United States. Congress. Senate. Committee on the.

The Industrial reorganization act. Hearings, Ninety-third Congress, first session [-Ninety-fourth Congress, first session], on S. 1167 (Volume pt. 7) online

. (page 89 of 140)
Online LibraryUnited States. Congress. Senate. Committee on theThe Industrial reorganization act. Hearings, Ninety-third Congress, first session [-Ninety-fourth Congress, first session], on S. 1167 (Volume pt. 7) → online text (page 89 of 140)
Font size
QR-code for this ebook


rental price. The total value of the installed stock for each manufacturer, and
for the industry as a whole, is obtained by summati(m over the appropriate
models. The data in Table 2 are based on the installation censuses printed in
Computers and Automation (through 1962) and in EDP Industry and Market
Report (thereafter), because I believe them to be the most reliable data that
are available to the public. However, it must be kept in mind that these so-
called censuses are in fact incomplete market surveys taken by private firms,
and they are subject to an imknown margin of error.

The period through February 1958 represents the beginning of the computer
industry, and the market shares in Table 2 show that only the earliest entrants
played a significant role. Even Honeywell and RCA, wliich do appear in the
1958 cen.sus, had installed only one and three, computers, respectively, and the



* The foregoing description Is offered primarily as a background for the analyses that
follow. The author does not wish to suggest that either the concepts or the factual asser-
tions are necessarily appropriate to any litigation under the present antitrust laws of the
•United States.



5624

size of their shares is a vivid reminder of how small the market was then. On
tlie other hand, NCR had considerable experience with military computers, but
this is not reflected in a tabulation of shares in the general purpose market.
Computers installed in January 1960 were still overwhelmingly first generation
(vacuum-tube) systems, and tlie years 195S and 1959 are thus the height of the
first generation. However, by 19.19 the market was looking toward solid-state
systems, though onlv a few' were being produced, and this period is thus in
some ways a transitional one— Burroughs retained a strong grip on third place
in January 1960 bv concentrating its efforts on the vaciuun-tui>e model B-220,
but it paid the price by performing very badly in 1960 and 1961. when only
solid-state computers were selling well, and it had fallen to sixth by January
1962. The second generation (transitorized computers) continued through 1964,
with 1965 another transitional year (note the relatively small increase in the
installed base from January 1965 to Jaimary 11>66). and 1966 marks the begin-
ning of the third-generation (integrated circuitry) boom.

The principal observation on Table 2 is that IBM dominntes the computer
industry, Mdth a market share in the neighborhood of 70 percent, and there is no
one firm with a strong second position. Table 2 does not indicate any clear trend'.
in IBM's market share. This share has fallen a few percentage points from its
195S-1969 plateau, but the decline was essentially complete by January 1962,
with only mild undulations thereafter. The decline from January 1965 to
January 'l966 marks IBM's changeover to its System /360 line of computers,
and the surge to January 196S reflects the final arrival of all the basic models
in System/360 to full production. The conclusion, then, is that IBM had a few
exceptionally good years toward the end of the firs^t generation, when Univac
was in disarray and before other competition could make itself felt strongly,
but there is no evidence of a trend in IBiNI's ninrket share in the .second and
third generations, nor does tlie approximately TO percent attained then differ
from IBM's early achievements, up to late 1956.

The clearest trend visible in Table 2 is the decline of Univac. The main decline
in Univac's share of new production was already complete l)y 1957, but there
was a further small decline in the 1960's after a plateau from 1963 through
1965. The decline in Univac's share of the installed l)ase, as shown in the table,
is both .-'.ower and more persistent (and therefore more easily spoted as a trend),
because the vestiges of Univac's former eminence did not disappear from the
computer stock until 1964.

It is also clear from these aggregate market share data that Control Data,
General Electric, and Honeywell all carved out important positions in the non-
IBM part of the computer market. Control Date did so quickly, in the five
years from 1960 through 1964. whereas Honeywell and General Electric experi-
enced longer and slower gains. Otlier firms with upward trends are DEC. SDS.
and perhaps NCR. though 1. 1 one of these has gone so far or so fast as did Control
Data, General Electric, and Honeywell. Finally, the shares of Burroughs and
RCA s^em to be approximately stable.

