Technology change & the rise of new industries


Published on

Using an analysis of many existing and emerging industries, this book (to be published by Stanford University Press) shows how one can analyze the timing of new industry formation. It does this by analyzing the improvements in cost and performance that have enabled new technologies to become economically feasible.

Published in: Business, Technology

Technology change & the rise of new industries

  1. 1. Technology Change andthe Rise of New Industries by Jeffrey L. Funk 1
  2. 2. Table of ContentsChapter 1. IntroductionPart I. What Determines the Potential for New Technologies? Chapter 2. Technology Paradigm Chapter 3. ScalingPart II. When do Technological Discontinuities Emerge? Chapter 4. Computers Chapter 5. Magnetic Recording and Playback Equipment Chapter 6. SemiconductorsPart III. Opportunities and Challenges for Firms and Governments Chapter 7. Competition in New Industries Chapter 8. Different Industries, Different ChallengesPart IV. Thinking about the Future Chapter 9. Electronics and Electronic Systems Chapter 10. Clean EnergyChapter 11. Concluding RemarksAppendix. Research MethodologyList of References 2
  3. 3. Chapter 1 Introduction The U.S. and other governments spend far more money subsidizing the production ofclean energy technologies such as electric vehicles, wind turbines, and solar cells than theydo on research and development for clean energy i. Why? A big reason governments isbecause many believe that costs fall as a function of cumulative production in a so-calledlearning or experience curve, and thus stimulating demand is the best way to reduce costs.According to such a curve, product costs drop a certain percentage each time cumulativeproduction doubles as automated manufacturing equipment is introduced and organized intoflow lines ii. Although such a learning curve does not explicitly exclude activities done outsideof a factory, the fact that these learning curves link cost reductions with cumulativeproduction focuses our attention on the production of a final product and implies that learningdone outside of a factory is either unimportant or is being driven by the production of a finalproduct. But is this true? Are cumulative production and their associated activities in a factory themost important sources of cost reductions for these types of clean energy or any othertechnology for that matter? Among other things, this book shows that most of theimprovements in wind turbines, solar cells, and electric vehicles are being implementedoutside of their factories and that many of these improvements are only indirectly related toproduction. Engineers and scientists are increasing the physical scale of wind turbines,increasing the efficiencies and reducing the material thicknesses of solar cells iii , andimproving energy storage densities of batteries for electric vehicles, primarily in laboratoriesand not in factories. This suggests that increases in production volumes, particularly those ofexisting technologies, are less important than increases in spending on R&D (i.e., supply-sideapproaches), an argument that Bill Gates iv and other business leaders regularly make. 3
  4. 4. Although demand and thus demand-based subsidies do encourage R&D v, only a smallportion of these demand-based subsidies will end up funding R&D activities. Should this surprise us? Consider computers (and other electronic products such as mobilephones vi). The implementation of automated equipment and its organization into flow lines inresponse to increases in production volumes has not been the main reason for the dramaticreduction in the cost of computers over the last 50 years. The cost of computers primarilydropped for the same reasons that their performance rose: continuous improvements inintegrated circuits (ICs) have led to improvements in the cost and performance of computers.Furthermore, the improvements in the cost and performance of ICs were only partly from theintroduction of automated equipment and their organization into flow lines. A much biggerreason was large reductions in the scale of transistors, memory cells, and other dimensionalfeatures where these reductions in scale required improvements in semiconductormanufacturing equipment. The equipment were largely developed in laboratories, thesedevelopments depended on advances in science, and their rate of implementation dependedmore on calendar time (think of Moore’s Law) than on the cumulative production volumesfor ICs vii.1.1 New Questions and New Approaches We need a better understanding of how improvements in cost and performance emergeand of why they emerge more for some technologies than others, issues that are largelyignored by books on management (and economics). While most such books are aboutinnovative managers, innovative organizations, and their flexibility and open-mindedness,such books don’t help us understand why some technologies experience more improvementsin cost and performance than do others. In fact, they dangerously imply that the potential forinnovation is everywhere and thus all technologies have about the same potential forimprovements. 4
  5. 5. Nothing can be further from the truth. ICs, magnetic disks, magnetic tape, optical discs,and fiber optics have experienced what Ray Kurzweil calls “exponential improvements” incost and performance in the second half of the 20th century while mechanical components andproducts assembled from them did not viii. Mobile phones, set-top boxes, digital televisions,the Internet, automated algorithmic trading (in for example hedge funds), and onlineeducation have also experienced large improvements over the last 20 years as they benefitedfrom improvements in the above-mentioned technologies. A different set of technologies (e.g.,steam engines, steel, locomotives, and automobiles) also experienced large improvements inboth cost and performance in the 18th and 19th centuries. An understanding of why sometechnologies have more potential for improvements than do others is necessary for firms,governments, and other organizations to make good decisions about clean energy and othernew technologies. We also need a better understanding of how science and technology determines thepotential of new technology. Although there is a large literature on how advances in sciencefacilitate advances in technology in the so-called “linear model of innovation ix,” many ofthese nuances are ignored once learning curves and cumulative production are considered.For example, improvements in solar cell efficiency and reductions in material thicknessinvolve different sets of activities and the potential for these improvements depend on thetype of solar cells and on the level of scientific understanding for each type. Lumpingtogether the cumulative production from the different types of solar cells together causesthese critical nuances to be ignored and thus prevents us from implementing the best policies. Part of the problem is that we don’t understand what causes a time lag (often a long one)between advances in science, improvements in technology that are based on this science, andthe commercialization of the technology. And without such an understanding, how can firmsand governments make good decisions about clean energy or more fundamentally how canthey understand the potential for Schumpeter’s so-called “creative destruction” and new 5
