 347-349. While SHARE fell short of early expectations, it became
important with the introduction of IBM’s 704 in 1955 and accumulated a library of 300 programs.
disk drives. In-house development was facilitated by growing use of higher level languages. One
of the first and most successful of these, FORTRAN (FORmula TRANslator), was introduced in
1957 and a 1958 survey of twenty-six SHARE user installations found that over 50% of these sites
employed it for the majority of their problems, a rate of use that soon accelerated with further
improvements in available compilers.26 The substantial improvements in programmer productivity
made possible by FORTRAN lessened the severity of bottlenecks in "in house" development
efforts, extended the range of feasible applications, and freed users to consider new machines with
compatible language features. Higher-level languages reduced the costs of "in house"
development, further tilting the "make or buy" decisions toward the "make" outcome because they
reduced the requirement for in-house creation of highly specialized (machine specific) "machine
language" programs freeing programmer resources that would otherwise have slowed
developments and forced many companies to employ external programming services.
While FORTRAN was used for wide range of applications, the demand for a high level
language for accounting and other business applications was keenly felt by the user community.27
These developments, as well as the sponsorship by the Defense Department of a committee to
develop a "common business language," led to the specification of a new language COBOL
(COmmon Business Oriented Language) in 196028 and two years later IBM offered COBOL for
several of its 1401 models, including one of the smallest such systems.29 Growth in COBOL
usage was even more rapid than had been true of FORTRAN. IBM estimated that while
FORTRAN use peaked in 1965, COBOL continued to grow rapidly growth through 1975,
accounting for about 50% of software usage and displacing FORTRAN and use of assembler
languages beginning in 1965.30 The development of higher-level language support for IBM
26 ibid., p. 357.
27 ibid., p. 364.
28 Flamm . See also Chapter 3 in this volume.
29 Bashe et al , p. 366.
30 OECD , p. 31. Assembly languages are machine-specific low-level languages that
are closely related to the actual instructions executed by a computer, or "machine language."
Higher level languages are "compiled" into machine language while assembly language programs
are "assembled," a term that denotes the close translation between assembly and machine language.
Although assembly language programs are typically more compact than the "compiled" version of
a program can perform the same functions in less time, assembly language is also typically more
time-consuming to write and debug than higher level languages such as FORTRAN or COBOL.
computers was an important factor in delaying the growth of an external market for computer
software. Despite rapid growth in applications demand and centralized computer facilities, these
higher-level languages supported productivity gains in software development that blunted the
demand for external programming services of large organizations with in-house software
In 1960, IBM introduced the IBM 1401, a less expensive general purpose machine
addressing the needs of the medium size user. This machine was sold with a new high-level
software language RPG, whose operations resembled those of punched card systems, and thus
could be employed by individuals without costly retraining in the more abstract FORTRAN and
By 1965, IBM as well as its competitors including Burroughs and Control Data, had
stimulated a market for programming services, software products, and professional services of $500
million in annual revenues. In 1965, the total value of shipments of U.S. computer manufacturers
were estimated to be $2.4-$2.8 billion.32 Much of the $500 million in revenues went to "service
bureaus," companies that specialized in developing applications software such as payroll systems
and selling information processing services to other, usually small, companies. By contracting
externally for information processing services, client companies avoided investments in both
computer hardware and software, albeit at the cost of having to redefine their information
processing requirements to fit the solutions offered by service bureaus. One of the leading service
bureau companies, Automatic Data Processing (ADP) was established in 1949 and by 1964 had
revenues of $4.7 million, which grew to $20 million by 1968.33 The McDonnell Automation
Center, formed in 1960 and merged with McDonnell Douglas’ California computer operations
(following the 1970 merger of McDonnell and Douglas aircraft), recorded $47 million in revenue
in 1970.34 These, and other, computer service bureaus were competitive alternatives to the
purchase of computer hardware and the in-house development or purchase of software. While
many of these companies had strong sales growth in the latter half of the 1960s, the period was
31 Bashe et al , p. 480.
32 Estimates of the value of domestic software and services market from CBEMA  as
excerpted in Juliussen and Juliussen .
33 Fisher, McKie, and Mancke , p. 321.
34 ibid., p. 320.
also marked by the rapid diffusion of smaller-scale computers such as IBM’s 1401 that offered
medium and smaller-sized organizations their own computer facilities while almost all larger
business organizations had installed computers by the end of the decade.35
Early commercial applications of computers were associated with in-house programming
using higher-level languages and the growth of service bureaus as an alternative supplier of
computing services. This structure for the supply of software, in which computer manufacturers
created "tools" for applications development, users developed application software, and a residual
of users employed service bureaus for their data processing needs was short-lived. Developments
that occurred from 1965-1970, including IBM’s success with the System/360 and IBM’s decision
to unbundle software from its supply of computers,36 increased the market for multi-installation
software sales. The new entrants that formed the base of the independent software vendor (ISV)
industry included software tool and utility program suppliers as well as "vertical market"software
companies that provided applications for particular industries and for common software needs such
as accounting systems. For these reasons, 1965-1970 were the pivotal years in the emergence of
the current structure of the U.S. software industry.
35 Phister .
36 IBM would also spin-off its service bureau operations to Control Data, DeLamarter ,
pp. 87, 94.
The number of independent software vendors grew rapidly during the late 1960s.
Lawrence Welke, president of International Computer Programs, Inc. (ICP) testified in U.S. vs.
IBM that, by 1965, 40 to 50 major independent suppliers of software and programming services
and several hundred smaller organizations had been established.37 Welke stated that these
companies were deriving most of their revenues from work with the U.S. government and from
systems development work on behalf of computer manufacturers.38 The largest of these
companies, Computer Sciences Corporation (CSC formed in 1959), had revenues of $5.7 million in
1964, $17.8 million in 1965, and $82 million in 1970. Much of CSC’s growth in the late 1960s
reflected the firm’s development of multi-client software packages for accounting, ticketing,
income tax preparation, commercial loans, and system operations. Similarly, Informatics, Inc., an
early database producer began offering its product, Mark IV, in competition with IBM’s unpriced
"generalized retrieval system" in 1967.39 By 1969, Informatics, Inc., revenues were $19.8 million
from sales of computer service and software and there were over 170 installations with the Mark
IV program.40 Welke estimated that the 1965 population of 40-50 major and several hundred
smaller companies had by 1969 grown to more than 2,800 such organizations. This enormous
growth was focussed on contract programming services, which accounted for $600 million of these
companies’ revenues in 1969 which dwarfed their sales of $20-25 million in software products.41
Although many companies were founded prior to this period, 1965-1970 marks the
beginning of the U.S. independent software industry. Before this period, software companies were
small and relied on government contracts and system development work for hardware companies.
In contrast with the service bureaus that were developing during the period, software companies
had little direct contact with users other than the federal government and therefore had a difficult
time marketing their services. Welke estimated that user expenditures on software creation
skyrocketed from $200 million in 1960 to $3-4 billion in 1965 and continued upward to $8 billion
37 Fisher, McKie, and Mancke , p.322.
39 ibid, p. 326.
41 ibid., p. 323.
in 1970 and $12 billion in 1975.42 Thus, with the contract programming and software sales of
$625 million in 1969 noted above, users acquired less than 10% of their needs externally in the
Three events made it possible for independent software vendors to improve their position
in the 1970s. The first was the 1964 introduction of the IBM System/360 "family." The
System/360 unified the operating system software of the IBM product line, allowing users to
develop software applications that could operate on systems ranging in monthly rental price from
$8,800 to $60,000, with corresponding increases in computational power. The unification of
software within the System/360 product line encouraged users to define a "migration path" from
smaller to larger mainframe computers that would maintain the value of their software
development effort as their use of software and computers intensified. These same opportunities
supported the growth of computer service companies whose regional processing operations could
be expanded through migration to System/360’s with greater computational capacity as their
business expanded. But the System/360 also provided the first instance of a broad "installed base"
of computers with a single operating system. Independent software vendors had, for the first time
in the industry’s history, the opportunity to market the same product to a variety of users. By the
late 1960s, the independent software industry still accounted for a small share of U.S. software
creation activities despite rapid growth during the previous decade.
The second important event supporting the growth of the independent software sector was
IBM’s decision to unbundle the sale of hardware and software during 1969. IBM, which by
various estimates accounted for between two thirds and three quarters of computer sales and
installations, had previously offered software tools for creating user applications as part of the
computer systems it leased and sold. Since IBM systems accounted for 66-75% of mainframe
sales and leases during this period, its software "bundling" policy was highly influential. The
practice was terminated on June 23, 1969 for new orders and January 1, 1970 for existing
installations. The shock from this change was cushioned by IBM’s announcement that it would
continue to provide system software and previously produced applications and development tool
software to both new and established users.