Behind IBM, no one firm has a strong hold on second place. Univac held that
position through January 1969. but by a decreasing margin, with Honeywell.
General Electric, and Control Data all challenging for the second position, and
RCA not much further behind.

The foregoing description of market shares has used data only on the installed
base of computer systems. Two other measurement l>a.ses are also commonly
discussed: revenue ; and the value of current production and shii>ments. Each
measurement basis has its advantages and disadvantages, primarily of a tech-
nical nature. But when the data are tallied, the scores do not differ by more than
a few percent, and these differences are too small to be of major concern in the
context of this inquiry. The reasons for this judgment is that both economic
theory and empirical evidence indicate that the difference of a few percentage
iwints in. say, the combined market shares of the top four firms is not likely
to have much effect on either the conduct or performance of an industry.' In
the computer industry, it is easily seen (and already well known) that there
is a very high degree of seller concentration. It is equally ckvar that this high
level of seller concentration is due to the overwhelming size of a single firm,
rather than to the presence of several firms, each with a moderately large market
share. Beyond this, there is nothing that further study of market shares can
reveal about seller concentratioai as. a static element of market structure. I
therefore turn now from the question of whetlier IBM dominates the computer
industry to the analytically more challenging question of how and why this
dominance was achieved and maintained.

6 Joe S. Bain, Industrial Organization, ch. x, especially pp. 413-314.



5625



■ -^ (Ti Ol CO OO 00 CD



* OO CO ' CTJ ^



^


ir>


•^


inr


■0 r^


UD


«


r^


'^


CO


r-


Oi


^


i-^


^'






rvi




CM


en




oo


r-,


r^




























oo






oo




*3-


CD












ro
















oo








— — ^mE

•2 5 ^5iJ



5626

HOW IBM ACHIEVED AND MAINTAINED ITS DOMINANCE OF THE COMPUTER INDUSTET

Identification of the factors accounting for IBM's dominance of the computer
industry is needed to facilitate consideration whether structural change in the-
industry is desirable and feasible. These fatcors fall into four classes. First,
there are some advantages of large scale in the computer industry. These are
advantages that accrue to any firm that is bigger than it rivals, and that thus
tend to destabilize a competitive market structure. Second, there are structural
characteristics tending to stabilize market shares. These forces work against
the creation of market dominance from a competitive situation, but they also
help preserve market dominance if it can once be achieved. Third, IBM has
some special advantages in the computer industry, which it owes to historical
circumstance. Finally, IBM's conduct in the market has itself had a impact on
the structural basis for market dominance, and it may or may not also have
involved repressive and disciplinary practices that have characterized some
other monopolies in the past. These four groups of factors are considered in turn.

The first set of explanatory factors are advantages of large scale, and the first
of these to come to mind are economies of large scale (lower unit costs at
higher rates of output). The classic locus for scale economies is in the i)roduc-
tion of goods, which in the computer industry encompasses the manufacture of
components, their assembly into complete units of equipment, and the design
and development activity associated with the manufacturing process. In the
first half of the 1960's tiie economies of large scale in the area were not sub-
stantial relative to the size of the market. But in the second half of the decade,^
and more specifically with the arrival of integrated or hybrid integrated circuitry,
the situation may have changed. One estimate by a knowledgeable but anony-
mous observer placed IBM's direct costs of hardware production as much as
one-fourth or one-third lower than the direct costs of its rivals. With typical
direct costs (for everyone except IBM) a bout 30 to 35 percent of selling prices,
this advantage, expressed relative to revenues, may give IBM a margin of as
much as ten percent over its rivals. More recent judgments by other observers
have questioned this conclusion, and several experts familiar with the industry's
cost structure believe that IBM has essentially no production cost advantages
due to economies of large scale.

The true picture of scale economies in computer production is very difficult
to discern, even with better information than is available to the public, and this
for two reasons. First, measurement of the product is so difficult that what are
thought to be cost differences are in fact product differences."

