Policy Research Working Paper 10164 Massive Modularity: Understanding Industry Organization in the Digital Age The Case of Mobile Phone Handsets Eric Thun Daria Taglioni Timothy Sturgeon Mark P. Dallas Development Economics Development Research Group September 2022 Policy Research Working Paper 10164 Abstract Digitization is transforming the organization and geog- Consensus” era: 1) it develops a broader view of modular raphy of industries. Once digitized, information can be and platform ecosystems than has been advanced so far, generated, collected, stored, monitored, analyzed, and pro- highlighting the overlapping and layered nature of digital cessed in ways not previously possible, and when common industry ecosystems; 2) it focuses on the multiplicity of standards are used as modular interfaces, data can be trans- standards that bind modular ecosystems together; and 3) it ferred and put to use with greater ease across organizations draws attention to the geographic and geopolitical implica- and geographic space. An important effect of digitization tions of what it calls Massive Modular Ecosystems (MMEs). on industrial organization is the emergence of global-scale The case study of the mobile phone handset industry reveals modular ecosystems associated with specific classes of three paradoxes associated with MMEs: 1) they allow for products, applications, and technologies. The modules extremely complex products to be produced at scale, unlike and sub-systems in these ecosystems can—albeit with sig- more traditional industries; 2) they simultaneously feature nificant engineering effort, because they are complex—be high degrees of market concentration at the level of complex reused, connected, and layered to drive innovation and sub-systems and components, and market fragmentation deliver products and services with immense complexity at at the level of the industry overall and at the level of com- scale. The nuances of this transformation have not been plementors; and 3) they are concentrated in geographic lost on the field of technology management and innovation. clusters, but because MMEs integrate work carried out in The primary focus of this literature has been on how to many specialized clusters in many countries, the system as capture value in modular ecosystems, mainly by focusing a whole is geographically dispersed. This leads to a fourth, on how to companies can influence or leverage industry policy-related paradox: MMEs generate strategic and geo- architectures and “win” in an era of digital platforms. This political pressures for decoupling when placed under stress, paper makes three contributions to these literatures, as well but the same set of circumstances also creates pressures for as to literatures on global value chains (GVCs), industry maintaining the business relationships and institutions that standards, and industrial policy in the post- “Washington have come to underpin global integration. This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at Dtaglioni@worldbank.org, eric.thun@sbs.ox.ac.uk, sturgeon@mit.edu, and dallasm@union.edu. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Massive Modular Ecosystems Eric Thun, Oxford Saïd Business School Daria Taglioni, World Bank Development Research Group Timothy J. Sturgeon, MIT Industrial Performance Center Mark P. Dallas, Union College and Council on Foreign Relations Originally published in the Policy Research Working Paper Series on September 2022. This version is updated on November 2022. To obtain the originally published version, please email prwp@worldbank.org. JEL codes: L14 Transactional Relationships • Contracts and Reputation • Networks L22 Firm Organization and Market Structure L23 Organization of Production L25 Firm Performance: Size, Diversification, and Scope O14 Industrialization • Manufacturing and Service Industries • Choice of Technology O32 Management of Technological Innovation and R&D O33 Technological Change: Choices and Consequences • Diffusion Processes Keywords: Global Value Chains and Global Supply Chains, Firm Organization, Industrial Development, Firm-to-firm linkages, Digital development Acknowledgements: This work has been supported by a World Bank trust fund with the Republic of Korea, acting through the Korean Development Institute School of Public Policy and Management (KDIS) on the KDI School Partnership for Knowledge Creation and Sharing (TF0B0356). We would like to thank Jing-ming Shiu at National Cheng Kung University for his insights and for sharing his original data on telecommunication standards and Android Open-Source Project code commits. In addition, we wish to thank Irene Iodice, Ritika Khandelwal, Jose Marzluf and Isak Falk for their excellent research assistance. For very insightful comments on a draft manuscript, we thank Ari Van Assche, Carliss Baldwin, Judith Biewener, Florian Butollo, Devika Narayan and Lindsay Whitfield. Furthermore, for their useful feedback, we thank participants of the 2022 Society of Socio-economic Annual Conference Network O Session: Digitalization and Digital Platforms: Implications for GVCs, and the 2022 Comparative Political Economy of Global Business Research Conference at the Saïd Business School, University of Oxford. MASSIVE MODULAR INDUSTRIAL ECOSYSTEMS TABLE OF CONTENTS 1 INTRODUCTION ............................................................................................................................................................................ 1 2 THE ROLE OF MODULARITY IN COMPLEX GLOBAL INDUSTRIES ............................................................................... 4 MODULARITY IN FIRMS ............................................................................................................................................................................................ 5 MODULARITY IN (GLOBAL) VALUE CHAINS ........................................................................................................................................................... 6 MODULARITY IN INDUSTRIAL ECOSYSTEMS .......................................................................................................................................................... 7 STANDARDS: THE SOURCE OF MODULAR INTERFACES ........................................................................................................................................ 8 4 DEFINING MASSIVE MODULAR ECOSYSTEMS ................................................................................................................. 10 THREE PARADOXES IN MMES .............................................................................................................................................................................. 11 Complexity at scale .................................................................................................................................................................................... 11 Market concentration and fragmentation ...................................................................................................................................... 12 Geographic clustering and dispersion ............................................................................................................................................... 13 5 MOBILE PHONE HANDSETS – A MASSIVE MODULAR INDUSTRIAL ECOSYSTEM ............................................... 14 DATA ........................................................................................................................................................................................................................ 14 THE EVOLUTION OF HANDSET INDUSTRY ARCHITECTURE – FROM INTEGRATED TO MODULAR ................................................................. 15 COMPLEXITY AT SCALE .......................................................................................................................................................................................... 20 Scale ................................................................................................................................................................................................................. 20 Product complexity .................................................................................................................................................................................... 22 MARKET CONCENTRATION AND FRAGMENTATION .............................................................................................................................................. 24 Handsets ......................................................................................................................................................................................................... 24 Sub-systems ................................................................................................................................................................................................... 26 GEOGRAPHIC CLUSTERING AND DISPERSION......................................................................................................................................................... 28 Handsets ......................................................................................................................................................................................................... 28 Sub-systems ................................................................................................................................................................................................... 29 Telecom standard-setting ....................................................................................................................................................................... 