The motives for IBM’s unbundling decision are disputed. One interpretation is that IBM’s
actions were made in response to anticipated litigation.43 Fisher, McKie, and Mancke 
dispute this view, noting that no direct evidence of relation between the announcement and the
DOJ antitrust action was discovered during subsequent litigation.44 These analysts instead argue
that IBM’s costs of software support were increasing rapidly. The costs of developing the
System/360 operating system software had proved a major trauma for the company, and seemed to
foreshadow still further cost increases.45 In addition, the growth of independent software vendors
made it possible by the late 1960s for IBM to consider separate pricing for software and to retreat
from its commitment to provide all of the software tools that users might need in order to purchase
or lease IBM computers.
The third important event of the late 1960s for the software industry was the development
of the minicomputer industry. Although IBM dominated sales of mainframe computers that
occupied a central position in the data processing activities of their customers, centralization of
computer operations had disadvantages. The inflexibility and inconvenience of mainframes
imposed costs on users that wished to develop new applications based on real time computer
control, research and development problem solving, and other specialized problems. For these
users, the mainframe computer was a barrier to gaining exclusive and intensive access to
computing time, achieving rapid turnaround in the development process, and linking devices such
as scientific instruments under computer control. These problems persisted despite the frequent
existence of excess computational capacity within mainframe computer installations.
Digital Equipment Corporation (DEC’s) pioneered the minicomputer market with its
introduction of the PDP-8 in 1965. The PDP-8 could be rented for $525/month, 6% of the cost of
IBM’s smallest System/360, the Model 30. The PDP-8 performed commercial type computational
43 On December 6, 1968, IBM announced that significant changes in IBM marketing practices
would be made in the following year. On December 11, 1968, Control Data Corporation (CDC)
sued IBM for antitrust violations following two years of data gathering on IBM’s practices that
CDC shared with the Department of Justice. U.S. vs. IBM was filed a month later with claims
similar to those offered by CDC. Brock , p. 170. The protracted resolution of U.S. vs. IBM
certainly prevented IBM from reconsidering its unbundling decision during the 1970s.
44 Fisher, McKie, and Mancke , pp. 176-77.
45 See Pugh, Johnson, and Palmer , pp. 331-345 who employ the term "trauma" to
describe this experience. For a poignant summary of System/360 system software development see
Brooks , especially pp. 47-48.
tasks at about 6% and scientific computation at 22% the speed of the Model 30.46 By
compromising speed in order to achieve a very low cost, the PDP-8 tapped user needs that were
not well served by the competing technological solution, time sharing, which had floundered in the
late 1960s.47 At one seventh the cost of a mainframe, users could afford to let the PDP-8 stand
idle, dedicate it to a single task, or share it among a small team engaged in a development
effort.48 The minicomputer also extended the application of "real time" control, i.e. applications
in which a computer directly controls scientific instruments or electro-mechanical systems.49
Moreover, minicomputers could be used in dedicated data processing applications such as data
collection and entry.
The ability to develop entirely new computer system architectures in which tasks were
"distributed" provided a basis for networked computing, a means for combining computers of
different sizes and computational capabilities and optimizing the system for both computational
46 Both comparisons are drawn from Phister , pt. 2, pp. 343-44 and pp. 350-351. The
speed comparison is based on the machines’ relative Knight index, a measure based on the speed
of executing instructions in different types of applications where the frequency of particular
instructions is used as a weight on their execution speed.
47 Experiments with time sharing had begun when John McCarthy and his colleagues sought a
means to address their own software development needs at MIT during 1960 and 1961, (Pugh,
Johnson, and Palmer , pp. 355-356.) IBM launched several experimental timesharing system
software development efforts including the System/360 Time Sharing System (TSS) in 1967.
Despite these efforts, the performance goals demanded of time sharing were not met. An history
of IBM concludes, "Having been victimized by over-optimism, time sharing temporarily floundered
during the last half of the 1960s." [Pugh, Johnson, and Palmer , p. 364.] While part of
IBM’s problems with developing time sharing systems stemmed from failure to embrace computer
architectures optimized to this application and the shortcomings of its early time sharing operating
systems, companies such as General Electric that had overcome both hurdles fared little better.
General Electric’s 265 system was adopted by MIT and Bell Laboratories and was responsible for
GE’s leading position in the time sharing services market of the late 1960s. Nonetheless, this
system (as well as IBM’s) experienced accelerating decline in performance as users were added
and, by several measures, was more expensive than batch computing on mainframe computers.
(see Sharpe , pp. 509-517.)
48 Two contemporary viewpoints on decentralization were Tomaszewksi , who
advocated mobile "gypsy" development teams for centralized computers, and Wagner , who
argued that dedicated use of cheaper computers makes sense even with enormous unused capacity.
Wagner’s hardware-based solution prevailed, largely because of the reductions in hardware costs
made possible by the minicomputer.
49 Such "real time" systems had previously been developed for cost insensitive applications
such as the control of military and space missions. The minicomputer allowed similar techniques
to be employed in commercial applications.
performance and response time. First developed in the late 1960s, distributed computing had its
greatest impact only during the 1970s. Minicomputers also made it possible for small
organizations to begin to purchase and use their own computers. In a 1968 survey, U.S.
manufacturing companies with fewer than 500 employees or less than $10 million in sales reported
no use of larger computers.50 In the same year about 30% of companies with 200-499 employees
or $5-10 million in sales were using smaller computers, including the smaller System/360
mainframes and comparable machines including minicomputers. Applications and affordability
were responsible for rapid growth in the minicomputer sector during the late 1960s, the period
when time sharing operating systems for mainframes were floundering. By 1970, as the CBEMA
estimates of Table 1 indicate, minicomputer unit shipments exceeded those of mainframes and,
consistent with their price, were achieving about one seventh of mainframe revenues.51
To summarize the remarkable developments affecting the software industry of the late
1960s--IBM decided to unbundle hardware and software as its System/360 moved to a dominant
position in the market for mainframe general purpose computers, independent software suppliers
began to carve out a market by competing in quality with IBM software products, programming
services and service bureaus were growing rapidly as they improved their self-developed software
and benefitted from price per unit performance reductions in computer equipment, and the advent
of the minicomputer created an enormous range of affordable commercial applications for
computer systems. Although disentangling the relative contributions of each of these developments
appears to be an impossible task, their collective impact, according to Welke, was to raise user
software creation investments from the $3-4 billion level in 1965 to $8 billion in 1970. In the
same five years, CBEMA estimates that software product and services revenue increased more than
ten-fold from $200 million in 1965 to $2.5 billion in 1970, (see Table 2).
50 Phister , pt. 2, p. 453.
51 Growth in U.S. shipments in minicomputers depends upon classification. Table 1 includes
estimates of both Phister and the Computer and Business Equipment Manufacturers Association
(CBEMA). Phister’s inclusion of smaller and more specialized machines raises his estimate and
pushes the date of first shipments of minicomputers back in time to 1957.
IV. Growth and its Discontents--The 1970s
By 1970, annual sales of computers and peripheral equipment (SIC 7573) were
approaching $5 billion or nearly 4.75% of domestic business investment in equipment and
structures.52 Software product and service revenue were $2.5 billion in 1970, 50% of hardware
revenues in that year. The entry of numerous software companies during the late 1960s had
created an industry structure that was so fragmented that estimates of the number of firms vary
from 1,500 to 2,800. Of the $2.5 billion in 1970 software product and service revenues, these
firms may have divided less than $1 billion in revenue in a market that was still dominated by
IBM and other system producers such as Burroughs and Control Data.53 This implies that
average firm size would have been $350,000 to $700,000, a revenue level that could support no
more than about a dozen employees. In the early 1970s independent software development and
sales was a "handicraft" industry" in which an extensive division of labor or specialization was
impossible. Moreover, users faced enormous problems in sorting out the offerings and capabilities
of these companies. This fragmentation of software and service suppliers helped sustain the
growth of larger computer service companies such as Computer Sciences Corporation, McDonnell
Douglass, ADP, and EDS.54
Early in the 1970s, improvements in hardware technology allowed faster "real time"access,
an impetus for software industry growth. IBM introduced the System/370 in 1971 with new hard
disk technology (the 3330 disk storage unit, also introduced in 1971) that made it possible for user
online disk storage to exceed online tape storage for the first time in the history of the industry.55
This development dramatically improved response time of time sharing systems, significantly
increasing the performance of a system architecture that had been a great disappointment in the late
52 To be precise it was 4.738%. The 1970 output of the computer and peripheral industry
(SIC 7573) was $4,984 million while gross private domestic nonresidential investment $105.2
53 Datamation estimated IBM’s 1975 software revenues at 10% of its $11.1 billion revenues,
54 During the 1970s, EDS began to provide on-site data processing services for large clients.
55 Pugh, Johnson, and Palmer , p. 532.
1960s and allowing the growth of time sharing service companies as a new source of software
As IBM’s System/370 extended the capabilities of centralized data processing, Digital
Equipment Corporation (DEC) and its competitors in minicomputers developed the market for
decentralized data processing. Minicomputer and mainframe shipments had been approximately
equal in number (but not in revenues) in 1970. By 1976, almost six minicomputers were sold for
every mainframe and by 1980 more than ten minicomputers were sold for each mainframe (see
Table 1). Each of these minicomputers required software, often of a very specialized nature
requiring extensive user development efforts. For example, the oil refinery business was a major
source of minicomputer use, sparking rapid expansion in the demand for complementary
Use of minicomputers as primary computers in smaller organizations, their use as "front
ends" for mainframes, use of minicomputers in data communications systems, and their embedding
in process control systems each required very different software. Although total demand for
minicomputer software was very large, the diversity of minicomputer applications meant that the
size of individual markets for "packaged" minicomputer software was limited. Small market size
limited the economies of scale from software packages and led to the development of a
minicomputer software industry structure that closely resembled that of the mainframe software
industry, in contrast to the "mass market software publishing" that emerged with the personal
computer in the 1980s.