  6. 6. industry formation? A new industry is defined as a set of products or services that are basedon a new concept and/or architecture where the products or services are supplied by a newcollection of firms and their sales are of a significant amount (e.g., greater than $5 billion).According to Joseph Schumpeter, waves of new technologies (that are often based on newscience) have created new industries along with opportunities and wealth for new firms asnew technologies have destroyed existing technologies and their incumbent suppliers. This is also a book about why specific industries emerge at certain moments in time. Forexample, why did the mainframe computer industry emerge in the 1950s, the personalcomputer (PC) one in the 1970s, the mobile phone and automated algorithmic trading ones inthe 1980s, the World Wide Web in the 1990s, and online universities in the 2000s? On theother hand, why hasn’t personal flight, underwater, or space transportation industries emerged,in spite of large expectations in the 1960s x? Similarly, why hasn’t electric vehicle, wind, andsolar industries yet emerged, or when will ones emerge that can exist without subsidies? Parts of these stories concern policies and strategies. When did governments introduce theright polices and when did firms introduce the right strategies? But parts of these stories alsoinvolve science and technology, and as mentioned above, these parts have been largelyignored by management books on technology and innovation xi, even as the rates of scientificand technological change have accelerated and the barriers to this change has fallen xii. Whenwas our understanding of scientific phenomenon or the levels in performance and price forthe relevant technologies sufficient for industry formation to occur? We need better answersto these kinds of questions in order to complement research on government policies and R&Dstrategies for firms. For example, understanding the factors that impact on the timing ofscientific, technical, and economic feasibility can help firms create better product andtechnology roadmaps, business models, and product introduction strategies. They can helpentrepreneurs understand when they should quit existing firms and start new ones xiii. Theycan also help universities better teach students how to look for new business opportunities 6
  7. 7. and address global problems; such problems include global warming, other environmentalemissions, the world’s dependency on oil and minerals from unstable regions, and a lack ofclean water and affordable housing in many countries. Examples of the problems that arise when firms misjudge the timing of economicfeasibility can be found in the mobile phone industry. In the early 1980s studies concludedthat mobile phones would never be widely used while in the late 1990s studies concluded thatthe mobile Internet was right around the corner. In both cases these studies misjudged the rateat which improvements in performance and cost would occur. In the former, the studiesshould have been asking what consumers would do when Moore’s Law made handsets freeand talk times less than 10 cents a minute. In the latter, the studies should have beenaddressing the levels of performance and cost needed in displays, microprocessor andmemory ICs, and networks before various types of mobile Internet content and applicationswere technically and economically feasible xiv. Chapters 2 and 3 (Part I) of this book address the potential of new technologies using theconcept of a “technology paradigm.” Primarily advanced by Giovanni Dosi xv, few scholars orpractitioners have attempted to use a technology paradigm to assess the potential of newtechnologies or to compare different ones xvi. One key aspect of a technology paradigm isgeometrical scaling, which is a little known concept that was initially noticed in the chemicalindustries (and in living organisms) xvii. Part I shows how a technology paradigm can help usbetter understand the potential for new technologies where technologies with a potential forlarge improvements in cost and performance often lead to the rise of new industries. Part Iand the rest of this book also show how implementing a technology and exploiting the fullpotential of its technology paradigm require advances in science and improvements incomponents. One reason for using the term “component” is to distinguish between components andsystems in what can be called a “nested hierarchy of subsystems xviii.” Systems are composed 7
  8. 8. of sub-systems, sub-systems are composed of components, and components may becomposed of various inputs including equipment and raw materials. This book will just usethe terms systems and components to simplify the discussion. For example, a system forproducing integrated circuits (ICs) is composed of components such as raw materials andsemiconductor manufacturing equipment.1.2 Technological Discontinuities and a Technology Paradigm A technology paradigm can be defined at any level in a nested hierarchy of subsystemswhere we are primarily interested in large changes in technologies or what many calltechnological discontinuities. Technological discontinuities are products that are based on adifferent set of concepts and/or architectures than are existing products and they are oftendefined as the start of new industries xix . For example, the first mainframe computers,magnetic tape-based playback equipment, and transistors (as were new services such asautomated algorithmic trading and online universities) were based on a different set ofconcepts than were their predecessors of punch card equipment, phonograph records, andvacuum tubes respectively. On the other hand, mini-, personal, and various forms of portablecomputers only involved changes in the architectures. Building from Giovanni Dosi’s characterization of them and using an analysis of manytechnologies (See Appendix for methodology), Chapter 2 and the rest of this bookcharacterize a technology paradigm in terms of: 1) a technology’s basic concepts or principlesand the tradeoffs that are defined by these concepts or principles; 2) the directions of advancewithin these tradeoffs where these advances are defined by a technological trajectory(s) xx; 3)the potential limits to these trajectories and their paradigms; and 4) the roles of componentsand scientific knowledge in these limits xxi . Partly since this book is concerned withunderstanding when a new technology might offer a superior value proposition, Chapter 2focuses on the second and third items and shows how there are four broad methods of 8
  9. 9. achieving advances in performance and cost along technological trajectories: 1) improvingthe efficiency by which basic concepts and their underlying physical phenomena areexploited; 2) radical new processes; 3) geometric scaling; and 4) improvements in “key”components. In doing so, Chapter 2 shows how improvements in performance and/or price occur in arather smooth and incremental manner over multiple generations of discontinuities. Whilesome argue that these improvements can be represented by a series of S-curves where eachdiscontinuity initially leads to dramatic improvements in performance and price xxii, Chapter 2and the rest of the book shows that such dramatic changes in the rates of improvements arerelatively rare. Instead, this book’s analyses suggest that there are smooth rates ofimprovements that can be characterized as incremental in nature over multiple generations oftechnologies and that these incremental improvements in a technological trajectory enableone to roughly understand near-term trends in performance and/or price/cost for newtechnologies.1.3 Geometrical Scaling Chapter 3 focuses on geometric scaling as a method of achieving advances in theperformance and cost of a technology. Geometric scaling refers to the relationship betweenthe geometry of a technology, the scale of it, and the physical laws that govern it. Or as othersdescribe it: the “scale effects are permanently embedded in the geometry and the physicalnature of the world in which we live xxiii.” As a result of geometric scaling, some technologies benefit from either large increases(e.g., engines or wind turbines) or large reductions (ICs) in physical scale. When technologiesbenefit from increases in scale, the output is roughly proportional to one dimension (e.g.,length cubed or volume) more than is the costs (e.g., length squared or area) thus causingoutput to rise faster than the costs, as the scale of the technology is increased. For example, 9