31 5 THE POLICY PARADOX ............................................................................................................................................................ 33 6 CONCLUSIONS ............................................................................................................................................................................. 36 REFERENCES ................................................................................................................................................................................... 38 APPENDIX – DATA SOURCES ..................................................................................................................................................... 43 LIST OF TABLES TABLE 1. DISTRIBUTION OF POWER AND STANDARD-SETTING PROCESSES IN MODULAR ECOSYSTEMS ......................................................... 10 TABLE 2. MOBILE HANDSET PERFORMANCE AND FUNCTIONAL IMPROVEMENTS (2009-2020) .................................................................. 23 TABLE 3. APP RELEASES BY COUNTRY, 2017 .......................................................................................................................................................... 31 LIST OF FIGURES FIGURE 1. MODULARITY AS ENABLER OF COMPLEXITY AT SCALE............................................................................................................................ 5 FIGURE 2. GEOGRAPHIC IMPLICATIONS AND STRATEGIC OBJECTIVES FOR DIFFERENT FORMS OF INDUSTRY ORGANIZATION ..................... 14 FIGURE 3. LAYERED MODULAR ECOSYSTEMS IN THE MOBILE PHONE HANDSET MME AND LINKS TO ADJACENT INDUSTRIES ................... 19 FIGURE 4. MOBILE HANDSET SHIPMENTS AND AVAILABLE SMARTPHONE APPS, 2007-2020 ....................................................................... 21 FIGURE 5. MOBILE HANDSET OS MARKET SHARE, JANUARY 2012 – APRIL 2021 .......................................................................................... 22 FIGURE 6. “SMARTPHONE” SHIPMENTS BY BRAND AND GEOGRAPHY OF OWNERSHIP, 2007 – 2020 ........................................................... 25 FIGURE 7. RISING CONCENTRATION AT THE MOBILE HANDSET SUB-SYSTEM LEVEL, 2008-2019 ................................................................ 26 FIGURE 8. TOP MOBILE HANDSET EXPORT SHARES AND VALUES , 2007-2019, US$ BILLIONS ...................................................................... 28 FIGURE 9. MOBILE PHONE COMPONENT COST SHARE BY SUB-SYSTEM AND COUNTRY, 2010 AND 2019 ..................................................... 29 FIGURE 10. CONTRIBUTIONS (SOFTWARE CODE “COMMITS”) TO GOOGLE’S ANDROID OPEN-SOURCE PROJECT ........................................ 30 FIGURE 11. MOBILE SUBSCRIPTIONS BY NETWORK INTERCONNECT STANDARD, 2011-2021, MILLIONS ................................................... 32 FIGURE 12. PARTICIPATION IN MOBILE TELECOM INTERCONNECT STANDARD SETTING BY COUNTRY (3GPP), 2001-2020 ................. 33 FIGURE 13. THREE PARADOXES FOUND IN MME ASSOCIATED WITH RELATIONAL (CENTRIPETAL TENDENCIES) AND MODULAR (CENTRIFUGAL TENDENCIES) COORDINATION, LEADING TO A POLICY PARADOX ................................................................................... 36 ii MASSIVE MODULAR ECOSYSTEMS 1 Introduction Digitization is transforming the organization and geography of industries. Once digitized, information can be generated, collected, stored, monitored, analyzed, and processed in ways not previously possible, and when common standards are used as modular interfaces, data can be transferred and put to use with greater ease across organizations and geographic space. A crucial effect on industrial organization is the emergence of global-scale modular ecosystems associated with specific classes of products, applications, and technologies. The modules and sub-systems in these ecosystems can – albeit with significant engineering effort, because they are complex – be reused, connected, and layered to drive innovation and deliver products and services with immense complexity at scale. The nuances of this transformation have not been lost on the field of technology management and innovation (Teece 2018, Baldwin 2020, Kretschmer et al. 2020, Furr et al. 2022). The primary focus of this literature has been on how to capture value in modular ecosystems, mainly by focusing on how to influence or leverage industry architectures (Jacobides et al. 2006, Gawer and Cusumano 2014) and “win” in an era of digital platforms (Gawer and Cusumano 2008, Kenney and Zysman 2016, Van Alstyne et al. 2016, Cusumano et al. 2019). Our goal is to make three contributions to this literature, as well as literatures on global value chains (GVCs), industry standards, and industrial policy in the post “Washington Consensus” era. Our method is to focus explicitly on the depth and complexity of digital ecosystems through a multifaceted case study of the mobile phone handset ecosystem (including its outward linkages to other industries) with an analysis based on longitudinal data sets rather than emblematic examples and anecdotes. Our first contribution is to develop a broader view of modular and platform ecosystems than has been developed in the literature so far. The existing literature either develops typologies of ecosystems and their internal dynamics (Gawer 2014, Gawer and Cusumano 2014, Jacobides et al. 2018, Baldwin 2020), focuses on one industry ecosystem at a time (Teece 2018), competition between ecosystems (Hazlett et al. 2011), or the strategic implications of alternative forms of governance (Murmann and Frenken 2006, Van Alstyne et al. 2016, Furr et al. 2022). What has been largely missing is a deeper consideration of how modular industry ecosystems overlap, layer, and interconnect to comprise broader sectoral ecosystems and infiltrate adjacent industries.1 This approach allows us to frame the information and communication technology (ICT) sector as an “ecosystem of ecosystems,” a structure that we characterize as a massive modular ecosystem (MME). 1 Kenney and Zysman’s (2016) concept of “platforms for platforms” comes closest to the layered and fractal nature of industrial organization we try to convey in this paper. Quoting Stuart Feldman (p. 65), they note that at some point, when delving seriously into the structure of platform ecosystems, one sees that “it is platforms all the way down.” 1 MASSIVE MODULAR ECOSYSTEMS Our second contribution is to focus on the multiplicity of standards that bind modular ecosystems together. The platform ecosystem literature focuses primarily on the de facto standards set by platform leaders and the winners of standards wars. This focus is understandable given the “winner-take-all” (or most) outcomes resulting from these competitive battles, and the rapid growth, extreme market power, disruptive effects, and vast rents (and personal fortunes) amassed by winners. However, our research has convinced us that the vast majority of standards that govern MMEs are created by more co-operative means. Most de jure standards, for example, are created by voluntary and consensus-based processes, even as they comprise an essential infrastructure that allows modular ecosystems to form and scale. Our third contribution is to draw attention to the geographic and geopolitical implications of MMEs. The existing literature is useful for understanding how firms can create and capture value in digital ecosystems, but a critical assumption stems from the neo-liberal international order that was taken as a given at the time when much of this literature was written: a time when strategies focused on integration in global value chains (GVCs), narrow vertical specialization, and lean supply chains made sense because people, capital, goods, services, and intellectual property could flow and connect across borders with relative ease. More recently, the geo-political order has shifted, quite suddenly, from a rules-based international order focused on lowering barriers to trade and investment, toward an outcomes-based system where rules can be set aside in pursuit of narrow national interests (Rodrik and Walt 2021). In short, the nation-state has reemerged as a more potent force in structuring global industries. What happens when geography is brought back in? Kenney and Zysman (2020) rightly point out that the power of dominant platforms, and their ecosystems, is such that they are able to reconfigure the spatial economy in terms of value creation, value capture, labor markets, and industrial location; and enable new business models for users. Their argument is that economic geographers need to pay more attention to the “platform economy.” We take the opposite (complementary) tack: that the literature on technology and platform ecosystems needs to pay more attention to geography. The dominant conception of modular ecosystems in the technology management and innovation literature is that they consist purely of information flows2, and with this comes the implicit – and mistaken – assumption that interfirm information can be transferred unimpeded by factors external to firm decisions, such as international trade frictions or geopolitical rivalry.3 Barriers to trade and investment have risen quite suddenly as governments have been willing to aggressively deploy national security and industrial policies to incentivize domestic production, block perceived bad actors’ access to crucial inputs, and exclude specific companies from “critical” infrastructure such as telecom, high performance computing, and data services. Essential public goods are at risk of 2 We would like to thank Carliss Baldwin for this insight. 