The fragmentation of the minicomputer industry is indicated in Table 3 which reports on
the results of Datamation’s 1977 user survey of satisfaction with packaged software.58 In the
1977 survey, 199 packages were rated by 5 or more users, the survey report’s threshold of use for
reporting the name of the software package. Mainframe package software applications dominated
56 The enhanced computational capabilities of the System/370 were complemented by it use of
integrated circuit memory, which proved to be faster and ultimately cheaper for storage of data and
software instructions than ferrite core memory. ( See Pugh  for a detailed history of ferrite
core memory, the critical technology for computer memory until the 1970s.)
57 Petrochemicals was the leading sector in demand for minicomputers; installations grew from
300 in 1964 to 2,000 in 1974. See Phister , pt. 1, p. 134.
58 . In all, 5,813 responses reporting on specific packages were received from Datamation
these "multi-user"packages, accounting for 72% of applications for which the computer system
could be identified.59 By comparison, non-IBM minicomputer package software, most of which
was system software, accounted for only 10 packages in the survey, the same number of software
packages that were listed for IBM’s System/3 minicomputer. Collectively minicomputer software
packages accounted for only 20% of the products mentioned by five or more respondents to the
survey.60 Thus, despite the far larger "installed base" of minicomputers, the relatively small
number of packaged software products for minicomputers indicates continued fragmentation in the
minicomputer software market as of 1977, despite rapid growth in the number of machines and
users during the previous decade.
For the software industry as a whole, the 1970s were a period of broad-based growth
accompanied by growing concerns about productivity in software development. Growth of
software related activities was led by a ninefold expansion in data processing service revenues; but
sales of software products and professional services also expanded by factors of 5.5 and 7
respectively during the decade (see Table 2). For mainframe computers, the 1970s were a period
of intensive rather than extensive growth. By 1970, growth in the number of firms with
mainframe computer systems, computers per dollar of revenue, and computers per employee had
all decelerated from the high growth rates of the 1960s.61 The organizations that had adopted
mainframes by 1970 were developing application programs that would more intensively use their
existing mainframe computers rather than increasing the number of mainframes used in their
organizations. Growth in the use of minicomputers was both intensive (within firms) and extensive
(in new user firms) during this period.
In both mainframe and minicomputer applications the problems of software development
and maintenance received growing attention during the 1970s. Users were beginning to experience
significant problems in managing their in-house programming efforts. The problems noted in the
technical and management literature of the time include efforts by programmers to shelter their
59 Of the 192 packages that could be linked to a particular type of computer system, 138 were
mainframe software packages.
60 Unfortunately, Gepner  did not report user characteristics. He does report that 30,000
surveys were sent out for the 5,813 returns, a response percentage of almost 20%. Unless the
mailing list was biased this is a large enough sample to capture a significant share of the
minicomputer users of the period.
61 Phister , pt. 1., p. 129.
positions by creating software with high maintenance costs,62 the intractability of deriving
meaningful measures of software and system performance metrics for increasingly complex
systems,63 the hurdles of properly specifying large software systems,64 the frustrations of
organizational politics in managing such efforts,65 and the growing disappointment with software
Problems with software did not prevent the 1970s from being a period of enormous growth
in computers applications and perhaps contributed to employment growth in data processing. As
early as 1972, computer equipment accounted for almost 23% of producer durables sold in the
U.S.67 Concerns about software noted above reflect continued growth in this investment, whose
share of producer durables increased more than 33% by 1982.68 Occupations related to data
processing grew relative to both non-computer service and manufacturing sector employment
62 For example, one software entrepreneur noted: "It is not at all uncommon for a
programmer to threaten resignation, while simultaneously generating the type of undocumented
programs that increases management’s dependence on him. Thus he is in a position of strength
from which he can (and, in the aggregate, often does) use mild blackmail to achieve greater status,
money or dominance over management." (Brandon ) See Kraft  for the alternate
view that programmer’s work was becoming "deskilled"and routinized.
63 See Kolence  and Boehm , .
64 Larsen 
65 Keen and Gerson .
66 The quality shortcomings of software creation activities were a frequent subject in the trade
press and in conferences of the 1970s. For at the example at the 1974 National Computer
Conference the President of the American Federation of Information Processing Societies "accused
the data processing industry of ignoring the growing inability of the programming profession to
produce enough good software." (See Dolotta et al  for the quoted paraphrase of G.
Glaser’s remarks and the cite to Glaser , the published version of the speech.) One example
from Datamation in 1974 of many in the trade press, quotes Tom Steel of Equitable Life
Assurance: "Quality is, of course, a complex attribute, but by whatever measure one chooses,
current software scores low. It is usually inadequate functionally, inconsistent between actuality
and documentation, error-ridden and inexcusably inefficient. Beyond all that, it costs far too much.
I can think of no other products (aside, perhaps, from pornography and telephone service in New
York) that have all these failing to anything like the degree found in software." (Kirkley , p.
67 CBEMA , p. 14.
68 ibid. It continued to grow until well into the 1980s, eventually leveling off at about one
throughout the 1970s.69 New methods for organizing the growing investment in data processing
equipment lagged behind the investment itself.
As the costs of supporting this new form of physical capital became more apparent, new
structures for decision-making and oversight were needed. The symptoms of efforts to come to
grip with these issues are the problems were increased attention to software and system
performance, software specification, organization of software creation, and quality assurance. But
because software was becoming the repository and mediator of information flows, productivity
problems competed for managerial attention with the problems of extending access and utilizing a
rapidly expanding flow of information.
69 Baumol, Blackman, and Wolff , Ch. 7.
V. The Revolution: Software in the 1980s
The 1980s were an extraordinarily complex period of evolution and growth for data
processing. Before beginning a discussion of major new actors and activities, two major
developments that distinguish the 1980s from the previous decade merit attention, the death of time
sharing as it was known in the 1970s, and the retreat of computer manufacturers except IBM from
software and service activities.
The process of creative destruction during this period razed an entire sector of the data
processing industry during the decade, the time-shared service company. Table 4 and 5 report on
the largest data processing companies in 1980 and 1991 that derived the majority of their revenues
from software and services. In 1980, there were eight major (as well as dozens of smaller regional
companies) timesharing services companies including Dunn and Bradstreet which is primarily
identified with business services and software . By 1991, seven of these companies had fallen out
of the Datamation 100, while the eighth (Dunn and Bradstreet) retained some online information
services. Information service companies such as Prodigy (a joint venture among Sears, IBM, and
other companies), Compuserve, and America Online have filled part of the vacuum left by the
timesharing service companies, and the sale of processing time on remote computers is still an
active but highly fragmented market. But remote computation services are now more closely tied
to companies that offer system integration or vertical market (specialized by industry) consulting
services. Thus, software creation activities for particular industry segments that might have been
provided by timesharing service firms are now provided by more specialized service providers.
These services may still be organized as a remote computing activity (e.g. ADP now receives much
of its data from users electronically rather than on paper forms), but the 1970s-style time sharing
company that offered a collection of applications within a single timesharing system has virtually
During the 1980s the software activities of computer manufacturers also underwent
dramatic change, as the importance of software and service activities declined significantly for
computer manufacturers other than IBM. Table 6 shows software and services revenues for the
major computer manufacturers for 1981 and 1991, excluding the IBM plug-compatible companies
(e.g. Amdahl) that have never been prominent in software production. Computer manufacturers’
total revenues from software almost doubled over the decade, but almost all of this increase was
attributable to IBM; software’s contribution to its total revenue grew from 17 to 20%. While DEC
expanded its software revenues as well, by 1991, software accounted only one-third of DEC’s
software and service revenue while IBM derived 84% of its 1991 software and services revenue
U.S. computer manufacturers, with the exception of IBM, withdrew from software
development during the 1980s through several different paths. For Unisys (formerly Sperry and
Burroughs), NCR (now a division of AT&T), and Hewlett-Packard, 1991 software and service
revenues fell below their 1981 levels and plummeted as a fraction of the firms’ total computer-
related revenues. Other firms, especially new entrants into workstations, minicomputers, and
personal computers (such as Sun, Tandem, Prime, and Apple) specialized in selling hardware rather
than software.70 These developments reflected the maturation during the 1980s of the
independent software industry. Sales of computers no longer required manufacturers to provide
software other than a basic operating system. Instead, other companies (including the in-house
development efforts of users) provided these complementary inputs.