  10. 10. consider the pipes and reaction vessels that make up chemical plants. While economies ofscale generally refers to amortizing a fixed cost over a large volume at least until the capacityof a plant is reached, geometrical scaling refers to the fact that the costs of pipes (surface areaof a cylinder) vary as a function of radius whereas the output from pipes (volume of flow)vary as function of radius squared. Similarly, the costs of reaction vessels vary as a functionof surface area (radius squared) whereas the output of reaction vessels vary as a function ofvolume (radius cubed). This is why empirical analyses have found that the costs of theseplants only rise about two-thirds for each doubling of output and thus increases in the scale ofchemical plants have led to dramatic reductions in the cost of many chemicals xxiv. Other technologies benefit from reductions in scale. The most well-known examples ofthis type of geometrical scaling can be found in ICs, magnetic disks and tape, and opticaldisks where reducing the scale of transistors and storage regions has led to enormousimprovements in the cost and performance of these technologies xxv. This is because for thesetechnologies, reductions in scale lead to improvements in both performance and cost. Forexample, placing more transistors or magnetic or optical storage regions in a certain areaincreases the speed and functionality and reduces both the power consumption and size of thefinal product, which are typically considered improvements in performance for mostelectronic products; they also lead to lower material, equipment, and transportation costs. Thecombination of both increased performance and reduced costs as size is reduced has led toexponential changes in the performance to cost ratio of many electronic components. Like Chapter 2, Chapter 3 and other chapters also show how geometrical scaling is relatedto a nested hierarchy of subsystems. It shows that benefiting from geometrical scaling in ahigher level “system” depends on improvements in lower-level supporting “components xxvi,”and large benefits from geometrical scaling in a lower level “key component” can drivelong-term improvements in the performance and cost of a higher level “system.” In thesecond instance, these long-term improvements in the cost and performance of components 10
  11. 11. may lead to the emergence of technological discontinuities in systems, particularly when thesystems do not benefit from increases in scale. Part II shows how exponential improvementsin ICs led to discontinuities in computer, magnetic recording and playback equipment, andsemiconductors as does Chapter 9 for other systems. In fact, most of the disruptive innovations covered by Clayton Christensen, who manyconsider to be the guru of innovation xxvii, benefit from geometrical scaling (and experienceexponential improvements) in either the “system” or a key “component” in the system. Thissuggests that there is a “supply-side” aspect to Christensen’s theory of disruptive innovationthat is very different from his focus on the demand-side of technological change. While histheory suggests to some that large improvements in performance and costs along atechnological trajectory naturally emerge once a product finds a low-end niche and thusfinding the low-end niche is the central challenge of creating disruptive innovations xxviii,Chapter 3 and several chapters in Part II show how geometrical scaling explains why somelow-end innovations became disruptive innovations and why these low-end technologicaldiscontinuities initially emerged. Thus, a search for potentially disruptive technologies shouldconsider the extent to which a system or a key component in the system can benefit fromrapid rates of improvement through for example geometric scaling.. Some readers may find the emphasis on supply-side factors in Chapters 2 and 3 (Part I) tobe excessive and thus classify the author as a believer in so-called technology determinism.Nothing could be further from the truth. I recognize that there is an interaction betweenmarket needs and product designs, increases in demand encourage investment in R&D, andthe technologies covered in this book were “socially constructed xxix.” The relevance of thissocial construction is partly reflected in the role of new users in many of the technologicaldiscontinuities covered in Part II, where these new users and changes in user needs can leadto the rise of new industries xxx. For example, the emergence of industries represented bymicrobreweries and artisan cheese are more the result of changes in consumer taste than 11
  12. 12. changes in technology. Some of these changes in consumer taste come from rising incomesthat have led to the emergence of many industries serving the rich or even super rich. Whenthe upper 1% of Americans receives 25% of total income, many industries that cater tospecialized consumer tastes will emerge xxxi. This book focuses on supply-side factors because industries that have the potential tosignificantly enhance most lives or improve overall productivity require dramaticimprovements in performance and cost. As Paul Nightingale says in a special issue onGiovanni Dosi’s theory of technology paradigms, where he draws on the research of NathanRosenberg and David Mowery xxxii, “Market pull” theories are misleading, not because theyassume innovation processes respond to market forces, but because they assume that theresponse is unmediated. As a consequence, they cannot explain why so many innovations arenot forthcoming despite huge demand, nor why innovations occur at particular moments intime, and in particular forms xxxiii.” For example, the world needs inexpensive solar, wind, andother sources of clean energy, and large subsidies are increasing demand and R&D spendingfor them. But even with these large subsidies, large improvements in cost and performancewill not be forthcoming if these technologies do not have the potential for dramaticreductions in cost. And if they don’t have such a potential, the world needs to look for othersolutions. A second reason for focusing on supply side factors is that unless we understand thetechnological trajectories and the factors that directly impact on them such as scaling, howcan we accelerate the rates of improvement in cost and performance? Since much of themanagement literature on learning primarily focuses on the organizational processes that areinvolved with learning, this literature implies that organizational issues have a bigger impacton the potential for improving costs and performance than does the characteristics of thetechnology xxxiv. Thus, while the management literature on learning implies that solvingenergy and environmental problems is primarily an organizational issue, geometrical scaling 12