3 An exception can be found in the comment by Richard Nelson on an article authored by David Teece anchoring a special section of Research Policy entitled "Profiting from Innovation in the Digital Economy." Nelson takes Teece to task for ignoring the role of public institutions, including the state, in funding and performing the basic R&D upon which platform innovation is built. Still, in our view, failure to grapple with the thoroughly international character of digital ecosystems is a significant oversight by these writers. 2 MASSIVE MODULAR ECOSYSTEMS being co-opted by nation-states seeking to intervene in pre-competitive, private sector-led standard-setting processes to create long-run competitive advantages for their own national champions. As a result, firms can suddenly find themselves cut off from crucial markets, inputs, and arenas for collaboration. To provide an example of the vulnerabilities and benefits that come with global integration in complex industries, we document the geography of ownership across various layers of the mobile handset MME, including standard-setting processes, and use the analysis to explore the tension it creates between decoupling and continued accommodation of global integration. . We structure our discussion by highlighting three paradoxes evident in the ICT MME. The first paradox is that MMEs can produce extremely complex products at scale. Traditionally, production in very large numbers has been limited to relatively simple and standardized products while extremely complex products have required more customization and small-batch production. In an MME, innovation and the creation of specific domain knowledge can advance within each “module” or node of a modular ecosystem without tight interdependencies or undue consideration of other system elements, enabling production at very high scale. The space for innovation in MMEs appears to have far fewer upward or outward limits than more tightly coordinated and hierarchical industry ecosystems. The second paradox is that MMEs exhibit both high levels of market concentration – within layers and nodes – and high levels of fragmentation, both across layers and at the MME’s margins. The high levels of concentration are a consequence of the depth of technological capabilities required to compete within individual component and sub- system layers of an MME. The dominant firms in each layer are not able to control the system as a whole, however, because the system is vast and modular interfaces allow for the constant addition of new sub-systems and connections with other ecosystems. The dynamic and decentralized nature of innovation within an MME prevents dominant actors in any layer from controlling the system as a whole, much less internalizing it. At the same time, barriers to entry can be relatively low for complementors, adding another dimension of market fragmentation. The third paradox is that MMEs are both geographically clustered and dispersed. Because excellence in any specific technological or operational aspect of the ecosystem tends to be geographically and historically rooted, the firms responsible for key components and subsystems are typically concentrated in specific countries and specialized industrial clusters that are difficult to replicate or substitute for. But, similar to market structure, geographic specializations exist across the MME, making even the most dominant industrial clusters highly dependent on clusters elsewhere. At the same time, low barriers to entry for complementors means that they can and do come from almost anywhere. The rest of the paper is organized as follows: In Section 2 we review literature on the industrial organization at the firm, value chain, and ecosystem levels to develop a framework for understanding MMEs, focusing on the role of modularity at each level and 3 MASSIVE MODULAR ECOSYSTEMS the standard-setting processes that enable modularity. In Section 3 we present the defining features of MMEs, structuring the discussion according to the three paradoxes we have introduced above. In Section 4 we illustrate these features through a multifaceted case study of the mobile phone handset industry, again using the paradoxes we have identified as a roadmap. In Section 5 we show how the three paradoxes in the organization of MMEs lead a fourth “policy paradox” in which companies and countries simultaneously experience pressure to decouple from and accommodate international interdependencies. In section 6, we offer a summary conclusion and discuss opportunities for future research. 2 The role of modularity in complex global industries A defining feature of MMEs is their ability to produce complex goods and services at scale, and this is enabled by modularity. As Baldwin and Clark (2000: 5-6) write, “the essence of a complex thing is that its parts are interrelated, the different pieces must work together, and the whole must achieve more than any subset of its parts.” Without modularity, the interrelationships between parts creates the need for intense “relational” coordination, an approach that proves highly effective for the creation and exchange of tacit knowledge but is difficult to scale.4 Modularity creates a middle ground between simple markets and relational exchanges by providing formalized, codified interfaces between tasks, business functions, or stages in a value chain – thereby reducing (but not eliminating) the need for tight coordination. Figure 1 offers a stylized depiction of the trade-off between scale and complexity with different forms of co-ordination. Both modular and relational inter-firm ties occupy a space between market transactions and internalization, but modularity allows firms to share complex information without 4 Classic theories of industrial organization (Coase 1937) stress that firms sharing critical tacit knowledge tend to create joint assets and intellectual property specific to the transaction, and this “asset specificity” increases the chance of hold-up risks for both parties Firms, especially buyer (or lead) firms, seek to balance the risk and rewards of this dilemma as they make sourcing decisions. Industrial organization can be defined as an amalgam of these make versus buy decisions. Theory predicts that lead firms will tend to internalize activities characterized by high degrees of tacit knowledge (or other sources of transaction costs) and rely on arms-length market linkages for the remainder. Hierarchy within firms provides a (presumably) low risk coordination mechanism. But there are limits to what can be internalized. The knowledge, capabilities, and/or investment resources required to internalize a given business process or technology may not exist internally, or may be impractical to create in time or at a cost to be useful. Internalization also inhibits the ability to scale. Arms-length market relationships can be scaled rapidly, but are poor at transmitting the complex information needed to innovate. Relational linkages have been put forward as middle ground between hierarchy and markets. Although relational linkages can lead to relationship-specific assets and joint IP as a means of transmitting complex information between firms, economic sociologists, in particular, emphasize that such inter-firm dependencies can be managed through long-term business relationships and personal ties rather than internalization (Granovetter 1985, Powell 1990, Adler 2001). Economic geographers (Scott 1988, Storper 1995) have further highlighted that relational ties tend to build up in geographically- and industry-specific districts, clusters and regions in order to take advantage of agglomeration economies and sector-specific support institutions. But again we are confronted by limits to scale. Dense, tacit-knowledge-rich relational ties may provide an effective coordination mechanism within a region or between firms or individuals with a long history of working together, but business linkages based on social and geographic proximity cannot be expected to scale easily or quickly. 4 MASSIVE MODULAR ECOSYSTEMS revealing critical knowledge or creating joint assets, thus mitigating risk from the build- up of transaction-specific assets (Gereffi et al, 2005). In a modular ecosystem, there are often – and even essentially – strong relational linkages within modules and between close alliance partners, but also a plethora of relatively attenuated linkages between modules (Baldwin and Clark 2000). As a form of economic coordination, modularity is unique in its ability to transmit complex information at scale. Modularity can emerge within firms, within value chains, and over broader industrial ecosystems. Figure 1. Modularity as enabler of complexity at scale Market coordination = limits to complexity Modular coordination = formalized, codified interfaces between tasks, business functions or value chain stages that enable both scale and complexity Internal Relational (hierachical) coordination = coordination = limits to scale limits to scale Modularity in firms A modular interface can be created by a firm for internal use. For example, a firm can develop a core product (e.g. a computer) and a “family” of complementary add-ons (e.g. printers, information storage, compatible software, etc.), thus increasing the potential for greater system variety and scope while retaining the ability to exploit economies of scale for the core product, which becomes a “platform” for its own “complements” (Gawer and Cusumano 2014). When the platform, its complements, and any standardized interfaces between the two are produced by the same firm, the platform and interface standards are said to be “closed” (Baldwin 2021). While internal modularity governed by closed interfaces can provide efficiencies and strategic benefits for the firm, markets for both core products and complements will be limited by the firms’ internal capabilities and market reach. This limitation is one of many factors that have driven firms toward outsourcing.5 5 Bernard et al. (1995) show that when firms export products along with complements made by third parties rather than internally produced complements, a pattern they refer to as “carry along trade,” there is a positive impact on the productivity of the exporter. 