The following sub-sections discuss three major developments of the 1980s that have
affected the software industry. The emergence of the personal computer at the beginning of the
decade provided a new and revolutionary organizing principle, mass publication of packaged
software. The introduction of the workstation in the middle of the decade fueled the creation of a
new, large class of software applications that exploited the workstation’s sophisticated graphics and
numerical computation capabilities. The productivity and organizational problems that first
appeared in the 1970s supported the rapid growth during the 1980s of system integration
companies and "outsourcing" of corporate data processing activities and management.
70 An important qualification is that Apple Computer, Inc.’s operating system software for its
Macintosh computers was bundled with the sale of the hardware and thus is not recorded as
The Personal Computer Revolution
A succession of products introduced during 1975-81 was overshadowed by IBM’s
introduction of its PC in August, 1981.71 IBM succeeded in learning from all of the experiments
that had been undertaken in the first six years of the industry and introduced a machine that
combined a reasonable level of computational power and an operating system that facilitated
application development. Customers quickly endorsed the new product, purchasing over 13,000 of
the machines in its first year.72 While Apple and Tandy machines continued to outsell IBM for
several years, IBM had attained 26% of the personal computer market by 1983. The market for
personal computers grew very rapidly during the 1980s. Table 7 reports the IDC’s data on U.S.-
based computer industry shipments by value and volume for the decade. By 1984, personal
computer sales accounted for more revenue than IDC’s large, medium, or small scale computer
market segments with shipments of 9.7 million personal computers for a total revenue of some
$17 billion, bringing the installed base in 1984 to 23 million machines.
The rapid growth in the installed base of personal computers provided an homogenous
market for operating systems and applications of unprecedented size. In 1984, the installed base of
both large and medium size computers was less than 200,000 units and about 1.9 million small-
scale systems were in use; there were 23 million personal computers in use in that year. The
enormous size of the personal computer market created unprecedented scale and profit
opportunities for software producers that were simply unavailable in the markets for larger
computers, even though software in the latter markets often had higher price tags.
Three new software companies emerged in the early years of rapid growth in the personal
computer market. As of 1985, Lotus with annual revenues of $226 million had become the 60th
largest U.S. data processing company.73 Microsoft at $163 million ranked 78th and Ashton-Tate
at $110 million ranked 100th among U.S. data processing companies in 1985.74 These companies
were joined in 1988 by WordPerfect Corporation, whose sales in that year were $179 million.
71 Langlois  provides a concise history of the early development of personal computers
prior to and including IBM’s introduction of the PC.
72 ibid., p. 23.
73 Datamation , p. 62.
Each of these companies revenues was dominated in 1986 by a single product that had penetrated
a large share of the installed base of personal computers and dominated a class of software--
spreadsheets for Lotus, operating systems for Microsoft, databases for Ashton-Tate, and word
processing for WordPerfect.
While the arithmetic of corporate size in the personal computer industry is straightforward,
it is less immediately obvious why a single product in any application class should garner a
dominant share of personal computer installations. In practice, however, several network
externalities can cause a single software product to become a de facto standard for an
application.75 First, there are advantages in establishing a single operating system standard.
IBM’s endorsement of the PC-DOS operating system provided Microsoft with an enormous
advantage that it quickly converted into a position as the dominant supplier of operating systems
for IBM and IBM-compatible personal computers.76 Second, it is desirable for a user firm to
choose a single application product within a particular class so that information created by one user
may be shared by another. Lotus Development’s spreadsheet application, Lotus 1-2-3, was chosen
by many business users of IBM personal computers (for which it had been carefully optimized as
Chapter 6 notes) and rapidly displaced the VisiCalc spreadsheet program. Third, the accumulation
of skills, training materials, and add-on products that facilitate the use of the product may reinforce
the dominant position of a software product. The development of numerous specific applications
written for the dBase II and, later, dBase III and IV products of Ashton-Tate, as well as training
materials and the accumulation of consultant skills in these products, are examples of such
externalities. Fourth, widespread use of a specific software product creates externalities in the
labor market; more individuals are available with skills in widely-used application programs. The
widespread use of Lotus 1-2-3 in business schools and corporations made it possible to hire
individuals who required no training in the use of the product, encouraging companies to expand
their use of the product and erecting a substantial entry barrier to other products.
75 See Chapter 6 for additional discussion. Network externalities increase the value of
participation in a particular network. The telephone network is the archetypal example of network
externalities; the value of a telephone connection increases with the number of others connected to
the telephone network. A de facto standard is established by market outcomes as opposed to a de
jure standard which is established by some deliberative process.
76 IBM initially retained a proprietary advantage in the embedded software of the IBM PC that
disappeared when small software companies succeeded in duplicating its functionality, a
development that launched the PC-compatible or "clone" market.
Collectively, these effects have propelled specific applications to dominance in their class.
Between 1981 and 1985, the share of the 15 largest independent software vendors increased from
37% to 72% of total personal computer software revenues. The three largest companies, Lotus,
Microsoft, and AshtonTate, accounted for 35% of the total in 1985.77 There are, of course,
limitations to the impact of network externalities as the disappearance of Ashton-Tate through
merger with Borland and the recent acquisition by of WordPerfect by Novell suggest. In
particular, major differences in the functionality of products may fragment user choices and create
fewer positive externalities.
Personal computer word processing software illustrates the limits of these externalities in
permanently establishing market dominance. WordStar, produced by MicroPro International,
originally competed against a number of products with more readily understandable user interfaces
(such as MultiMate) or with more features (such as XyWrite). But WordStar and its competitors
were displaced by WordPerfect, which provided extensive features and an attractive user interface.
WordPerfect, in turn, has been unable to dislodge Microsoft Word as the primary word processing
application on Apple’s Macintosh and now faces major competition from the recent success of
Microsoft’s new operating system Windows, for which Microsoft Word is a commonly chosen
word processing application. As I noted earlier, the acquisition of WordPerfect in March 1994 by
Novell reflects the growing problems of competing against Word and Windows. Ashton-Tate’s
dBase products were challenged late in the decade by a new line of database products such as
Paradox, leading to Ashton-Tate’s acquisition by Borland, which had previously acquired one of
Ashton-Tate’s primary competitors. Thus, positive externalities reinforce the establishment of a
single product standard for a class of personal computer applications, but they do not guarantee the
emergence of a single standard.
The market in personal computer software resembled many aspects of publishing or mass
entertainment. The similarity of promotion and distribution methods with book and record
publishing was especially marked.78 Use of independent distributors that could provide stocks
of popular software to immediately satisfy demand for a "hit"product so that users needs could be
77 Business Week , p. 129.
78 The Software Publishers’ Association, established in the early 1980s, immediately began to
publish information about the unit shipments of particular products and award "gold" and
"platinum" status to bestselling products. (See Software Publishers Association , , and
 for recent examples.)
immediately satisfied became an important competitive weapon. The growth of personal computer
magazines, with extensive advertising for software and hardware, further strengthened the
similarities between this new software market and the recording or publishing industries. The
retail distribution channels that had been established to sell personal computers also supplied their
software while mail order suppliers or direct distribution provided a convenient source for more
specialized products and served as discount outlets for the major products. Major software
companies supplemented these promotion and distribution methods with direct sales efforts
characteristic of earlier packaged software.
The economic advantages for a firm from creating and maintaining a leading software
product were enormous. Most of the costs in the packaged software market were the fixed costs of
product development. When amortized over millions of users, very large development efforts
could be financed from current revenues. During the 1980s, the largest personal computer
software companies invested 10-11% of their revenues in R&D.79 Their development efforts led
to ever more sophisticated products that could utilize increasing personal computer performance.80
High development costs have introduced elements of monopolistic competition into the personal
computer software industry, as the costs of software development have begun reduce the high gross
margins of producers.81
The huge size of the market for personal computers also created dynamic markets for
complementary hardware products which in turn created new opportunities for applications
markets. Markets for hard disks, display monitors, modems, and printers all grew very rapidly
during the 1980s. The storage capacity of hard disks enabled software producers to deliver larger
programs, and created a demand for utility software to maintain the larger collections of files
stored on personal computer systems. Similarly, display monitors encouraged the creation of more
79 Hodges and Melewski .
80 A major transition in the personal computer industry occurred as Compaq and IBM
introduced new MS-DOS based computers based on the Intel iAPX 386 microprocessor. The
iAPX386, also called the 80386, employed a memory addressing method that made it possible to
develop software programs far larger than those that could be created for previous models of IBM
and IBM-compatible computers. In combination with the greater availability of high capacity hard
disk drives, the iAPX386 and its successors made it possible to devise much larger and more
complex software products for the personal computer including Microsoft’s Windows.
81 Nonetheless, the gross margins of Lotus Development and Microsoft were 81 and 74%
respectively as of 1989. These are impressive margins, even with the higher product development
(R&D) costs of 14-15% experienced in the late 1980s. See Business Week .
sophisticated graphic display programs and communication modems increased the demand for
personal computer communications programs. Perhaps the most important development came from
the improvement of printer technology. Canon’s laser printing engine, used by Apple and Hewlett-
Packard to produce laser printers, provided a raw capability to place images on a page at the
resolution of 300 dots per inch. Many of the software programs of the late 1980s were devised to
take advantage of this capability and an entirely new segment of personal computer, desktop
publishing, emerged as a means to take advantage of the high resolution printing in the creation of
announcements, newsletters, and other "print shop" quality documents, all of which could be
produced by a single personal computer.