  13. 13. and the other three methods of achieving advances in performance and cost remind us that thepotential for improving the cost and performance of a technology depends on thecharacteristics of the technologyxxxv. Without a potential for improvements, it would bedifficult for organizational learning to have a large impact on the costs and performance of atechnology no matter how innovative the organization is.1.4 The Timing of Technological Discontinuities Chapters 4 through 6 (Part II) analyze technological discontinuities, partly becausediscontinuities often form the basis for new industries. For example, the first mainframecomputers, mini- computers, personal computers, personal digital assistants, audio cassetteplayers, video cassette recorders, camcorders, memory ICs, microprocessors, automatedalgorithmic trading, and online education are typically defined as technologicaldiscontinuities that formed the basis of new industries. Like other discontinuities, they werebased on a different set of concepts and/or architectures than were existing products. Thecharacterization of a system’s architecture is also considered important because the ability tocharacterize a system’s concept partly depends on one’s ability to characterize a system’spotential architectures. But what determines the timing of these discontinuities? Since the characterization of aconcept or architecture and an understanding of the relevant scientific phenomenon usuallyprecede the commercialization of a technology, we can look at the timing of technologicaldiscontinuities in relation to them. How long before the emergence of technologicaldiscontinuities were the necessary concepts and/or architectures characterized Second, why isthere a time lag, and in many cases, why is there a long time lag between a characterization ofthese concepts and architectures and both the commercialization and diffusion of thetechnology xxxvi? These questions are largely ignored by academic researchers. While there is wide 13
  14. 14. agreement on the descriptions and timing of specific technological discontinuities, mostresearch on technological discontinuities focuses on the existence and reasons for incumbentfailure and in doing so mostly treats these discontinuities as “bolts of lightning.” For example,the product life cycle, cyclical and disruptive models of technological change do not addressthe sources of technological discontinuities and instead their emphasis on incumbent failureimplies that any time lag is due to management failure such as cognitive ones xxxvii. But do we really believe that management failure for either cognitive or organizationalreasons is why it took more than 100 years to implement Charles Babbage’s computingmachine in spite of early government funding xxxviii? Although Charles Babbage defined thebasic concept for the computer in the 1820s and subsequently built a prototype, generalpurpose computers did not emerge until the 1940s or diffuse widely in developed countriesuntil the 1980s. Is this time lag merely due to narrow-minded managers and policy makers, oris something else going on? More importantly, in combination with a theory thattechnological discontinuities initially experience dramatic improvements in performance andprice, an emphasis on incumbent failure as the main reason for a long time lag suggests thatthere are many technological discontinuities with a potential for dramatic improvements inperformance and price just waiting to be found. According to this logic, if only managers andpolicy makers could overcome their cognitive limitations, firms and governments could findtechnologies that could quickly replace existing ones and thus solve global problems such asglobal warming. This book disagrees with such an assessment and shows how the timing of discontinuitiescan be analyzed. Building from research done by Nathan Rosenberg and his colleagues on therole of complementary technologies in the implementation of new technologies xxxix, Part IIshows how insufficient components were the reason for the time lag between theidentification and characterization of concepts and architectures that form the basis oftechnological discontinuities and the commercialization (and diffusion) of the 14
  15. 15. discontinuities xl. Chapters 4 through 6 present a detailed analysis of the discontinuities incomputers, magnetic recording and playback equipment, and semiconductors respectively.One reason for choosing these “systems” is because few argue there were market failures fordiscontinuities in them, unlike those of more “complex network” systems such asbroadcasting or mobile phones that are addressed in Part III xli. A second reason is that therehave been many discontinuities in these and related systems and thus there are a lot of “datapoints” to analyze xlii. Third, the time lag for each discontinuity in these systems the wasprimarily due to one or two types of insufficient components, which is very different from themechanical sector where novel combinations of components have played a more importantrole than have improvements in one or two components xliii. Partly because it possible todesign many of these systems in a modular way xliv, the performance of systems addressed inPart II were primarily driven by improvements in “key” components (which is the fourthbroad method of achieving advances in the performance and cost of a system) andimprovements in key components also drove the emergence of discontinuities in the systems. For example, the implementation of mini, personal, and most forms of portable computersprimarily depended on improvements in one type of component, ICs, as the discontinuitieswere all based on concepts and architectures that had been characterized by the late 1940s xlv.Similarly, the implementation of various discontinuities in magnetic-based audio and videorecording equipment primarily depended on improvements in one type of component, themagnetic recording density of tape, as these discontinuities were all based on concepts andarchitectures that had been characterized by the late 1950s. In other words, in spite of theincreasing variety of components that can be combined in many different ways,improvements in a single type of component had a larger impact on the emergence of thesediscontinuities (and on the performance of these systems) than did so-called novelcombinations of multiple components (or technologies). This conclusion enables us to gobeyond the role of complementary technologies in the time lag and analyze the specific levels 15
  16. 16. of performance that were needed in single types of components before new systems, i.e.,discontinuities, could be implemented.1.5 Systems, Components and Discontinuities Chapters 4, 5, and 6 explore the relationship between improvements in single types ofcomponents and the emergence of discontinuities in systems in two ways, where both ofthese ways are facilitated by the smooth and incremental manner in which improvements inperformance and/or price have been occurring. First, building from the role of tradeoffs intechnology paradigms and marketing theory xlvi, these chapters show how improvements incomponents have changed the tradeoffs that suppliers and users make when they considersystems and how this leads to the emergence of discontinuities. Technology paradigms definea set of tradeoffs between price and various dimensions of performance and designersconsider these tradeoffs when they design or compare systems while users make tradeoffsbetween price and various dimensions of performance. In both cases, improvements incomponents can change the way these tradeoffs are made by both designers and users.Second, economists use the term “minimum threshold of performance” to refer to theperformance that is necessary before users will consider purchasing a system xlvii. For example,users would not purchase a PC until the PC could perform a certain number of instructionsper second. When a single type of component such as a microprocessor has a large impact onthe performance of a system such as a PC, a similar threshold exists for the components inthese systems. For example, PCs could not perform a certain number of instructions persecond until a microprocessor could meet certain levels of performance. Part II draws a number of conclusions from these analyses. First, the new concepts orarchitectures that form the basis of discontinuities in systems were known long before thediscontinuities were implemented. In other words, the characterization of concepts orarchitectures was usually not the bottleneck for the discontinuities and thus for the creation of 16