5 MASSIVE MODULAR ECOSYSTEMS Modularity in (global) value chains Standard interfaces simplify the handoff between value chain stages. In the words of Baldwin (2007), they create “thin crossing points.” In the context of inter-firm value chains and networks, they allow specialized modules from various producers to be interconnected and interoperate, thus overcoming limitations on firm-specific scaling.6 An external designer of a complementary module needs to be deeply knowledgeable about her own domain and fully cognizant of the interface, but less knowledgeable of the inner workings of other modules or the platform that incorporates or combines them.7 Each participating firm can therefore focus on a narrower set of “core competencies” and innovation can take place more or less independently within various modules (Prahalad and Hamel 1999, Baldwin and Clark 2000, Baldwin 2007). But modularity has broader, industry-level effects as well. As Kenney and Zysman (2016, p. 64) put it, “Digital platforms are complicated mixtures of software, hardware, operations, and networks. The key aspect is that they provide a set of shared techniques, technologies, and interfaces to a broad set of users who can build what they want on a stable substrate.” In a modular value chain, suppliers can leverage this attribute to scale their production beyond the needs of buyers with which they have established relational ties, and buyers can source from any supplier that can meet their requirements, manage information flows cross the standard interface, and deal effectively with the “exceptions” that can be expected to arise in any complex transaction.8 Modularity in value chains therefore allows not only complex tasks to be decomposed, but it can also facilitate the entry of new actors who contribute both knowledge and competitive pressure to the system (Baldwin 2020). Resiliency is created as the system scales, because disturbances within any particular module or subsystem are localized and substitutions can easily be made (Sanchez and Mahoney 1996), so long as markets for specific modules and subsystems are not highly concentrated or the entire system is disrupted at the same moment, a point we return to in the policy section. The concept of a “modular” form of value chain linkage supporting a more loosely coupled network organizational form gained salience with the acceleration of outsourcing and offshoring in both manufacturing and services in the 1990s and 2000s (Sturgeon 2002, Dossani and Kenney 2003, Bodrožić and Adler 2017). When standardized interfaces are deployed at the level of global industries, they provided a means of forging cross-border linkages without relying entirely on dense bilateral knowledge sharing. They offer a mechanism for the accumulation of weak ties across entire populations of buyers and suppliers. This vastly expands the scale and scope of production systems to include a range of industrialized and less developed countries, each more or less 6 As Baldwin explains, “…through standard interfaces, modularization creates numerous thin crossing points, where transaction costs are low. The thin crossing points offer points of entry for would-be entrants [complementors] who produce and sell modules rather than whole systems ” (Baldwin 2020, p. 25). 7 Of course, to participate in some platform ecosystems, complementors need to be willing to share 20-30% of proceeds with the platform owner 8 The commonality of such “exceptions” is what differentiates transactions in modular value chains from simpler market transactions, where exceptions are rare and relationships between buyers and sellers can be arms-length (Ponte and Sturgeon 2013). 6 MASSIVE MODULAR ECOSYSTEMS “vertically specialized;” potentially playing one or a small number of specialized roles in GVCs (Gereffi et al. 2005, Mudambi 2008, Turkina et al., 2016, Timmer et al., 2019). GVC research has highlighted variation in governance forms within industries, in transactions based on the institutional settings where GVCs touch down, across transactions between lead firms and various suppliers, and even across transactions between the same two parties over time (Berger 2006, Ponte and Sturgeon 2013). Since GVC governance theory starts with the amalgam of bi-lateral relationships between buyers and sellers as its window into industrial organization, its utility in capturing all of this variation is limited. While several solutions have been put forward to increase the scope of GVC governance theory and analysis9, and value chains can be conceptualized as nested within ecosystems, we believe that the MME concept provides a framework better suited for capturing and explaining the complexity and variation that is so often evident within and across knowledge-intensive, digitally-enabled global industries. Modularity in industrial ecosystems Recent theoretical work on industrial ecosystems and platforms provides a unit of analysis above the level of the firm and its dyadic transactions with suppliers (Moore 1996, Adner 2017, Bogers et al. 2019, Baldwin 2020). Unlike a (global) value chain, which begins at the product level and traces the various types of contractual relationships that lead firms forge with key suppliers (especially across borders), an ecosystem involves a larger set of interdependencies, many of them informal (Furr et al. 2022). An ecosystem encompasses “providers of complementary innovations, products, or services who might belong to different industries and need not be bound by contractual arrangements – but have significant interdependence nonetheless” (Jacobides et al. 2018). An industry ecosystem is typically rife with modular linkages. Relational ties necessarily form within the ecosystem when information cannot be adequately codified – as is often the case when cutting edge products and services are being created – but if the ecosystem is dominated by relational ties, coordination challenges constrain the ability of the ecosystem to scale (and cross borders). Modular ties, by contrast, ease coordination within the ecosystem, while also allowing for each individual part of the ecosystem to retain a high degree of autonomy and follow paths of innovation independently (Jacobides et al. 2018).10 The combination of easy coordination and autonomous 9 Sturgeon (2008) explicitly suggests using the linkage between the lead firm and its most important, “1st tier” suppliers to characterize the broader industry. The GPN literature lists other important influences on the shape of GVCs (domestic institutions and norms) but offers no systematic theory for how various global industries evolve and why one industry might differ from the next. Ponte and Sturgeon (2013) provide a framework for how influences might “travel up and down” to shape GVCs at different levels (from linkages, nodes, conventions to the whole chain), and allow for the existence of “multi -polar’ GVCs populated by multiple powerful firms. In a typology of power in GVCs developed by Dallas et al. (2019) several forces structure the evolution of industries (demonstration effects, institutions that make rules and set standards, and a more amorphous constitutive form of power coming from less formal groups acting according to – often geographically-situated – norms). 10 As Jacobides et al. (2018: 2260) explain, “…technological modularity allows interdependent components of a system to be produced by different producers, with limited coordination required. While 7 MASSIVE MODULAR ECOSYSTEMS innovation enables the design and production of complex, dynamically innovative products at scale. A platform – an intermediary functional structure that facilitates transactions and governs interactions between distinct user groups – is a particularly powerful example of modular coordination within an ecosystem. A central actor offers a standard way for suppliers to connect optional complements to customers across a core platform, creating a “two- sided” market with buyers or end-users on one side and complementors on the other. Complementors have a high degree of autonomy, and can gain access to the platform’s customer base as long as they are able to conform to the platform’s design rules. The result is a system that encourages experimentation and creativity, and is able to scale extremely rapidly. Innovation occurs among a diverse set of independent actors, and failed contributions and complements do not threaten the overall system because either the platform owner or the end user can reject or remove the option with little if any penalty (Gawer and Cusumano 2014, Jacobides et al. 2018, Baldwin 2020, Baldwin 2021). While a value chain is typically controlled by a lead firm, and an ecosystem may include firms that are able to exert strong leadership, it is often the case that no single firm “controls” the complicated constellation of firms that are working together to create collective value (Furr et al. 2022), especially when the lens is pulled back to observe the institutions supporting the creation of standards and the adjacent firms and industries that can be drawn into or indirectly support the ecosystem. Standards: the source of modular interfaces The role of modularity in firms, value chains, and ecosystems is well established in the literature. However, with the exception of research on how lead firms strategically cultivate de facto standards at key interfaces to earn rents (Leiponen 2008), the varied processes of standards setting appear infrequently in literatures on industrial organization, GVCs, or ecosystems (Shiu et al, forthcoming). There is an active discussion on how firms can promulgate standards that allow them to produce complex hierarchical systems with many suppliers (Teece 2018), and how “two- sided” platforms can draw in complementors to generate network effects (e.g., Van Alstyne et al, 2016). There is literature on important roles played by non-competitive (pre-competitive, open source, and de jure) standards,11 but the topic of how standards the that govern more “mundane” transactions are generated and function in industrial ecosystems is underemphasized in the innovation and technology management literature the overarching architecture design parameters may be set by a hub, organizations have a large degree of autonomy in how they design, price, and operate their respective modules, as long as they interconnect with others in agreed and predefined ways.” 11 Here we are referring to literature on industry standards. Ponte and Gibbon (2005) argue that the existence of quality standards in industries such as wine, for example, has allowed lead firms to outsource more easily to loosely coupled suppliers. There is also a vast and dynamic literature on labor and environmental standards, for example, especially in the GVC field (e.g. Nadvi 2008). Farrell and Simcoe (2012) provide a useful typology of alternative paths to compatibility. 