By the end of the 1980s, personal computer users were able to choose from thousands of
programs for specialized applications and dozens of major software products for more general
applications. The resulting software network is certainly as complex as the network of software
applications that has been developed for all other types of computer systems. Although personal
computers have been used as terminals for mainframe and minicomputer systems, limitations in
transfer of information across systems have, until recently, limited the use of the personal computer
as a node in the larger network of computational resources available in modern businesses. Recent
efforts to "re-integrate" the personal computer into a corporate computer network that itself has
become more "distributed" between mainframe and smaller computers is considered in the
conclusion of this chapter.
The Workstation: Continuity and Change
The workstation, introduced by Apollo in 1981 and Sun Microsystems in 1982, was a
hybrid between the personal computer and minicomputer. Like the personal computer, the
workstation took advantage of dramatic reductions in the price per unit of performance of
microprocessor integrated circuits. Like the minicomputer, the architecture and peripherals
attached to the workstation’s central processing unit provided high performance in computation and
display. IBM compromised the computational capabilities of its PC to penetrate the mass market
for desktop computers; by contrast workstation producers entering the market at roughly the same
time sought to attract engineers and other technically sophisticated users who would otherwise use
minicomputer or mainframe computers.82 A major appeal of the workstation was its graphics
capability. Graphics intensive applications involving computer-aided design (CAD) and computer-
aided engineering (CAE) had been developed for minicomputers but suffered from limitations in
graphic display capability and the fragmentation in the minicomputer software market.
The most successful of the workstation companies, Sun Microsystems, adopted a corporate
strategy based on "open" standards involving the use of UNIX, a widely available operating system
that had first been developed at Bell Laboratories. Sun’s version of UNIX had been modified by
U.C. Berkeley computer science researchers with support from the Defense Advanced Research
Project Agency (DARPA).83 UNIX was already widely used on minicomputers and Sun
Microsystem’s strategy was to persuade technical users and software developers that applications
for its workstations would be "portable" to ever more powerful workstation products, imitating
IBM’s System/360 marketing strategy. Along with liberal licensing of its microprocessor
architecture (see Chapter 4), this strategy proved enormously successful in inducing investment in
software for Sun workstation products.
From 1985 to 1990 the number of suppliers listed in Sun Microsystem’s Catalyst guide to
software for Sun workstations grew from 177 to 1,325, despite a major revision in the Sun
operating system that interrupted the growth of suppliers in 1989.84 The majority of companies
82 Langlois  reports that IBM eschewed using the 8086 microprocessor for the IBM PC
to avoid creating a machine that would compete more directly with other IBM products, p.22.
83 See Flamm  and Chapter 3 in this volume.
84 From research notes by Carolyn Judy, Center for Economic Policy Research, Stanford
offered a single software product.85 Their products were, in turn, often specialized, high-
performance solutions to problems such as oil field management, molecular modeling, and
electronic circuit design. The growth in applications software spurred further sales of Sun’s
workstations in the same virtuous cycle that supported sales of IBM PC and the Apple Macintosh
personal computer products. A larger installed base stimulates software development, and the
availability of software stimulates the purchase of additional hardware platforms compatible with
that software. Although a similar cycle operated in mainframes and minicomputers, the relevant
installed base of both workstations and personal computers was vastly larger, even in specialized
markets, because of their lower price per unit of performance. The "virtuous cycle" dynamic thus
operated with greater speed and economic impact.
85 The percentage of vendors offering a single product were 68%, 67%, 69%, 66%, 61%, and
60% for each of the years 1985-1990 (Calculated from research notes provided by Carolyn Judy,
see previous note). The lower fraction in 1989 and 1990 may reflect the change in the Unix used
for Sun’s operating system, a change that was supposed to encourage software development.
The Rise of System Integration and Outsourcing
The personal computer of the 1980s did not have the performance or capacity to replace
mainframe computers and organizations continued to face the problems of maintaining and
expanding mainframe software applications. Solutions to these problems appeared to require a
sophisticated internal corporate business operation within the company whose major output was
data processing services. Many companies discovered that these internal capabilities were an
expensive drain on investment resources and, even more importantly, a distraction from the
business activities necessary to the company’s success. The continued growth in the intensity of
computer operations during 1970s and 1980s provided a growing challenge for internal
development capabilities. In many cases, internal bureaucracies ossified, or were simply unable to
keep up with the pace of technological change. Accordingly, firms began to reconsider the make
or buy decision for their entire data processing activity. If another firm could provide the
technological knowledge and human resources to implement specialized software solutions, choose
among complex competing hardware offerings, and deliver useful information services to internal
users, why not buy these services rather than produce them in-house? The growing complexity of
data processing technologies and markets pushed companies toward the "buy" solution and a
number of companies emerged to satisfy this demand.
During the 1970s as was noted in Table 4, two companies had developed large system
integration operations. These companies, Computer Sciences Corporation and EDS, originally
established as computer service firms, delivered large scale system solutions to the federal, and to
many state governments. By 1991, thirteen system integration companies had joined the ranks of
Datamation 100, the largest U.S. data processing companies (see Table 5.) Among these companies
were four of the big eight accounting firms, the 1970s leaders EDS and Computer Sciences
Corporation, and seven other companies, most of which had been involved in providing computer
programming and consulting services at a smaller scale during the 1970s. Among these
companies, EDS is noted for the extensiveness of its intervention in building data processing
departments, managing the requirements specification, subcontracting for software creation, and
eventually staffing the day to day data center operations of client companies.86 By comparison,
most of the other companies have chosen a more limited role with regard to clients, alternately
providing consulting, programming service, "change" management, or other specific services in
varying proportions over time in partnership with client company data processing personnel.
86 EDS also is active in more the market for more limited services.
The size and scope of these new outsourcing and computer service firms appears to signal
an important change in the methods of developing large corporate information systems. Unlike the
computer service bureaus or time-sharing services, software development and system design is
performed in partnership or on behalf of single corporate users. This method of delivering
programming services creates software that is not only organization-specific, but that also benefits
from the supplier’s multi-client business. The large computer service organization is able to
finance the fixed costs of large scale software development and accumulation of specific technical
competences in hardware and software that can be amortized over a large number of customers.
The computer service organization can also have major impact on the supply of published software
by negotiating multi-client site licenses and requesting specific software modifications for their
clients’ needs. Earlier computer service bureaus and timesharing services pursued economies of
scale through investment in computer hardware and the delivery of "remote" computing; the new
service organizations are achieving these economies by organizational economies of scale that
allow "direct" delivery of their services. Their success, however, is likely to attract competition
from computer manufacturers, software suppliers, and perhaps a new generation of timesharing
services using improved data telecommunications.
Summing Up the 1980s
The 1980s were a complex period in the development of the U.S. software industry. The
growth in mainframe and minicomputer applications and sales continued through the decade, but
additional layers of complexity were introduced by the widespread adoption of workstations and
personal computers. The variety and volume of hardware and software mushroomed, and so too
did problems of compatibility and complexity in organizing and managing the much larger
The common theme of 1980s developments was the creation of methods for realizing
economies of scale in the development of software. Personal computer software companies
achieved economies of scale in software development with a "publishing"approach that tapped the
immense installed base. The leading firms’ positions were further reinforced by the positive
externalities in skills and data compatibility. The large installed base of workstations and the use
of UNIX as a common operating system supported development of specialized, computationally-
intensive software. For mainframes, scale advantages were achieved in the creation of
organization-specific software through the growth of the new service organizations. In each of
these areas, software was developing characteristics of a mature industry with established actors,
large scale organizations, effective distribution methods, and a stable population of users.
Although user-produced software continued to absorb substantial resources, the viability and
stability of the independent software vendor industry seemed assured.
The developments of the 1980s have many implications for the future of software creating
activities, organizational design, and the future hardware and software markets. They also have
played an important role in the U.S. software industry’s international position. Beginning with a
brief overview of this international position, the conclusion examines the implications of the
growth of networking and the uncertain development of new techniques for software engineering,
development, and maintenance.
VI. U.S. Software in the International Market and The 1990s: Signs of Reintegration?
Network Growth and New Programming Methods
The rapid pace of change in hardware and software markets have given the U.S. software
industry an advantage in international competition where. In packaged software, U.S. independent
and system software producers holds very strong positions in domestic and foreign markets.