Second, IBM's scale of operations is at least a full order of magnitude larger
than any of its rivals (this margin may have fallen to a factor of about seven
with the Honeywell acquisition of GE's computer business in 1970), and the
market therefore offers no experience showing whether the scale economies, if
any, enjoyed by IBM could be achieved with only ten or thirty or fifty percent
of the market.

A second locus for advantages of large scale in the computer industry is in
marketing and distribution, including maintenance. Here it is generally agreed
that IBM has substantial cost advantages that result from its large scale. How-
ever, it is not clear that these cost advantages reflect anything more profound
than the high success ratio that must attend the marketing efforts of a firm
with 70 percent of the market. IBM obviously does not dissipate a large fraction
of its sales effort on customers it eventually loses, whereas its rivals do bear
a much greater biirden of fruitless sales calls.

The third of the advantages of large scale in the computer industry is the
benefit from external econonues in the use of computers. External economies
are benefits derived by each user of computers from the activities of other
users. Owing to product differentiation, the external economies created by instal-
lations of one computer model benefit primarily the other users of computers of
that model. The nature of these external economies is described first. Then it
is shown that they redound to the benefit of the computer manufacturers, and
that the benefits derived by each manufacturer are directly related to its share
of the computer market.



8 As an analogy, suppose the existence of extreme product differentiation made It pos-
sible for General Motors to sell its Chevrolet Vega for the same price as Chrysler's much
larger Dodge Dart. An observer unable to identify the size differences between the two
cars would still note GM's lower costs, and they might well be attributed to economies of
large scale. But the proper explanation of the postulated situation is that GM's product
differentiation advantages enable it to sell a smaller and less expensive product for the
same price as its rival sells a larger and more expensive product.



5627

External economies in the use of computers occur in five ways. The first is
in the market for computer time. It is a rare user who needs all of the time
on his own computer, and never anything more. If for no other reason, this
is true because the computer may break down at a crucial moment when data
must be processed ; but even if the average load is very close to capacity, the
load is apt to rtucturtte from day to day (or hour to hour). One way to deal
with periods of excess demand (or overcapacity) is to buy time on another
computer like one's own (or sell one's own surplus time). But the market
for computer time can operate successfully only if the number of participants ia
large enough so that during most time periods the random flnctuations of the
different users' supply and demand are canceled by averaging.

Exactly paralled to this argument is one concerning the market for used com-
puters. Since brokerage is clearly subject to economies of scale in the relevant
range, and since savings in it are apt to be passed on to the owners of the hard-
ware, the more widely used a particular model, the less it depreciates. The pros-
pective saving in transportation costs — because buyers and sellers of used com-
puters can be more closely matched in space, the larger the number of trans-
actions — also reduces the depreciation of computers used in larger numbers.''

The third source of external economies in the use of computers is the market
for programming services and other assistance in the use of computers, as they
are provided by service bureaus, software houses, and EDP consultants. These
firms are apt to have more experience in working with a more widely used,
model, and this greater experience makes it easier for a user of such a model
to obtain help from them.

User organizations are a fourth source of external economies. They offer, in
addition to exchange of programs, everything from advice about running' an
EDP shop to a semi-annual junket for the shop's chief. And despite the frequent
criticism that junketing is the only "benefit," there is little doubt that the exist-
ence of a user organization does add to the desirability of the model it repre-
sents. The larger the user organization, the more programs there are to exchange,
and thus the greater the benefit to each individual user.