8 MASSIVE MODULAR ECOSYSTEMS (for details on this critique, see Langlois (2006) and Baldwin (2008)). Such standards are crucial for our study because they foster the interoperability that connects ecosystem to ecosystem, thus enabling massive modularity. The ICT industry is rife with “mundane” standards. ICT-related industry standards comprise 43% of all standards issued by the twenty largest standard-setting organizations in the United States (Farrell and Simcoe 2012: 28). A personal computer alone is estimated to implement between 250-500 distinct standards (Biddle et al. 2010), though only a handful (Windows operating system and Intel CPUs, most prominently), are de facto dominant platforms or designs. The vast majority of standards are akin to public goods, and many are all but taken-for-granted. Still, they are central to the functioning of an MME, not least because many span and often allow for the interconnection of products made in distinct ICT industries. Given that no single firm coordinates the creation of these standards, nor ensures their interoperability within an integrated end-product, it is important to explore how they emerge and eventually lead to the creation of an MME. Table 1. highlights two dimensions of standard-setting in modular industry ecosystems. The first is the distribution of power among the actors that define a standard, ranging from concentrated to diffuse. Along this spectrum we identify three ideal types: monopolistic, where one or a very small number of actors set the standard, to oligopolistic, where a larger number of strong actors work in concert or in alliance to define a standard, to multi-stakeholder arrangements that are more consensus-based. The second dimension is whether the resulting standard is de jure or de facto. De jure standards are created, ratified, and endorsed by formal organizations with clear membership criteria and rules for how standards will be created and used. De jure standards can be set by governments with authoritative relationships within a jurisdiction, or through a variety of voluntary organizations led mainly by the private sector, ranging from smaller, less formalized, less egalitarian and more closed-membership consortia, to consortia with more open membership rules that may convene temporarily to solve a particular industry-level problem, to larger and more formalized and permanent standard- setting organizations (SSOs). With de facto standard-setting, stakeholders do not come to formal agreement on how a standard is developed; they are outcomes of market competition. “Competitive dominant designs and platform rules” emerge as a result of competitive success of the design’s owner, and earn monopoly rents (e.g., Microsoft Windows or Apple’s iOS mobile operating system), while “pre-competitive dominant designs” are standards that may have been widely adopted as a result of the long-ago competitive success of a dominant actor, but have, over time, become part of an industry’s taken-for-granted infrastructure (e.g., Ethernet, JavaScript, or the protocols defining the layout of electronic circuit boards or microchips discussed later). The categories in Table 1. are ideal types, and there are many hybrid and mixed forms. In almost all cases, standards need to evolve and be updated over time, and the process of shepherding a standard through various iterations, and sometimes ensuring “backward compatibility” with prior standards, is a main occupation of all standard-setting 9 MASSIVE MODULAR ECOSYSTEMS processes. Without known standards to govern interactions between firms in an ecosystem, be they owned by dominant actors or freely published, modularity will be limited and the ecosystem will not scale. In our view, MMEs cannot be fully understood without documenting the main standards that bind the system together. Table 1. Distribution of power and standard-setting processes in modular ecosystems Standard setting process De facto De Jure Government agencies Competitive dominant designs Concentrated and platform rules (e.g., U.S. FCC bandwidth auctions Monopolistic and frequency requirements, (e.g., Android and iOS APIs, location services created by semiconductor foundry military and other agencies: compatibility rules) GPS/GLONASS/BeiDou) Distribution of power Consortia, associations, for aa, Standards battles and alliances Oligopolistic (e.g., Betamax vs VHS) (e.g., Blu-Ray Association, WiMax Forum) Pre-competitive dominant designs (e.g., Ethernet, Gerber, GDSII) Standard-setting organizations Multi- Open-source projects that include (SSOs) stakeholder widely accepted interface (e.g., 3GPP, WIFI Alliance) Diffuse standards (e.g., Linux; various software languages, compilers, converters, & translators) 4 Defining massive modular ecosystems Following the literature on complex systems and modularity in design (Simon 1962, Sanchez and Mahoney 1996, Baldwin and Clark 2000, Murmann and Frenken 2006), we view the organization of an MME as a vertically nested hierarchy of ecosystems – final products, subsystems, components, and other inputs. While both vertical slices and horizontal layers can be, and often are, viewed as ecosystems unto themselves, closer observation reveals that they are linked to ecosystems in adjacent industries, not only vertically across value chain stages, but laterally and diagonally, since sub-systems and components flow into multiple industries. An MME is, therefore, an ecosystem of ecosystems, each based on a variety of standards, some distinct and some shared. While industry ecosystems and the standards that govern inter-organizational linkages in the ICT sector can take various forms, our research suggests that digitization and the elaborate array of standards that support it are driving the MME organizational form across broad swaths of the global economy. 10 MASSIVE MODULAR ECOSYSTEMS Three paradoxes in MMEs A stable ecosystem, Baldwin (2020) argues, has a balance between its centripetal and centrifugal forces. On the one hand, the centripetal force exerted by a standardized interface between tasks (or business functions) binds the ecosystem together through complementarities and interdependence. On the other hand, the need to access capabilities of highly specialized firms that are widely dispersed across an ever- expanding ecosystem serves as a centrifugal force within the ecosystem, preventing it from collapsing into a single dominant organization controlling all of its elements: final products, standard interfaces, inputs and complements. The co-existence of centripetal and centrifugal forces in modular ecosystems gives MMEs multiple characteristics that appear contradictory. In this section we focus on three paradoxical qualities of MMEs: 1) they can produce highly complex products and systems at scale; 2) they demonstrate high market concentration within various ecosystem layers and market fragmentation across the MME as a whole; and 3) the capabilities required to dominate any particular layer or node tend to be concentrated geographically, but the MME is typically spatially dispersed overall. Complexity at scale The key to understanding the concept of an MME is to recognize the multiplicity of platforms, dominant designs, open-source standards and complements that reside in various nodes and across layers. As we explained in the previous section, the looseness of coordination afforded by standardized interfaces, the optional nature of complements, and the presence of network effects allow modular systems to scale rapidly. An MME, like most complex systems, is not composed of elementary components all joined together, but is structured in a set of layered, nested, and overlapping systems, subsystems, and key components (Simon 1962, Tushman and Murmann 1998, Murmann and Frenken 2006). Complex products and systems have also been a focus of literature on “systems integration” (Prencipe et al. 2003, Brusoni 2005), a term that usually refers to the organizational requirements for designing, building and coordinating the production of hugely complex products and systems in low volumes, such as highway bridges and military systems such as aircraft carriers. The range of knowledge required to perform all of the various tasks required to produce, maintain, and operate a complex system are vast, the level of specialized knowledge that is required to perform each task is deep, and the coordinating function of systems integration is extremely demanding (Hobday 1998). Although an ecosystem of firms can be coordinated by multiple means (i.e., prices, bilateral contracts, multilateral contracts and platforms), according to Baldwin (2021), ecosystems without modular platforms tend to be smaller and more fragmented since platforms create network effects that create additional possibilities for scaling and attracting complementors. By balancing the degree of centralized coordination provided by a platform leader with the diversity and experimentation of complements, only “platform ecosystems support the coordination of modular systems and distributed decision-making at scale” (Baldwin 2021: 21, emphasis in the original). The ease of connecting to or “complementing” to modular ecosystems means that the milieu of potential complementors, partners and competitors is open-ended. 