According to IDC, one of the leading market research companies in the industry, the 1992 the U.S.
domestic market share of U.S. independent software companies in packaged software was 58%
and the share of U.S. system companies was 30%.87 In Europe, U.S. independent software
companies held a 60% share of this market, the sum of 34% from independent suppliers and 26%
from system vendors. As Malerba and Torrisi note in Chapter 7 of this volume, software in
Europe is often delivered by hardware manufacturers and service companies and therefore it is
possible that U.S. companies hold a smaller position in the European market than the above figures
would indicate. It is true, however, that European packaged software consumption is comparable
to that of the U.S. and many of the U.S. service companies, such as Arthur Andersen, have strong
positions in the European market. In Japan, U.S. independent software companies hold a 27%
share of the packaged software market and system companies a 33% share. The overall level of
packaged software sales in Japan is, however, only one quarter that of Europe and the U.S. for the
reasons suggested by Baba, Takai, and Mizuta in Chapter 5. U.S. service companies have a
smaller position in the Japanese market.88
The most obvious explanation for the international competitive position of U.S. companies
is that they have enjoyed a first mover advantage in all of the software industry’s market segments.
European and Japanese computer production and rates of utilization have historically lagged behind
those of the U.S. providing domestic producers with smaller markets and fewer economies of scale
advantages than those available to U.S. firms. First mover advantages were generated not only be
commercial activity, but also by government R&D policy and the early development of computer
science education in U.S. universities. The importance of software for national defense systems
led to generous U.S. government support for basic and applied research in software, often through
the Defense Advanced Research Projects Agency (DARPA).89 The Association for Computing
87 IDC .
88 See Siwek and Furchgott-Roth  for further discussion.
89 See Chapter 3 for more discussion of U.S. government programs.
Machinery (ACM), a professional association, played an important role in developing curricular
standards for college-level study of computer science and aided in the rapid growth of university
courses and degree programs in software engineering.
By the end of the 1980s, U.S. companies were engaged in another major advance in
information technology, the linking of personal computers into extensive networks. Electronic
mail (e-mail), file transfer, and "work group" software applications provided one impetus for
networking efforts. Another impetus was the continuing importance of mainframe computers as
repositories of organizational databases "of record" that were updated, edited, and analyzed by
users employing personal computers. In addition to inter-user and mainframe communications,
networks provided companies with new opportunities for sharing software and computational
resources and for servicing and maintaining software within the organization. U.S. software
development in the 1990s will be influenced by these growing network capabilities, reinforcing the
U.S. software industry’s position in international markets where network developments are likely to
follow a similar path in the future. One of the largest U.S. software companies currently is Novell,
which provides the leading local area network operating system.90 Many of the currently
available database programs have introduced features that permit their use on networks (e.g. record
locking which prevents two users from simultaneously attempting to change database information.)
These developments are likely only the first attempts at deriving value from the rapid growth in
Networks provide new entry points for the introduction of externally produced software a
user organization. Instead of convincing each individual user to adopt a particular software
solution, software companies can market applications for many users at a single site or within a
single organization and thereby generate higher revenues from a single purchase decision. In many
respects, this marketing approach resembles the direct sales approach of earlier periods in the
software industry. But it also allows software producers to deal with the problems of coordinating
software needs within a firm. As noted earlier, system integration and consultant companies
currently are addressing these coordination and compatibility problems. To the extent that users
find software-embodied applications that help them coordinate computer use in the network, the
need for external services may decline. The result may be greater competition between system
90 Another example of software developed specifically for networks include Lotus
Development’s Notes, which provides a means to distribute and comment on documents over
integrator companies and larger software producers who can add consulting activities to their
software development, assisting a company in decisions on the complementary purchases necessary
to use their application or operating system software. This competition may influence the structure
of the software industry as consulting companies and software producers merge or acquire one
Assuming that security problems do not block the further extension of networking in
software and computer systems, the re-integration of personal computers with other data processing
resources within organizations and the linkage of these resources through data communication
pathways will become more important in the next decade. In principle, this process should result
in new software systems linking manufacturing companies with their suppliers and customers to
manage delivery schedules, improving the efficiency of ordering and pricing procedures, and
reducing the costs of managing the flow of material and labor inputs. Such gains will require the
development of new technical compatibility standards for software such as EDI (electronic data
interchange) and the creation of software applications that facilitate inter-organizational links. It
seems likely that these developments will ensure growth in the market for externally-developed
software solutions at the expense of internal development efforts.
A second issue likely to grow in influence during coming years is alternative techniques
for software development. The personal computer revolution has demonstrated that substantial
advantages can be gained from the use of standard application programs. These applications
programs, however, have not eliminated the need within user organizations to develop custom
applications software tailored to specific organizational needs. How these applications are
implemented, and whether they can benefit from new development techniques will be major
concerns in coming years. For example, there are two current approaches for improving the
custom applications software development process.91 The first approach uses computer assisted
software engineering (CASE) methods to simplify, document, and maintain software. The second
approach uses a new class of object oriented programming languages (OOP) to provide standard
modules for the creation of more complex software systems including application programs.
From an economic viewpoint, the two approaches suggest somewhat different principles.
A main objective of CASE is reduction in the costs of creating custom software while maximizing
91 See Cusumano  and Nakahara  for different views of the Japanese approach to
the same problem.
the flexibility and heterogeneity of traditional approaches to this task. Expanded use of CASE will
accelerate the growth of customized software applications by lowering development costs. But the
flexibility that makes CASE attractive also may limit its application. Flexibility allows
programmers to address virtually any problem but also reproduces some of the productivity
problems of earlier custom software. For example, although CASE techniques simplify code
generation, they do not necessarily encourage the use of well-tested code modules. Thus, CASE
techniques can lead to errors and inconsistencies similar to those of custom programs where
programmers frequently reinvent (often with error) the methods of performing commonly
OOP presents an entirely different tradeoff. An "object" can be used across units within an
organization and across organizations. Application programs are assemblies of objects, some of
which are custom-coded for a particular organization’s needs. A measure of flexibility is
embedded within the design of each object. This approach can generate the same sorts of
externalities that were created by the mass market for personal computer software, as frequently
reused objects are optimized for specific environments and as users improve their skills in the use
of such objects. Moreover, as the software network of the corporation integrates objects, the
opportunity for external suppliers to enter is enhanced.92 But realizing advantages and funding
the costs of developing sophisticated objects requires widespread adoption of a limited number of
object libraries. Independent software vendors will offer many different object libraries in an
attempt to become dominant suppliers and unless a "shakeout" reduces their number, a classic
monopolistic competitive equilibrium (where costs of differentiation absorb potential profits) is
likely to occur. Such an outcome would divert resources from the incremental improvement of
object libraries and user skills in their use toward efforts to secure user adoption of one of many
competing object libraries. The latter goal is likely to result in object libraries that are more
specialized to particular classes of users and the outcome will be similar to existing product
differentiation within application programs. OOP may thus move the long-established tension
between standardization and customization in applications software to a new level, without
92 The intellectual property debates over "look and feel" and the alleged infringement by
Microsoft Windows on Apple’s operating system are only the beginning of these kinds of disputes
as users seek to construct a web of interactions among their programs.
The contest between CASE and "object" approaches to software development is only one
example of the new tradeoffs that flow from the enormous installed base of personal computers
and the growing use of networks. The contest between efficient solutions derived from custom
solutions and the broad adoption of "standard" approaches that rely on the economies of mass
reproduction are of growing importance in explaining the rate and direction of technical change in
software. A better economic understanding of the emergence of technical compatibility standards
and the conflicts between variety-enhancing competition and cost-reducing coordination are needed
to more fully understand these developments.
Future developments in the U.S. software industry are likely to be shaped by the mature
character of the supply "infrastructure" for software creation and the large installed bases of such
particular information technology systems as personal computers. The U.S. software industry
emerged from a complex "infrastructure" involving the interaction of computer producers, users,
and independent software vendors. These actors were joined, for a time, by the timesharing
vendors and, more recently, by the computer service consulting companies. The complexity of this
infrastructure provides the U.S. with an enormous variety of possible solutions to new challenges
presented by changes in hardware and application needs. Although there is no fundamental reason
why other nations cannot, in time, develop similar infrastructures for software creation, the U.S.
has an enormous lead in this process that is likely to be maintained. Among the sources of this
diversity were important events unique to the U.S. market, such as the early creation of an
independent software vendor sector because of IBM’s software unbundling decision, the U.S. lead
in creating new computer "platforms" along with rapid adoption of these platforms that have
stimulated complementary software development, and the aggressive adoption of information
technology by public and private U.S. institutions. These events have been reinforced by early
U.S. development of computer science education and widespread investment in on the job training
in the use of software systems, generous funding of software research and development by the U.S.
government, and the enormous size of the maturing U.S. market for hardware and software.
We now know that large installed bases, and the expectation of large future installed bases,
of computer systems enhances investment in software development. The existence of such
software supports expansion in the sales of compatible computer hardware as Chapter 4 in this
volume argues. Change in the software industry is likely to be driven by the proliferation of
computational "platforms," such as widely accepted personal computer and workstation models,
and the associated demand for large investments in software development to support application of
these platforms. Many "platforms" that will influence software development in coming years are
not yet available. The recent introduction of personal digital assistants (PDAs), handheld
computers that accelerate the trend toward portability, and plans for the development of customer
equipment, including PDAs, that will allow access to services offered over a "national information
infrastructure" of fibre optic telecommunication networks suggests that the coming decade may
initiate another cycle of investment growth in software creation that involves all of the actors that
have emerged to date and perhaps new ones as well.