  17. 17. the industries that many of these discontinuities represent. Instead, the bottleneck was in oneor two types of components that were needed to implement the discontinuities. Thus,improvements in components can gradually make new types of systems, i.e., discontinuities,possible and the thresholds of performance (and price) that are needed in specific componentsbefore a new system is economically feasible can be analyzed. Second, finding new customers and applications, which partly reflect heterogeneity incustomer needs xlviii, can reduce the minimum thresholds of performance for the componentsthat are needed to implement discontinuities. Chapters 4, 5, and 6 provide many examples ofhow new customers and applications (and also methods of value capture) enableddiscontinuities to be successfully introduced before the discontinuities provided the levels ofperformance and/or price that the previous technology did. In other words, these newcustomers, applications, and method of value capture reduced the minimum thresholds ofperformance for these systems and their key components. However, although this wasimportant from the standpoint of competition between firms, the impact of these newcustomers, applications, and methods of value capture (and the heterogeneity in customerneeds that they reflect) on these thresholds were fairly small when compared to the manyorders of magnitude in system performance that came from improvements in componentperformance. Third, one reason that discontinuities emerged in computers and in magnetic recordingand playback equipment is because they did not benefit from geometric scaling to the extentthat their components did. ICs and magnetic recording density experienced exponentialimprovements in cost and performance because they benefited from dramatic reductions inscale, i.e., geometric scaling. However, since computers and magnetic recording systems donot benefit much from geometric scaling (in some cases they exhibit diseconomies of scale),it was natural that smaller versions emerged and replaced the larger versions. Fourth, the demand for many of these improvements in components was initially driven 17
  18. 18. by other systems and/or industries. This enabled many new systems/industries to get a “freeride” on existing industries as improvements in components “spilled over” and made newindustries possible. This provides additional evidence that the notion of cumulativeproduction driving cost reductions is misleading and impractical, a point that others such asWilliam Nordhaus have made using different forms of analysis xlix. Not only is it by definitionimpossible for learning curves to help us understand when a potential discontinuity (one notyet produced) might provide a superior value proposition, Part II shows how improvements incomponents (e.g., ICs) gradually made new discontinuities economically feasible where thedemand for these components was coming from other industries.1.6 Challenges for Firms and Governments Chapters 7 and 8 of Part III address a different set of questions, ones that concern thechallenges for firms and governments with respect to new industries. While Parts I and IIfocus on when a discontinuity might become economically feasible and thus imply that firmseasily introduce and users easily adopt new technologies, Chapters 7 and 8 summarize thecomplexities of new industry formation and thus the challenges for firms and governments.These complexities may cause the diffusion of new technologies to be delayed or they mayenable new entrants or even new countries to dominate an industry whose old version waspreviously dominated by other countries. Chapter 7 focuses on competition between firms. Incumbents often fail when technologicaldiscontinuities emerge and diffuse, particularly when these discontinuities destroy anincumbent’s capabilities l. New technologies can destroy a firm’s capabilities in many areasincluding R&D, manufacturing, marketing, and sales where the destruction of the capabilitiesmay be associated with the emergence of new customers. For example, Clayton Christensenargues that incumbents often fail when a low-end innovation displaces the dominanttechnology (thus becoming a disruptive innovation) largely because the low-end innovation 18
  19. 19. initially involves new customers and serving these new customers requires new capabilities li.Helping firms analyze the timing of technological discontinuities, which is the subject of PartII, can help firms identify and prepare for discontinuities through for example identifying theappropriate customers and creating the relevant new capabilities to serve the new customers. Other research has found that the total number of firms in an industry declines quicklyfollowing the emergence of a technological discontinuity in some industries or sectors morethan in others where the number of firms is a surrogate for the number of opportunities lii. Thisdecline occurs through mergers, acquisitions, and exits, in what many call a “shakeout” in thenumber of firms. The occurrence of such a shakeout depends on whether large firms haveadvantages over smaller firms through economies of scale in operations, sales, and/or R&D.For example, economies of scale in R&D (or other activities) favor firms with a large amountof sales in a new industry because they can spend more on total R&D than can firms withfewer sales. Initially, their greater spending on R&D leads to more products, their moreproducts leads to more sales, and thus positive feedback leads to larger firms dominating anindustry where the smaller firms are acquired or exist the industryliii. Chapter 7 focuses on these issues in more detail and on how two factors, the number ofsubmarkets and the emergence of vertical disintegration, impact on the importance ofeconomies of scale, a shakeout in the number of firms and thus the number of opportunitiesfor new entrants. The existence of submarkets can reduce the extent of economies of scale inR&D when each submarket requires different types of R&D and thus the existence ofsubmarkets can prevent the emergence of a shakeout. This enables a larger number of firms,including entrepreneurial startups, to exist in an industry or sector with many submarkets thanin one with few submarkets. Vertical integration enables the late entry of firms, sometimes long after a shakeout hasoccurred. Furthermore, since vertical disintegration can lead to a new division of labor in aneconomy in which there is a set of new firms providing new types of products and services, 19
  20. 20. vertical disintegration can also lead to the rise of new industries. While Chapter 7 primarilyfocuses on the emergence of high-technology industries such as computer software,peripheral, and services and semiconductor foundries and design house, verticaldisintegration has also led to the formation of less high-tech, albeit large industries such asjanitorial, credit collection, and training services liv. Chapter 8 focuses on how the challenges for firms and governments vary by type ofindustry using a typology of industry formation. While these industries might emerge fromeither vertical disintegration or technological discontinuities, most of the examples are forthose that emerged from discontinuities. The typology focuses on system complexity andwhether a critical mass of users or complementary products is needed for growth to occur.Although the formation of most new industries depends on when a new technology becomeseconomically feasible and thus provides a superior “value proposition” to an increasingnumber of users, industries represented by complex systems and/or that require a criticalmass of users/complementary products for growth to occur face additional challenges lvwhere these challenges may delay industry formation. Meeting these challenges might requireagreements on standards, new methods of value capture and industry organization,government support for R&D, government purchases, new or modified regulations, newlicenses, or even new ways of awarding licenses.1.7 Thinking about the Future Chapters 9 and 10 of Part IV use the conclusions from previous chapters to analyze thepresent and future of selected technologies. Chapter 9 looks at a broad number ofelectronics-related technologies such as displays, wireline and mobile phonetelecommunication systems, the Internet and on-line services (including financial andeducational ones), and human-computer interfaces. Building from the notion of a technologyparadigm, it shows how improvements in specific components such as ICs have enabled new 20