11 MASSIVE MODULAR ECOSYSTEMS Unlike value chains (global or otherwise) which have identifiable lead firms at the product level (i.e., brands) or intermediaries that control access to the supply chain, MMEs often lack a single leader or orchestrator that fully creates or directs the system as it scales. In complex systems governed by modularity the integration function is simplified. Instead of being managed by a heavyweight system integrator, MMEs invite the participation of many system and sub-system integrators, allowing the MME to scale rapidly and produce at high volumes. The processes of co-evolution in broad sectoral and industry ecosystems are more organic, since the modular nature of MMEs allows new functions to be added (and removed) in response to changes on both the demand- and the supply-side without severely impacting or putting at risk the other elements of the system. The result are sectors and industries that can grow and evolve in dynamic and unpredictable ways. Market concentration and fragmentation In an MME, the processes of technological upgrading and functional accretion means products and services are constantly gaining performance improvements and new capabilities. Concentration tends to come with specialization in technologically demanding and dynamic sub-systems for several reasons. First, the depth of the required technological capabilities and the huge and ongoing capital investments required to remain at the cutting edge of the technology create ever higher barriers to entry. Outsourcing of even critical inputs is the norm because after a certain level of complexity, it is unlikely that any single final product level firm will have the range of capabilities or necessary investment resources to master all of the highly specialized and technologically demanding inputs needed for the system to function. This drives specialization and market concentration at the sub-system and (complex) component levels. Given fierce competition, buyers place intense demands on suppliers, and laggards in fiercely contested sub-system markets tend to drop out or migrate to less demanding industries, leaving the supply-base for key sub-systems and components with fewer participants. Second, both dominant system integrators and supplier firms that dominate key input sub-industries are sometimes able to build, along with their capabilities, strong patent positions that protect them from new entrants, control which firms have access to their IP licenses, and provide a stream of licensing revenue, thus enabling a virtuous cycle of further investment in the R&D that created the dominant position in the first place. Third, firms in MMEs can sometimes exert platform leadership within their niche by setting standards and rules for sub-system complementors, as discussed earlier. Despite the tendency towards market concentration at the system and sub-system levels within an ecosystem, the MME as a whole is unlikely to become fully comprised of monopolistic or oligopolistic markets. The powerful tendency towards concentration within an MME is balanced by the ease with which new actors are able to join: the rapid pace of innovation and functional accretion enabled by modularity is constantly adding new layers to the MME, the weak ties and strong modular interfaces of platform ecosystems create opportunities for rapid addition of complementors, and a reliance on 12 MASSIVE MODULAR ECOSYSTEMS open source technologies and global standard setting organizations (SSOs) provide opportunities for external actors to participate in design and governance. The result is that the MME as a whole is segmented into a series of oligopolistic markets rather than one, and the fragmentation of the overall system works to diffuse the power of even the most dominant actor. Geographic clustering and dispersion A focus on the geographic pattern of clustering and dispersal shifts the unit of analysis from the firm to the nation-state and the interlinking of sub-national industrial clusters. Research on industry GVCs has shown that locations within specific global industries have shifted from industrial to functional specialization (Turkina et al., 2016; Timmer et al., 2019), and that lead firms across a variety of knowledge-intensive industries increasingly share common global suppliers (Sturgeon, 2002; Sturgeon and Lester, 2004; Sturgeon et al., 2017; Turkina and Van Assche, 2018). Although the pattern of concentration and fragmentation is related to the market dynamics described in the previous section, they are not the same. Geographic concentration, as measured by the headquarter location of firms, can accentuate the level of concentration in an MME layer or node. A module that has an oligopolistic market structure, for example, will be a monopoly at the country level if all of the market-leading firms are located the same country. Similarly, a module that has a fragmented market structure may still be concentrated geographically if all industry players are headquartered in one or a small number of countries. This intensification of concentration at the geographic level is the natural result of path-dependent competitive advantages at the national level, agglomeration effects, and industrial clustering. If an industry is dominated by a small number of firms, it is not unusual for them to be located in industrial clusters supported by specialized labor markets, supporting firms, and institutions (Scott 1988, Storper 1995, Saxenian 1996, Locke 1997, Thun 2006). The geographic dispersion of specialized activities is not new, but in MMEs the patterns of dispersion are more complex than when business functions are spread across borders by multinationals and their affiliates, or via international sourcing relationships in GVCs, as depicted stylistically on the left and center of Figure 2. A multinational firm can access capabilities, engage in cost and currency arbitrage, and tap into markets in multiple countries by creating affiliates in many countries to carry out specific business functions (e.g., R&D, manufacturing, sales, software development, back-office administration). A GVC, by contrast, involves both offshoring and outsourcing, creating a chain (or web) of business relationships orchestrated by a lead firm, with suppliers distributed across different geographies. The lead firm in a GVC follows roughly the same business motivations as multinational firms, but the additional element of “outsourcing” implies a shift in strategic objectives to include leveraging the capabilities of suppliers and capturing and protecting the high value-added activities within the chain. 13 MASSIVE MODULAR ECOSYSTEMS Figure 2. Geographic implications and strategic objectives for different forms of industry organization The objectives and strategies of an actor in an MME are even more complex, given the multiple layers and standard-setting processes in the system. As depicted on the right side of Figure 2, MNEs and GVCs are embodied within MMEs. Actors in GVCs will try to gain capabilities and market power within their business function and industry niche and leverage the capabilities of suppliers and other business partners, while actors in MMEs must also try to control, influence, or at least understand the interface standards relevant for their niche as well as niches across layers. This has implications for the activities the firm needs to engage in. It can mean, for example, allotting substantial engineering hours to contributing to and leveraging the activities of standard setting organizations (SSOs) and open-source communities.12 For instance, Qualcomm, the U.S.-based semiconductor design firm that dominates the market for mobile handset CPUs, is a member in over 160 distinct standard-setting organizations, crossing over many MME layers (Casaccia 2017). In addition, there is no single lead firm orchestrating the entire MME. 5 Mobile phone handsets – a massive modular industrial ecosystem Mobile handsets provide an ideal case study for examining how multi-layered modularity emerges, impacts competition, and shapes the organization and geography in a complex industry. As in other industries within the ICT sector, modularity in mobile telecom is enabled by the hundreds of de-facto and de-jure standards that have emerged through an uncoordinated process of technological evolution, market competition, specialization, and standard-setting. At the same time, mobile telecom comprises a distinct industry MMEs embedded within the much larger sectoral MME of ICT. In this section, we provide a brief account of how the mobile phone handset portion of the mobile telecom MME evolved from an industry producing products with highly integrated product architectures into an MME. We then provide evidence demonstrating the three paradoxes of a MME. Data The evidence we present on the mobile phone industry in this section relies on pooled longitudinal data from multiple sources. Bill of materials (BOM) data for 456 mobile 12 As Shaikh and Levina (2019) argue, one of the best ways for companies to tap these resources is to become an important contributor. 