Finally, the software for a widely used computer system gets debugged faster
than for a less widely used model, because greater use uncovers the bugs faster.*

An important characteristic of these external economies is that their amount
is greater, the larger the number of installations of a computer model. They are
thus external economies in the Marshallian sense of reducing costs within aa
industry as the scale of production in that industry increases." The benefits,
however, need not go to the firms using the production process subject to external,
economies (in a competitive industry in the long run, nothing benefits the pro-
ducers). Instead they may accrue to factors of production used in the process
.subject to external economies, because they raise the demand schedule for those
factors relative to the demand for factors of production used in closely substitut-
alile processes not benefiting from these external economies to the same extent.^"

In EDP, the process subject to external economies is the use of a particular
computer model, along with complementary inputs, to process data electronically.
Since the use of EDP is atomistic, and since all the inputs except computers are
atomistically supplied, each computer manufacturer is able to appropriate the
value of the external economies obtained in the use of its comnuters (which the
manufacturer does by charging a higher price, or by attracting more customers
with a price that does not fully recover the value of the external economies).
Finally, since the amount of the external economies is directly related to the num-

■?To the extent that manufacturers act as brokers for both time and used machines

the former in service bureaus, the latter owing to the prevalence of leaslns these -idv^n-

ta«es may be interpreted as scale economies In distribution. This means that manufacturers
benefit directly by Internalizing these external economies, rather than bv the indirect'
process described in the text, infra.

l-^."''] [^^fferently, the more users there are, the fewer bugs each one discovers and is

ocp'^'o''ti^ Mars;l]all, Principles of Economics (Sth ed. ; New York: Macmillan, 1948), pp.
2(,G. 271-(2. Marshan repeatedly emphasizes the parallelism between external and
the'srale^oT^he^^trv depending on the scale of the firm, the former depending on

w The external economies also tend to lower the price of the product produced bv the-
process benefiting from them. But in the case of EDP. the services produced bv different
computer models are so closely substitutable that their relative valuercan hardlv be
f llf'lK^^^ differences in the cost of obtaining these services. Rather, what must happen
is that these cost differences wiH be absorbed by the manufacturers of the widely used
moaels as econoinic rents, or they will be used to keep computer prices below the point
driven from%heindusS' ''''''''^''"y ^^""^ external economies to so great an extent a?e.



5628

lier of computers of a model in use — that is, to the marlcet share of the manu-
facturer, given the total demand for computers, the larger manufacturers stand
to gain more than the smaller ones.

The last major advantage of large scale in the computer industry is the benefit
that comes from being the dominant firm in an industry where product differ-
entiation and integration are extremely important. Electronic data processing
is a complex of interrelated activities involving much more than just the pro-
duction of computer systems, and IBM's great size enables it to enfold more of
these activities than its rivals. This integration enables IBM to relieve the cus-
tomer of some of the responsibilities and woi-k of using computer systems to
achieve results in his business. In contrast, customers of other manufacturei-s
are forced to learn more about what they are doing in EDP, because the other
computer manufacturers do not have the broad expertise to step in and help run
their customers' businesses. IBM also benefits from its large size even in the areas
where it is not integrated. Firms offering products and services for use with
computer systems, such as digital communications equipment, plotters or other
specialized peripherals, and tailored .software packages, must all achieve a proper
interface with the computer systems their products complement IBM's dominance
of the computer industry ijennits it to in effect establish the standards for these
interfaces, to which the complementary products are designed ; whereas IBM's
rivals mn.st adjust to the market (which really means adjusting to IBM) in-
stead of having the market adjust to them.

Related to these two points is the way that product differentiation and IBM's
image of dominance affects the motivations of the corporate executive choosing a
compuer for his firm : it is said that no vice president for EDP has ever been fired
for choosing IBM, even when the new computer system failed to live up to the
expectations of top management ; but many have been fired for not choosing IBM.
The point is that some mistakes are always going to be made in computer procure-
ment, but top management will not second-guess against the choice of the dominant
firm, and the safe course for the middle-level executive making the decision is
therefore to stick with the leader.

The second major group of factors explaining IBM's dominance of the computer
industry contains the forces that tend to stabilize market shares. Chief among
these forces is product differentiation, which is even more important here than
as an advantage of large scale. Product differentiation tends to stabilize market
shares because it tends to make computer users captives of their current sup-
pliers, and most of the demand for computers is a demand for the expansion or
replacement of existing systems. Customers are captives of their suppliers
because their applications programs, their communications equipment, and their
entire EDP experience are compatible only with systems produced by their current
supplier. Expansion or replacement with compatible computer equipment is thus
essential to the avoidance of large conversion co.sts, and owing to product differ-
entiation this compatibility is ordinarily found only in the product line of the
current supplier.