  21. 21. system-based discontinuities to become technically and economically feasible. Moreimportantly, it shows how one can use an understanding of the technological trajectories in asystem or a key component of such a system to analyze the timing of new discontinuitiessuch as three dimensional displays, cognitive radio in mobile phone systems, cloud/utilitycomputing for the Internet, and gesture and neural-based human-computer interfaces. Chapter 10 looks at three types of clean energy and how the four broad methods ofachieving advances in performance and cost can help us better analyze the potential forimprovements in wind turbines, solar cells, and electric vehicles, and thus can provide betterguidance on appropriate policies than can the typical emphasis on cumulative production. Anemphasis on cumulative production says that the costs of clean energy fall as more windturbines, solar cells, and electric vehicles are produced, that this “learning” primarily occurswithin the final product’s factory setting as automated equipment is introduced and organizedinto flow lines, that the extent of this learning depends on organizational factors, and thatdemand-based incentives are the best way to achieve this learning. Governments haveresponded to this emphasis on cumulative production by implementing demand-basedsubsidies and firms have responded to these demand-based subsidies by focusing on theproduction of existing technologies such as existing wind turbine designs, crystallinesilicon-based solar cells, and hybrid vehicles with existing lithium-ion batteries. However, applying the four broad methods of achieving advances in performance andcost - notably improvements in efficiency, geometric scaling lvi, and key components - toclean energy lead to a different set of conclusions about policies where these policies involvethe development of newer technologies and ones that appear to have more potential forimprovements than the ones being currently emphasized. For wind turbines, the key issue isgeometrical scaling. Chapter 8 describes how costs per output have fallen as the physicallength of the turbine blades and towers have been increased where increases in scale requirestronger and lighter materials. Thus, government policies should probably focus on the 21
  22. 22. development of these materials through supply-based incentives such as R&D tax credits ordirect funding of research on new forms of materials. Furthermore, some evidence suggeststhat the limits to scaling have been reached with the existing wind turbine design, particularlyusing existing materials, and thus new designs are needed. Again, supply-based incentivessuch as R&D tax credits or direct funding of new forms of wind turbine designs will probablyencourage manufacturers to develop new designs than will demand-based subsidies. For solar cells, improvements in them come from a combination of increases in efficiencyand reductions in cost per area where the latter is primarily driven by both reductions in thethicknesses of material and increases in the scale of production equipment (both are forms ofgeometrical scaling). The largest opportunities for these improvements are in new forms ofsolar cell designs such as thin-film ones that are already cheaper on a cost per peak Watt basisthan are crystalline silicon ones. Unfortunately, crystalline silicon ones are manufactured farmore than are thin-film ones because turnkey factories are more available for crystallinesilicon than thin film ones and thus firms can more easily obtain demand-based subsidies forthe former than the latter ones. Therefore, like wind turbines, governments should probablyfocus more on supply-based incentives such as R&D tax credits or direct funding of newforms of solar cells to realize the necessary improvements in efficiency and reductions inmaterial thicknesses that appear possible with thin-film solar cells. For electric vehicles, the key component is an energy storage device (e.g., battery) and thusappropriate policies should focus on this device and not the electric vehicle. Chapter 10describes how improvements in lithium-ion batteries, which currently receive the mostemphasis by vehicle manufacturers, are proceeding at a very slow pace and that largeimprovements are not expected to emerge in spite of the fact that large improvements areneeded before unsubsidized electric vehicles become economically feasible. Therefore, inorder to encourage firms to look at new forms of batteries (or other forms of energy storagedevices such as capacitors lvii or compressed air), governments should probably focus on 22
  23. 23. supply-based incentives such as R&D tax credits or direct funding of new forms of energystorage devices.1.8 Who is this book for? This book is for anyone interested in new industries and in the process of their formation.This includes R&D managers, hi-tech marketing and business development managers, policymakers and analysts, professors, and employees of think tanks, governments, hi-tech firms,and universities. This book helps firms better understand when they should fund R&D orintroduce new products that can be defined as a new industry. It helps policy makers andanalysts think about whether technologies have a large potential for improvement and howgovernments can promote the formation of industries that are based on this technology. It alsohelps these people find those technologies that have a potential for large improvements andthus a potential to become new industries, which is much more important than devising thecorrect policies for a given technology. This book is particularly relevant for technologies in which the rates of improvements inperformance and cost are large and thus the frequency of discontinuities is high. For firmsinvolved with these kinds of technologies, understanding when technological discontinuitiesmight emerge is a key issue. This is because technological discontinuities often lead tochanges in market shares and sometimes lead to incumbent failure. They may even lead tochanges in shares at the country level; for example, the emergence of technologicaldiscontinuities have impacted on the rising (and falling) shares of U.S., Japanese, Korean andTaiwanese firms in the electronics industries in the second half of the 20th century. This bookcan help firms, universities, and governments better understand when these discontinuitiesmight emerge and thus the bridge the gaps between advances in our understanding ofscientific phenomenon, the characterizations of concepts and architectures, and thecommercialization of technological discontinuities. On one hand, scientists such as Michio 23