14 MASSIVE MODULAR ECOSYSTEMS phone handsets introduced between 2008 and 2019 is from the market research firm IHS Markit. Data on mobile phone handset specifications were obtained from various sources and websites, including the Teoalida and PhoneDB websites. Data on company contributions to various releases of Google’s Android operating system were scraped from folders found on the Android Open-Source Project website. Data on company contributions to each generation of mobile telecom standards was scaped from folders found on the 3GPP website. Other descriptive evidence was taken from the data aggregator Statista, with original sources indicated. Additional detail on the data sources used in this paper can be found in the appendix. The evolution of handset industry architecture – from integrated to modular The first-generation mobile telecom interconnect standard (1G) was developed in the early 1980s, a time when mobile communications technology was mainly analog and handsets had integrated product architectures, more or less unique to each device. We can refer to them without elaboration as “phones” because this was the primary function at the time. Levels of vertical integration were much higher than they are today, and when components were outsourced, semiconductor firms and other key suppliers typically had either relational or captive ties with the lead firm (Park and Ogawa 2009: 792). The second-generation standard (2G) emerged in the early1990s, during the shift from analog to digital: when a person spoke into a handset, the sound generated an analog wave that was converted into a digital signal represented by 1s and 0s, with the process reversed for incoming signals.13 In a shift that would become critically important later on, digitization also opened the door to transceiving data signals, a feature that ultimately opened a pathway toward greater functionality and modularity. However, this potential was not achieved immediately. The increase in modularity across 1G and 2G was a 15- year process, and even when the design of low-end handsets became highly modular, leading-edge firms continued to pursue more integrated strategies and relational linkages to key suppliers for high-end handsets, a pattern that, to some degree, persists today (e.g., with Apple, and a lesser degree Samsung and Huawei).14 13 This enabled faster speeds and fewer dropped calls because signals could be sub-divided into data “packets” and sent over multiple pathways instead of being transmitted in a linear stream. Sound quality was also improved with digital signal processing (DSP) semiconductors that filtered out noise and otherwise digitally enhanced the signal. 14 In 2G, the technical challenges associated with DSPs, in particular, led the leading handset firms – which at the time included Nokia (Finland), Motorola (USA), and Ericsson (Sweden), all of which were telecom equipment producers with close ties to the network interconnect standard-setting process – to initially adopt integrated approaches to handset design. By the mid-1990s, specialized semiconductor firms such as Texas Instruments (TI hereafter), Analog Devices, Lucent (all USA-headquartered), and Philips (the Netherlands) began bundling the complete stack of interconnect and analog-to-digital conversion protocols into highly functional technology platforms, or “chip sets,” that made it easier for a broader range of handset brands t o design basic, low-cost handsets for mass markets. Industry leaders such as Nokia responded by deepening relational ties with key suppliers (TI in particular) to develop cutting-edge handsets closely aligned to the newest generation of telecom infrastructure (Imai and Shiu 2010). 15 MASSIVE MODULAR ECOSYSTEMS The introduction of 3G technology in the 2000s elevated the importance of the operating system (OS) software. Initially, the OS for higher-end phones was a closed platform running applications that were developed internally by phone and chip-set designers for exclusive use on phones produced by the handset company,15 while lower end phones, especially in China, shifted quickly toward more modular systems when MediaTEK (, China) introduced highly integrated chip set platforms and reference designs, substituting for internal capabilities.16 Over time, the combination of increased capacity for data transmission made possible by 3G technology and improvements to CPUs and graphics processing units (GPUs) included in phones – innovations that responded to the popularity of video playback and on-line gaming applications in particular – began to transform the mobile handset into the internet-connected computing platform that we know today as the smartphone (Thun and Sturgeon 2019). In the late 2000s, a pivotal shift toward massive modularity in mobile handsets was led, not by an incumbent from the telecom industry, but by two companies from adjacent industries: a personal computer company and an internet search company: Apple and Android (Google). Not surprisingly, their approaches were fundamentally different than the one used by handset incumbents. With the launch of the iPhone in January 2007, Apple created a platform that combined closed and open access elements.17 The 15 Finland’s Nokia Symbian OS software was efficient and reliable, but the lack of a standardized modular interface between the OS and applications – what is referred to as an application programming interface (API) today – created a barrier for third party app developers, truncating the network effects of the platform and therefore its scalability. Even within Nokia, almost every handset was coupled with a different version of Symbian, and because there were dozens of versions that were not entirely compatible with each other, third-party app developers were frustrated by constant delays and uncertainty caused by the lack of a common platform or API for Nokia phines, much less across multiple handset producers (Lamberg et al. 2021). Perhaps not surprisingly, given its origins as a telecom equipment company, Nokia’s focus was on matching software designs with appropriate CPU features so as to optimize network performance (e.g., improved voice clarity and fewer dropped calls) rather than focusing on the capabilities of handset applications. As a result, telecom operators were often relied upon to work with app developers to bundle applications with handsets – and in fact, the features available on mobile phones at the time were largely under the control of network operators. The lack of scaling motivated Nokia to develop a consortia including several other handset producers (Samsung, Motorola, and Sony-Ericsson) in the Symbian ecosystem, but the challenge of assuring the compatibility of platforms and complements in an increasingly cost-competitive consumer-oriented market continued to be daunting, especially when spread across several handset brands using different processors and product architectures. Furthermore, there was no centralized marketplace for Symbian applications where users could discover and purchase software. If the app was not bundled with the handset, users had to purchase the software directly from each developer. 16 MediaTEK, a fabless semiconductor design firm based in Taiwan, China, went a step further when it began offering an integrated system on a chip (SOC) for mobile handsets, along with software and detailed instructions on how to implement them. These “reference designs” covered not only how phones connected to the network, but also included many of the new functions that were being added at the time, such as address books, audio playback, image processing for basic cameras, and interfaces to the memory chips needed to store this information. These high modular solutions resulted in very similar phones, yet allowed phone designers some leeway for customization, albeit in relatively superficial aspect of handset appearance (Imai and Shiu 2010; Brandt and Thun 2011). 17 The OS of the iPhone (iOS) was developed in-house – adapted, ironically, from an open-source kernel of Linux – as a closed platform unable to be altered by external developers or used by other handsets brands. Critically, Apple published its APIs so external developers could build compatible apps for sale on its on- line App Store, a model drawn from the ecosystems surrounding its line of MacIntosh personal computers, 16 MASSIVE MODULAR ECOSYSTEMS innovation was to provide software developers with an application programming interface (API) that would allow the software access to the iPhone’s Operating System (OS). In this way, app developers could retain their ability to innovate freely at the level of user experience and application interfaces. In return they had to abide by Apple’s rules governing participation on the platform and pay Apple a 30% share of revenue. Google’s Android, which was launched in 2008, adapted a hybrid approach to platform development. Like Apple, Google developed Android Open-Source Platform (AOSP) as a powerful de facto platform that connected Android-based handsets with a rich ecosystem of complements, and it used its monopoly power over the platform to create rules within the ecosystem to exert substantial control over both handset architecture and rules for complementors. Unlike Apple, however, Google adapted an open-source and non-proprietary approach to developing its platform.18 Control was maintained by providing differential levels of access to AOSP: the average firm was granted access to pre-packaged, standardized and mature versions of the operating system, while experimental next-generation, customizable versions were only available to a select number of handset firms with which Google maintained close relational ties. This system of rules allowed Google to retain substantial control over the platform and to use its leverage to create network effects to its advantage.19 The shift to modularity in mobile handsets was not bounded by the advent of iOS or Android; it was part of a much broader transformation within the ICT sector that began and the iTunes music store set up for its iPod portable music player. Apple had the market power to set a de facto standard with its APIs because of the instant popularity of the iPhone, which was driven in part by the prior popularity of the iPod/iTunes platform (iPod functionality was included in the iPhone until 2022), and Apple’s strengths in system and industrial design. To give app developers more freedom to innovate and customize the user experience, the iPhone dispensed with the fixed keypad of earlier smartphones, allowing the user interface to be entirely customized according to the needs of the application. This also eliminated requirements to synchronize app development with hardware design, and gave app developers the opportunity to innovate more freely, as long at the modular interface between applications and the OS (defined by Apple’s API) was adhered to Apple received its cut of revenue. In this way, the combination of the iPhone and the App Store created powerful network effects and the touchscreen smartphone quickly became the dominant form factor in the industry. 