Geoffrey D. Austrian, Herman Hollerith: Forgotten Giant of Information Processing, New York:
Columbia University Press.
Bashe et al 
Charles J. Bashe, Lyle R. Johnson, John H. Palmer, and Emerson W. Pugh, IBM’s Early
Computers, Cambridge: MIT Press.
Baumol, Blackman, and Wolff 
William J. Baumol, Sue Anne Batey Blackman, and Edward N. Wolff, Productivity and American
Leadership: The Long View, Cambridge: MIT Press.
James R. Beniger, The Control Revolution: Technological and Economic Origins of the
Information Society, Cambridge: Harvard University Press.
Barry W. Boehm, "Software and Its Impact: A Quantitative Assessment," Datamation, May, pp.
Barry W. Boehm, Software Engineering Economics, Englewood Cliffs, N.J.: Prentice Hall.
Dick H. Brandon, "The Economics of Computer Programming" in George F. Weinwurm, On the
Management of Computer Programming, New York: Petrocelli Books, p. 8.
Bresnahan and Greenstein 
Timothy F. Bresnahan and Shane Greenstein, "Technological Competition and the Structure of the
Computer Industry," Center for Economic Policy Research, CEPR Publication #315, Stanford
Gerald W. Brock, The U.S. Computer Industry: A Study of Market Power, Cambridge: Ballinger.
Frederick P. Brooks Jr., The Mythical Man-Month: Essays on Software Engineering, Reading,
Business Week 
Business Week, "Software: The Growing Gets Rough," Business Week, March 24, pp. 128-134.
Business Week 
Business Week, "Software: It’s a New Game," Business Week, June 4, pp. 102-106.
Computer and Business Equipment Manufacturers Association, The Computer and Business
Equipment Industry Marketing Data Book, Washington, D.C.: CBEMA.
John D. Cooper, "Corporate Level Software Management," IEEE Transactions on Software
Engineering, Vol. SE-4, no. 4, July, pp. 319-26.
Michael A. Cusumano, Japan’s Software Factories: A Challenge to U.S. Management, New York:
Oxford University Press.
Datamation, "The Top 50 Companies in the Data Processing Industry, Datamation, June, pp. 49ff.
Datamation, "The Datamation 100," Datamation, June, pp. 91ff.
Datamation, "The Datamation 100," Datamation, June, pp. 115ff.
Datamation, "The Datamation 100," Datamation, June 15, pp. 43ff.
Datamation, "The Datamation 100," Datamation, June 15, pp. 7ff.
Datamation, "The Datamation 100," Datamation, June 15, pp. 12ff.
Richard Thomas DeLamarter, Big Blue: IBM’s Use and Abuse of Power, New York: Dodd, Mead
Dolotta et al 
T. A. Dolotta, M. I. Bernstein, R. S. Dickson, Jr., N. A. France, B. A. Rosenblatt, D. M. Smith,
and T. B. Steel, Jr., Data Processing in 1980-1985: A Study of Potential Limitations to Progress,
New York: John Wiley.
Fisher, McKie, and Mancke 
Franklin M. Fisher, James W. McKie, and Richard B. Mancke, IBM and the U.S. Data Processing
Industry: An Economic History, New York: Praeger.
Kenneth Flamm, Creating the Computer: Government, Industry, and High Technology,
Washington, D.C.: Brookings Institution.
Herbert L. Gepner, "User Ratings of Software Packages," Datamation, December, 1977, pp. 118ff.
G. Glaser, "Keynote Address, 1974 National Computer Conference," Montvale, N.J.: American
Federation of Information Processing Societies.
Robert J. Gordon, "The Postwar Evolution of Computer Prices," in Dale W. Jorgenson and Ralph
Landau, eds., Technology and Capital Formation, Cambridge: MIT Press, pp. 77-126.
Robert J. Gordon, The Measurement of Durable Goods Prices, Chicago: University of Chicago
Shane Greenstein, "Lock-in and the Costs of Switching Mainframe Computer Vendors: What Do
Buyers See?," Faculty Working Paper 90-1718, Political Economy Series #48, University of
Gurbaxani and Mendelson 
Vijay Gurbaxani and Haim Mendelson, "Software and Hardware in Data Processing Budgets,"
IEEE Transactions on Software Engineering, Vol. SE-13, No. 9, September, pp. 1010-1017.
Hodges and Melewski 
Judith Hodges and Deborah Melewski, "Top 50: Profiles of the Leading Independent Software
Companies," Software Magazine, June, pp. 23ff.
IDC, Computer Industry Reports: The Gray Report, Framingham, Massachusetts.
International Data Corporation, Application Development Tools: 1993 Worldwide Software Review
and Forecast, Framingham, Massachusetts, December.
Juliussen and Juliussen 
Karen Juliussen and Egil Juliussen, The Computer Industry Almanac: 1991, New York: Simon and
Schuster (Brady Books).
Keen and Gerson 
Peter G. W. Keen and Elihu M. Gerson, "The Politics of Software Systems Design," Datamation,
November, pp. 80-84.
John L. Kirkley, "The Critical Issues: A 1974 Perspective," Datamation, January, pp. 64-67.
Kenneth Kolence, "Software View of Measurement Tools," Datamation, January, pp. 32-38.
Philip Kraft, Programmers and Managers, New York: Springer-Verlag.
Richard N. Langlois, "External Economies and Economic Progress: The Case of the
Microcomputer Industry," Business History Review, Vol. 66, Spring, pp. 1-50.
Gerald H. Larsen, "Software: A Qualitative Assessment," Datamation, November, pp. 60-66.
Tetsushi Nakahara, "The Industrial Organization and Information Structure of the Software
Industry: A U.S.-Japan Comparison," Center for Economic Policy Research, Stanford University,
CEPR Publication #346, May.
Organisation for Economic Co-Operation and Development (OECD), Software: An Emerging
Industry, Paris: OECD.
Montgomery Phister Jr., Data Processing Technology and Economics, Santa Monica: The Santa
Monica Publishing Co.
Emerson W. Pugh, Memories that Shaped an Industry: Decisions Leading to IBM System/360,
Cambridge: MIT Press.
Pugh, Johnson, and Palmer 
Emerson W. Pugh, Lyle R. Johnson, and John H. Palmer, IBM’s 360 and Early 370 Systems,
Cambridge: MIT Press.
Stephen S. Roach, "America’s Technology Dilemma: A Profile of the Information Economy,"
New York: Morgan Stanley (Special Economic Study), April 22.
William F. Sharpe, The Economics of Computers, (A Rand Corporation Research Study), New
York: Columbia University Press.
Siwek and Furchgott-Roth 
Stephen W. Siwek and Harold W. Furchgott-Roth, International Trade in Computer Software,
Westport, Connecticut: Quorum Books.
Software Publishers Association 
Software Publishers Association, Publishers’ Software Sales Report: 1988 Annual Report,
Washington, D.C.: SPA.
Software Publishers Association 
Software Publishers Association, Publishers’ Software Sales Report: 1989 Annual Report,
Washington, D.C.: SPA.
Software Publishers Association 
Software Publishers Association, Publishers’ Software Sales Report: 1990 Annual Report,
Washington, D.C.: SPA.
Paul A. Strassmann, The Business Value of Computers, New Canaan, Connecticut: The Information
L. A. Tomaszewski, "Decentralized Development," Datamation, November, pp. 61-64.
Frank V. Wagner, "Is Decentralization Inevitable?," Datamation, November, pp. 86-97.
Michael R. Williams, A History of Computing Technology, Englewood Cliffs, N.J.: Prentice-Hall.
Table 1. Phister and CBEMA Estimates of U.S.
Domestic Mainframe and Minicomputer Shipments
by Number and Value 1960-1990
Mainframe Mainframe Minicomputer Minicomputer
(nos.) ($ million) (nos.) ($ million)
Phister  Estimates
1960 1,500 560 300 30
1961 2,300 850 400 30
1962 3,100 1,060 400 30
1963 3,800 1,220 400 80
1964 5,100 1,570 500 100
1965 5,300 1,910 800 150
1966 7,000 3,200 1,000 130
1967 8,500 3,900 2,000 130
1968 7,400 4,650 3,500 185
1969 6,600 4,642 6,700 277
1970 5,040 4,073 9,500 282
1971 8,560 3,975 9,350 300
1972 10,970 5,170 15,100 450
1973 14,000 5,405 24,700 540
1974 8,900 6,220 34,000 810
CBEMA Estimates (from Juliussen and Juliussen )
1960 1,790 590 n/a n/a
1965 5,350 1,770 600 66
1970 5,700 3,600 6,060 485
1975 6,700 4,960 26,990 1,484
1976 6,750 5,060 39,320 1,887
1977 8,900 6,940 56,780 2,780
1978 7,500 6,230 68,340 3,690
1979 7,200 6,340 81,250 4,712
1980 9,900 8,840 105,870 6,238
1981 10,700 9,640 121,990 7,290
1982 10,600 9,860 128,000 7,770
1983 9,980 9,780 146,800 8,979
1984 11,330 11,900 205,400 12,817
1985 10,910 11,890 190,800 11,696
1986 10,990 12,200 198,200 11,872
1987 11,200 12,660 205,800 12,080
1988 11,540 13,270 218,100 12,656
1989 11,890 13,790 227,700 13,093
1990 12,130 14,190 232,000 12,650
Note: n/a means not available
Table 2. CBEMA Estimates of U.S. Domestic
Software and Service Revenues 1965-1988
Processing Software Professional
Year Services Products Services Total
1965 n/a n/a n/a 200
1970 1,200 500 800 2,500
1975 3,300 1,000 2,200 6,500
1980 10,800 2,850 4,350 18,000
1981 11,550 3,950 5,500 21,000
1982 12,650 4,900 5,950 23,500
1983 14,400 6,900 6,900 28,200
1984 17,150 10,000 8,100 35,250
1985 19,310 12,120 9,270 40,700
1986 20,750 14,150 10,100 45,000
1987 23,600 18,500 11,750 53,850
1988 26,900 27,850 13,300 68,050
Note: n/a means not available
Source: As cited in Juliussen and Juliussen .