  24. 24. Kaku (Physics of the Future) lviii discuss the scientific and technical feasibility of differenttechnologies. On the other hand, business professors discuss the strategic aspects of newtechnology in terms of for example a business model lix. This book helps one understand whenscientifically and technically feasible technologies might become economically feasible andthus when firms, universities, and governments should begin developing business models andappropriate policies for them. This book is also for young people. Young people have more at stake in the future thananyone else and this book is written to help people think about their future. It helps studentsthink about where opportunities may emerge and thus the technologies they should study andthe industries where they should begin their careers. In terms of opportunities, while theconventional wisdom is to focus students on customer needs or on what is scientifically ortechnically feasible, it is also important to help students understand those technologies thatare undergoing improvements and how these improvements are creating opportunities inhigher-level systems, something which even few engineering classes do partly because theyfocus heavily on mathematics (and are criticized for this) lx. For example, helping students(and firms and governments) understand how reductions in the features sizes of ICs,including bio-electronic ones and MEMS (micro-electronic mechanical systems), can helpstudents search for new opportunities. My students have used such information to analyze 3Dholograms, 3D displays, MEMS (micro-electronic mechanical systems) for ink jet printing,3D printing, different types of solar cells and wind turbines, cognitive radio, and new formsof human-computer interfaces (e.g., voice, gesture, neural), including the opportunities thatare emerging from these technologies lxi; some of these presentations are a source of data forChapter 9. Among other things, the final chapter discusses how this book can be used inuniversities courses to help students think about and analyze the future. Furthermore, the ideas discussed in this book can helps students and other young peoplelook for solutions to global problems that will not be easily found. Without an understanding 24
  25. 25. of technology change, how can we expect students to propose and analyze reasonablesolutions? To put it bluntly, discussions of policies, business models, and socialentrepreneurship are necessary but insufficient. New technologies and improvements inexisting ones provide tools that our world can use to address global problems and thusproposed solutions should consider the potential for and rate of improvements in technologies.For example, Chapter 10 uses this book’s ideas to analyze three types of clean energy andconcludes that the potential for improvements in them is mixed and thus more radicalsolutions are probably necessary. We need to ask students the right questions and give themthe proper tools so that they can do this type of analysis and propose more radical solutions.i The U.S. government expects to spend $150 billion between 2009 and 2019 on clean energy of which less than $5billion is expected toinvolve research and development of solar cells and wind turbines. Presentation by Dan Arvizu at National University of Singapore,November 3, 2010, Moving Toward a Clean Energy Future.ii Analyses of costs using cumulative production can be found for a variety of industries in (Arrow, 1962; Ayres, 1992; Huber, 1991; Argoteand Epple, 1990; March, 1991). For clean energy, these analyses can be found in (Nemet, 2006; Nemet, 2009). The notion that cumulativeproduction is the primary driver of cost reductions is also implicit to some extent in theories of technological change (Abernathy andUtterback, 1978; Utterback; 1994; Christensen, 1997; Adner and Levinthal, 2001). For example, Utterback (1994) and Adner and Levinthal(2001) focus on cost reductions through improvements in processes where the locus of innovation changes from products to processes andthus the locus of competition changes from performance to cost following the emergence of a dominant design. Although Christensen’s(1997) theory primarily focuses on the reasons for incumbent failure, his theory also emphasizes demand, how demand drives learning, andhow this demand leads to improvements in both cost and performance. More specifically, once a product finds an unexplored niche, anexpansion in demand leads to greater investment in R&D and thus improved performance and cost for the low-end product. Therefore, thekey to achieving improvements in performance and cost is to find these unexplored niches. An exception to these examples can be found in(Nordhous, 2009).iii One analysis of solar cells makes this point (Nemet, 2006).iv See Jason Pontin’s interview of Bill Gates in Technology Review, Q&A: Bill Gates, The cofounder of Microsoft talks energy,philanthropy and management style, August 24, 2010,, accessed on August 26,2010. See also Ball, 2010v (Schmookler, 1966)vi For example, less than 5% of the iPhone 3GS’s manufacturing cost in 2009 consisted of assembly costs and the majority of the 25
  26. 26. component costs were standard ICs whose costs depended more on advances in Moore’s Law than they did cumulative production of theiPhone. Last accessed on September 29, 2011.vii Increases in the number of transistors per chip, better known as Moore’s Law, are always presented as a function of time and notcumulative production.viii Kurzweil, 2005ix This book’s distinction between science and technology and of the linear module of innovation is roughly consistent with Arthur’s (2007,2009) characterization. He distinguishes between: 1) an understanding of a scientific phenomenon; 2) the definition of a concept orprinciple; and 3) solving problems and sub-problems in a recursive manner. For a broader discussion of the linear model, see (Balconi et al,2010)x See (Albright, 2002)xi Exceptions include: (Rosenberg, 1963, 1969; Freeman and Soete, 1997; Mowery and Rosenberg, 1998; Freeman and Louca, 2001)xii For example, the timing of the industrial revolution differed by decades if not centuries between countries, even among European onesand the timing of industry formation for the initial banking, insurance, and finance industries may have differed by even larger time spans.xiii For example, see (Klepper, 2007, 2010).xiv The mobile Internet is explored in more detail in (Funk, 2004, 2006, 2007a, 2007b).xv Dosi’s technology paradigm builds from Thomas Kuhn’s (1970) notion of a paradigm shift.xvi While there are many good descriptions of how technologies change, for example, see Arthur’s (2009) and Constant’s (1980)descriptions of jet engines and Hughes’ (1983) description of electricity, the potential for improvements in competing technologies is rarelyaddressed.xvii Many different terms are used by scholars. Nelson and Winter (1982) initially used the term economies of scale but laterWinter (2008) used the term scaling heuristics. Sahal (1985) used the term scaling while Lipsey, Carlaw and Clifford Bekar (2005) use both geometrical scaling and increasing returns to scale. For organisms, see Schmidt-Nielsen, 1984.xviii See for example, (Simon, 1962; Alexander, 1964; Tushman and Rosenkopf, 1992; Tushman and Murmann, 1998; Malerba, et al, 1999)xix For example, see (Abernathy and Clark, 1985; Tushman and Anderson, 1986; Utterback, 1994; Henderson and Clark, 1990). Whiletechnological discontinuities are defined in terms of differences with previous products, dominant designs are defined by the degreesimilarity among existing products (e.g., architectures and components) (Murmann and Frenken, 2006). Technological discontinuities andnew industries can be thought of as the second stage of Schumpeter’s three stage process of industry formation: 1) invention; 2) innovation;and 3) diffusion.xx Many scholars have emphasized directions of advance; these include Rosenberg (1969),who used the term focusing devices, Sahal(1985), and Vincenti (1994).xxi In order, these elements are similar to Dosi’s emphasis on a “specific body of understanding,” a “definition of the relevant problems tobe addressed and the patterns of enquiry in order to address them,” a “specific body of practice,” and “the operative constraints on prevailingbest practices and the problem-solving heuristics deemed promising for pushing back those constraints.” (Dosi and Nelson, 2010)xxii Foster (1986) focused on S-curves and Tushman and Anderson (1986) focused on dramatic rates of improvements, which they describeusing the term punctuated equilibrium. They borrowed this term from the field of biology where the theory of punctuated equilibrium saysthat most sexually reproducing species exhibit little evolutionary change except in rapid and localized cases (Gould and Eldredge, 1977).They concluded that technologies also undergo dramatic improvements following their introduction by looking at the speed ofmini-computers, seat-miles per year capacity of aircraft, and size of cement plants. Others (Kurzweil, 2005; Koh and Magee, 2006; Koomey 26