One other factor is significant among those tending to stabilize market shares.
This factor is the impact of the extra orflinnrily rapid growth of the computer
industry as a whole. In this hothouse environment, it is an enormous ta'^k for a
firm to acquire, absorb, and organize the factors of production — both labor and
capital — needed merely to keep pace with the industy. One may well ask how a
firm might be expected to increase its share of the computer market from two or
three percent to five. ten. or twenty percent in. say, three, five, or ten years, when
it must double in size every two or three years (as in the 1960's) merely to avoid
losing in market share. Yet growth potential in this range is exactly what was
needed to reduce IBM's dominance of the computer industry.

The third group of factors explaining IBM's rlorainance of the computer in-
dustry are the special historical circumstances that attended IBM's rise to
dominance. In a period of five years, from 19.51 to 19.")^. the computer was trans-
formed from an engineering tool, built and used by engineers, into a commercially
manufactured product, built in quantity and sold for use on a wide variety of
problems. The story of this transformation is one of the great tales of entre-
preneurship, and it is the story of IBM's success and Univac's failure."^



"This «torv has loner boen a lesrenfl in the compntpr inrtnstry. Genrjre .'^ohnssel ("IBM
v^ RKArR\Nb" D.atnmntion XI. Nos. 5 and [196.51) has recorder! the story, examined
it f-arefiilly. and verified the legrend behind it. The account here draws extensively, and
without further citation, on Schussel's work.



5629

Univac was the creation of Eckert and Manchly." After producing the ENIAO
and doing mucli of the work on its successor, the EDVAC, they left the University
of Pennsylvania to go into EDP for profit. Striking out alone, they established a
small corporation, which quickly obtained a contract to sell a Univac I to the
U.S. Bureau of the Census. In 1950, with the first Univac I not yet finished, Eckert
and Mauchly sold out to Remington Rand, joining its new Univac Division in
high executive posts.

The first Univac I was finally delivered to the Census Bureau in March 1951,
and the model was juade available to other buyers (over 30 were eventually sold).
It was the first stored-prograni computer and the first computer of substantial
scale to be sold commercially. When it appeared, IBM offered nothing better than
the older Card-Programmed Calculator, though it had begun developing the IBM
701 late in 1950. First deliveries of the 701 began in 195o, almost two years after
Univac I, but not much more than two years after development had begun. In
contrast, development of Univac I had started in 1947, and there were many
revisions before the model was finally put !nto production.

Univac, having established a lead in powerful computers with Univac I, held it
until 1956, when the appearance of the IBM 705 and 704 tipped the balance the
other way." There are several reasons for this swing. First is salesmanship.
Remington Rand was a poor second to IBM in the office machinery bu.siness, and
its sales force was thin, if not seriously understaffed. Called upon to sell com-
puters for business use, and therefore to the same group of customers it had
always been dealing with, it was unable to bear this additional burden as well as
was IBM's large sales force. Moreover, IBM had long had the attitude that it was
selling data processing services, not office machinery. This attitude was extremely
appropriate when data processing became electronic and computers were added
to- punched cards : obtaining data processing services from computers turned out
to be much more difficult than producing the hardware itself. IBM's attitude led it
to provide mur-h greater assistance in using computers than did Univac; and
since this assistance, rather than hardware, is what customers needed, they
bought from IBM.

The growth of business EDP. which IB^M pioneered through its effective sales
force, marks the ascendancy of IBM. Univac initiated this growth, but in so
doing was unfortunately involved in a fiasco. The installation of a Univac I in
1954 at the Gi-neral Electric plant in Louisville was widely regarded as a major



Online LibraryUnited States. Congress. Senate. Committee on theThe Industrial reorganization act. Hearings, Ninety-third Congress, first session [-Ninety-fourth Congress, first session], on S. 1167 (Volume pt. 7) → online text (page 89 of 140)