Table 3. Packaged Software and Computer Systems:
The Datamation 1977 Survey of Computer Software
Datamation, at that time the leading business publication in the data processing industry, mailed surveys to
30,000 computer installations and received 5,813 usable responses. Datamation then reported names and
other information about software packages including user satisfaction for software packages mentioned by 5
or more respondents. Although 1,200 packages were mentioned by users, only 199 were mentioned by 5 or
more users. The following tabulation is based on the information about computer system compatibility from
the description offered. In some cases it was not possible to determine for which computer system the
sofware package was designed and in many cases the class of computers is inferred from other information,
e.g. operating system compatibility.
Computer System Type Number Total
Mainframe Computers 138 72
IBM System/3 and 32 20 10
Minicomputer Systems 19 10
Cross Platform or
High Level Language 10 5
Other (e.g. IBM 1130) 5 3
Total for Percentages 192 100
Cannot Determine 7
Total in Listing 199
Source: Gepner .
Independent Software and Service Companies--1980
Among the Datamation 100, the 100 largest U.S. data processing companies based on EDP revenue, 25
companies were predominantly engaged in the provision of software or services in 1980.
Type Predominant Source of Revenue Number
A Software and Programming Services 4
B Timesharing and On-Line Services 7
C Systems Integration 2
D Specialized Services 12
These 25 companies, by type, were:
1980 DP Revenue Datamation Comment or Specialized
Company ($ million) 100 Rank Service (if applicable)
Type A: Software and Programming Services
SDC 187 43 Large software systems
Informatics 126 56 Large software systems
Dunn and Bradstreet 97 67 "Nomad" database product
Management Sciences America 53 96 Finance and Human
Type B: Timesharing and On-Line Services
General Electric 475 17
McDonnell-Douglass 280 25
Tymshare 211 38
Boeing 125 57
United Telecom 115 61
Comshare 88 68
Martin Marietta Data Systems 78 74
Type C: Systems Integration
Computer Sciences Corp. 560 15
EDS 408 18
Type D: Specialized Services
ADP 505 16 Accounting/Payroll Services
General Instruments 172 45 Wagering Systems
Bradford National 143 50 Financial Services
Planning Research 127 55 Government services
Reynolds and Reynolds 118 59 Auto dealer services
Shared Medical Services 106 64 Medical & hospital
The Sun Company 87 69 Financial and disaster
Interactive Data Corporation 69 81 Financial and data services
(subsidiary of Chase Manhattan)
Commerce Clearing House 67 83 Tax preparation
Anacomp 57 89 Financial services
National Data Corporation 53 94 Cash management services
First Data Resources 53 95 Credit card authorizations
(subsidiary of American Express)
Source: Datamation 
Software and Service Companies--1991
Among the Datamation 100, the 100 largest U.S. data processing companies based on EDP revenue, 41
companies were predominantly engaged in the provision of software or services in 1991.
Type Predominant Source of Revenue Number
A Software and Programming Services 17
B Timesharing and On-Line Services 0
C Systems Integration 13
D Specialized Services 11
These 41 companies, by type, were:
1991 DP Revenue Datamation Category Specific
Company ($ million) 100 Rank Annotation
Type A: Software and Programming Services Platform
Microsoft $2,276 12 Personal Computer
Computer Associates 1,438 23 Mostly Mainframe
Oracle 1,085 29 Mini and Mainframe
Lotus 829 40 Personal Computer
Novell 710 44 Network software
Dunn & Bradstreet Software 549 56 Mainframe
Wordperfect 533 57 Personal Computer
Borland 502 60 Personal Computer
Mentor Graphics 400 67 Specialized Workstation
ASK Computer Systems 395 68 Minicomputer
Cadence 353 73 Workstation
SAS Institute 295 79 Mini and mainframe
Autodesk 285 82 Personal Computer
Adobe 230 92 Personal Computer
Sterling Software 229 93 Mainframe
Legent 208 97 Mainframe
Information Builders 202 99 Mini and Mainframe
Type B: Timesharing and On-Line Services
[no companies were listed for the 1991 Datamation 100. The new leading companies are Prodigy,
Compuserve, and America Online.]
Type C: Systems Integration Parent Company
EDS 3,666 7
Andersen Consulting 2,260 13 Accounting Firm
Computer Sciences Corp. 1,945 15
Price Waterhouse 733 43 Accounting Firm
Science Applications Intl. 653 46
Planning Research 623 49
SHL Systemhouse 601 50
Coopers & Lybrand 571 52 Accounting Firm
Martin Marietta 561 53
Ernst & Young 551 54 Accounting Firm
Systematics Information 377 70
MAI Systems 329 76
Computer Task Group Inc. 285 81
Table 5. (cont.)
Software and Service Companies--1991
Type D: Services Predominant Service
ADP 1,933 18 General
American Express/First Data 995 33 Transactions processing
Nynex 601 51 Telecom and Financial
Bell Atlantic 550 55 Maintenance
Shared Medical 439 66 Hospitals
Policy Management Systems 342 75 Insurance
Boeing Information Services 325 77 Aerospace (Services to
Boeing were 80% of
American Management Systems 285 80 Telecom and Financial
Fiserve 281 84 Financial
Reynolds and Reynolds 233 90 Auto Dealers
National Data 221 95 Financial
Source: Datamation .
Table 6. Software and Service Revenues
of Major Computer Manufacturers 1981 and 1991
Company Software & Services Software & Services
(ranked by total DP revenue) Revenue Revenue as %
($ million) of Total Revenue
IBM 4,480 17.0
DEC 911 25.4
Control Data 1,154 37.2
NCR 1,029 33.5
Burroughs 838 28.6
Sperry 695 25.0
H-P 545 29.1
Honeywell 835 47.1
Xerox 209 19.0
Data General 130 17.0
Total 10,826 22.9
(ranked by total DP revenue) Software+Services Revenue as % Software Services
Revenue of Total Revenue Revenue Revenue
IBM 12,542 20.0 10,524 2,018
DEC 2,366 16.6 796 1,570
H-P 345 3.2 345 0
AT&T (incl. NCR) 850 10.4 250 600
Unisys (incl. Burroughs/Sperry) 1,200 15.0 600 600
Apple 250 3.8 250 0
Sun 175 5.1 175 0
Xerox 220 7.5 100 120
Tandem 140 7.2 110 30
Prime 247 17.9 247 0
Data General 210 17.3 210 0
Control Data 460 39.2 160 300
Total 19,005 15.5
Honeywell’s computer division was spun--off and purchased by Bull of the U.K in 1987
Sources: Datamation (1982), Datamation (1992).
Table 7. U.S. Based Company Computer Industry Shipments by Value and Number 1980-1990
Large-Scale Large-Scale Medium-Scale Medium-Scale Small-Scale Small-Scale Personal Computer Personal Computer
Shipments Shipments Shipments Shipments Shipments Shipments Shipments Shipments
Year by Number by Value ($ millions) by Number by Value ($ millions) by Number by Value ($ millions) by Number by Value ($ millions)
1980 2,380 7,880 16,100 7,300 197,800 7,700 486,000 1,642
1981 1,510 6,150 19,200 9,160 212,500 8,800 905,000 2,707
1982 2,300 11,140 24,100 9,100 248,300 9,100 3,775,000 5,358
1983 3,390 14,460 28,000 9,530 282,390 9,800 7,623,000 11,304
1984 3,750 16,100 38,650 13,400 338,800 11,100 9,670,000 17,168
1985 3,240 16,970 39,650 14,610 325,400 12,320 8,828,000 19,070
1986 3,260 17,560 44,900 15,750 354,700 13,500 9,816,000 20,720
1987 3,380 18,340 49,900 17,500 391,200 14,980 11,130,000 22,650
1988 3,530 19,360 56,000 19,600 447,000 16,810 12,380,000 25,150
1989 3,780 20,520 63,000 21,850 512,300 18,720 13,500,000 27,800
1990 4,000 21,900 68,900 24,000 582,800 20,500 14,560,000 30,100
Source: IDC (1992)