  27. 27. et al, 2011) have shown that new computers did not experience dramatic improvements in performance following their introduction while Iargue that large increases in seat miles and plant size are merely an artifact of infrequent introductions of large aircraft and largermanufacturing plants. Furthermore, neither of these measures of performance are relevant unless one discusses scaling, which Tushman andAnderson do not. Koh and Magee (2008) explicitly deny the existence of punctuated equilibrium in their analysis of energy storagetechnologies.xxiii (Lipsey et al, 2005). Geometric scaling is also different from network effects (Arthur, 1994; Shapiro and Varian, 1999) and increasingreturns to R&D (Klepper, 1996, Romer, 1986).xxiv (Haldi and Whitcomb, 1967; Levin, 1977; Freeman and Louca, 2001; Winter, 2008). Rosenberg (1994, p. 198) estimates the increasesin capital costs with each doubling to be 60%.xxv See for example, (Sahal, 1985; Lipsey et al, 2005; Winter, 2008).xxvi The first instance extends Richard Lipsey’s notion that the “ability to exploit [geometric scaling] is dependent on the existing state oftechnology.”xxvii For example, the Economist devoted at least five articles to him and his ideas in 2010 and 2011. However, analyses by other scholarssuggest that Christensen’s analysis may have exaggerated the challenges of disruptive innovations for incumbents (McKendrick, 2000; Kingand Tucci, 2002)xxviii Even Christensen’s newest book (Dyer, Gregersen and Christsensen, 2011) implies these things by focusing solely on the skillsneeded for creating a low-end innovation and ignoring the improvements in performance and price that are needed for a low-end innovationto displace the dominant technology and thus become a disruptive innovation.xxix See Bijiker et al (1989) for discussion of social construction of technologies, Schmookler (1966) for an analysis of R&D and demand,and others for analysis of the interaction between market needs and product designs (Clark, 1985: Vincenti, 1994; Arthur, 2009).xxx One example of a change in user needs can be found in (Trispas, 2008). New or existing users might also be the source of innovations(von Hippel, 1986).xxxi Lambert, R. Its camp is gone but the occupy movement will grow. Financial Times, November 15, 2011.xxxii Mowery and Rosenberg, 1998; Rosenberg: 1982, 1994)xxxiii (Nightingale, 2008). (Teece, 2008) and others make similar arguments. For example, Freeman (1994) concludes that “the majority ofinnovation characterized as ‘demand led’ were actually relatively minor innovations along established trajectories” and Walsh (1984) andFleck (1988) claim that supply-side factors drove innovation during the early stage of innovation in synthetic material, drugs, dyestuff, androbotics.xxxiv (Arrow, 1962; Huber, 1991; March, 1991)xxxv (Gold, 1981)xxxvi Although Arthur (2007) is one of the few to consider this time lag, others have considered the time lag between advances in scienceand the commercialization of the technology that is based on this science. For example, (Klevorick et al, 1995; Kline and Rosenberg, 1986;Mansfield, 1991).xxxvii Kaplan and Tripsas (2008) argue and Dyer et al (2011) suggest that cognitive bias is the main reason for any delay while otherslargely ignore the issue (Anderson and Tushman, 1990; Utterback, 1994; Christenson and Bower, 1996; Christensen, 1997; Klepper, 1997;Kaplan and Tripsas, 2008). Exceptions include Levinthal (1998), who uses the notion of speciation to describe the “emergence” (Adner andLevinthal, 2002) of new technologies and Windrum (2005), who focuses on heterogeneity.xxxviii Although some (Gleick, 2011) argue that a relay-based machine could have been constructed in the 19th century, this does not 27
  28. 28. invalidate my logic that components were the main reason for the time lag.xxxix Nathan Rosenberg and his colleagues (Rosenberg, 1963, 1969; Kline and Rosenberg, 1986; Mowery and Rosenberg, 1998)emphasizes the need for complementary technologies while others emphasize a novel combination of technologies (Basalla 1988; Ayres,1988; Iansiti, 1995; Fleming, 2001; Hargadon, 2003)xl I am making a distinction between an ability to analyze and an ability to predict or forecast.xli One exception is personal computers where Microsoft’s bundling of software is seen by some as anti-competitive. This is brieflymentioned in Chapters 7 and 8.xlii By related sectors, I refer to other types of magnetic storage and electronic systems.xliii Analyses of automobiles (Abernathy and Clark, 1985), machine tools, electrical generating equipment (Hughes, 1983), and aircraft(Constant, 1980; Vincenti, 1994; Arthur, 2009) suggest that novel combinations of components probably played a larger role indiscontinuities than did improvements in single types or components in the mechanical sector. On the other hand, there have been fewerdiscontinuities in the mechanical than for the electronics sector.xliv Modular design is a necessary but insufficient situation for a component to have a large impact on the performance and cost of asystem. For more on modular design, see (Langlois, 1992, 2003, 2007; Ulrich, 1995; Sanchez and Mahoney, 1996; Baldwin and Clark,2000)xlv Malerba et al (1999) make a similar argument but they focus on radical innovations in components while this book explains theemergence of discontinuities in terms of incremental innovations.xlvi The notion of tradeoffs is a fundamental property of indifference curves (Green and Wind, 1979), Christensen’s theory of disruptiveinnovation (Adner, 2002, 2004; Adner and Zemsky, 2005), and innovation frontiers (de Figueiredo and Kyle, 2006).xlvii (Green and Wind, 1973; Adner, 2002)xlviii Windrum (2005) explicitly uses heterogeneity to examine discontinuities while others (Levinthal, 1998; Adner and Levinthal, 2002)imply that heterogeneity is important..xlix Althouth Nordhaus (2009) makes the strongest argument about the problems with using the learning curve, others (Agarwal, Audretsch& Sarkar, 2007; Yang, Phelps, & Steensma, 2010) have also noted problems with learning curves.l (Afuah and Bahram, 1995; Anderson and Tushman, 1990; Tusman and Anderson; 1986; Utterback, 1994)li (Christensen, 1997). for an opposing viewpoint, see (King and Tucci, 2002; McKendrick, Haggard, and Doner, 2000)lii See for example (Gort and Klepper, 1982; Klepper and Grady, 1990; Agarwal and Gort, 1996; Klepper, 1997; Klepper & Simons, 1997;Tegarden et al, 1999)liii (Klepper, 1996; 1997)liv (Klepper, 1997; Klepper and Thompson, 2006). For vertical disintegration, also see (Jacobides, 2005; Jacobides and Winter, 2005;Cacciatori and Jacobides, 2005; Jacobides and Billinger, 2006),lv This analysis builds from the research of: (Rohlfs, 1974, 2001; Tushman and Rosenkopf, 1992)lvi Although Levitt and Dubner in their book Superfreakonomics (2009) also apply the concept of scaling to some solutions for globalwarming, they ignore the role of scaling in wind turbines, solar cells, and batteries.lvii Koh and Magee, 2008; Personal communication with Chris Magee, May 13, 2011lviii (Deutsch, 2011; Kaku,2011)lix One way to characterize a business model is in terms of value proposition, customer selection, method of value capture, scope ofactivities, and method of strategic control. 28
  29. 29. lx (Drew, 2011)lxi Slides for several chapters and slides from presentations by students in a course based on this book are available on slide share: Furthermore, these and other slides are discussed in my blog: 29