18 AOSP contained the core source code (including an open-source Linux kernel, which itself was a hybrid of many ad hoc de facto software standards), along with open-source apps (e.g., maps, calendars, etc.), security and testing functions, extensive documentation, and a wide variety of design and development tools. The AOSP drew from a creative and freewheeling open-source community to create a vast codebase and toolkit from which different versions (‘skins’) of Android could be customized and innovated upon by OEMs, component companies and Google itself. 19 To allow handset-makers the flexibility to customize the look of Andoid, Google requires devices to be approved through Google’s Android Compatibility Program to assure that the modified OS is not a non- compatible “forked” variant that would splinter the ecosystem (and undermine network effects). If handset brands are approved, they gain access to Google’s proprietary assets that are not technically part of Android, including Google Mobile Services, GooglePlay (the gateway to millions of third-party apps), Google’s proprietary apps (e.g., YouTube, GoogleMaps, Chrome) and other features (e.g., voice commands, GooglePay). This leverage allowes Google to maintain coherence and uniformity across Android devices, as well as exert control over OEMs by directing their billions of users onto Google’s licensed apps and services (e.g., search), where Google earned revenue from advertising placement. In short, Android combines contrary elements – a monopoly and de facto platform supported by multiple open-source communities and complimentors. 17 MASSIVE MODULAR ECOSYSTEMS much earlier, and is in fact “baked” into how the industry has evolved from the 1970s onward. Two key examples are Gerber and GDSII, de facto standard interfaces between design and manufacturing at the semiconductor and circuit board levels, respectively. Standardizing the interface between ICT design and manufacturing enabled the organizational and spatial separation of product innovation from manufacturing, a signature feature of GVCs and a main driver of industrialization in places like southern China, where the bulk of the world’s mobile handsets are produced.20 Shifting to the current situation that these earlier changes helped to enable, Figure 3 depicts some of the key the subsystems, business processes, and standards that comprise the mobile handset MME. It depicts how, in a contemporary mobile handset, multiple platforms operate in different hierarchical layers, linked vertically, with codified links to subsystems that do not operate as platforms. The triple blue lines represent modular linkages connecting both platform and non-platform elements within the mobile handset MME and how each of these links rely on standards created through one of more of the processes listed in Table 1. While lead firms in MMEs (sometimes referred to as system integrators) can and do occasionally opt for internalization of critical components as a strategic measure, and MMEs as a whole will necessarily include many relational ties, modular interfaces are essential because they are required to manage high and rising complexity as well as to allow innovation to proceed independently within various module domains. At the top of the figure, in the center, is the handset itself, linked to a set of contract manufacturers based on the requirements of such firms but also with the aid of industry standard design files (such as the Gerber standard for circuit board layout). On the right- side top are the “front end” radio modules that send and receive signals from the cell towers based on prevailing network connectivity standards as deployed by specific carriers. In the center of the figure is the handset operating system, which manages the 20 The design data of a printed circuit board – a foundational building block of modern electronic devices – is captured in the de facto standard of a Gerber file, which is then used to transfer layout specifications between the designer and a fabricator. The file specifies the physical layout of the circuitry embedded in the board, which provides design information to the circuit board manufacturer, and since information on where the wiring breaks the board’s surface is included, it can also be used in the assembly phase to program the robotic assembly equipment that places components onto boards prior to final product assembly. All software programs that hardware designers use to generate design data, including the U.S.- based Autodesk, the dominant player, generate “Gerbers” as the final step in the design process. Gerber files are named after Joe Gerber, an American developer of digital drafting machines. The original focus of his company, Gerber Scientific, was mechanical plotters, but by the 1980s the firm had shifted to computer- aided design (CAD) systems. Although other firms developed similar formats, because Gerber had published a full specification of their format (RS-274-D) in 1980, the file format and its successive generations became the industry standard. Many product level firms (system integrators) and component producers that had been taking an integrated approach to design and manufacturing shifted their focus to design and relied on contract manufacturers for production, a shift that was enabled by the modular interface offered by the Gerber standard (Sturgeon 2002). This was also the case in mobile handsets: by 2015, 46% of handsets were being assembled by contract manufacturers (IHS Markit), with Apple famously relying almost exclusively on the company Foxconn of Taiwan, China, for assembly (mainly in plants located in China). The Republic of Korea’s Samsung continued to assemble most of its phones in its own factories, but most others, including Huawei and smaller Chinese brands, turned to the robust contract manufacturing base in China. 18 MASSIVE MODULAR ECOSYSTEMS flow of data within the phone, including an ever-evolving set of non-platform sub- systems and components (memory, displays, etc.), and also provides a platform upon which to run compatible third-party applications via the API. Application oftware is tpically designed with the aid of industry-dominant application design software and plug- ins. Depicted at the bottom center of the figure are platforms providing additional wireless connectivity: to the internet (via WIFI), satellite-based location services (GPS), and to close-by terminals such as payment kiosks (nearfield transmitters). These capabilities enable a host of 3rd party devices to be connected to, and in many cases, controlled by the handset. The left-center portion of the figure depicts the mobile-specific CPUs (e.g., from Qualcomm) which, working together with the operating system, control the operation of the phone. Like the design of the handset, CPU designs can be handed off to contract manufacturers (semiconductor “foundries”) with the aid of a standard design format (e.g. GDSII) as long as close attention is paid to the requirements of the foundry. In another distinct layer of the handset MME, depicted in the figure’s lower left corner, we see that CPUs are almost always designed according to circuit logic (IP blocks) licensed by ARM, a UK-based company that has its own ecosystem of complementors and tool makers which allow CPU and handset-designers to implement its architecture.21 Figure 3. Layered modular ecosystems in the mobile phone handset MME and links to adjacent industries 21 Application Programming Interfaces (APIs) are the instructions platform owners publish so software developers can build applications compatible with the platform. Central Processing Units (CPUs) are semiconductors that perform a myriad of functions in complex ICT-products, including running compatible software applications. Some CPUs are designed and produced by the same company (e.g., The US’s Intel and Republic of Korea’s Samsung) but most are designed by “fabless” chip design firms (e.g., the US’ Qualcomm and Taiwan, China’s MediaTEK) and fabricated by contract manufacturers, or “foundries” (e.g., Taiwan, China’s TSMC and Singapore’s SSMC). IP cores are blocks of function -specific code that can be used to design blocks of function- specific circuity on semiconductors (especially CPUs) or as a rules-setting architecture that integrates various aspects of the CPU (e.g., the UK’s ARM). IP blocks are available from specific vendors (e.g., ARM) and can also be drawn from “libraries” that are included with IC design software (e.g., the US’ Cadence and Ansys). 19 MASSIVE MODULAR ECOSYSTEMS While complex products rarely share identical subsystems and components, their producers typically serve multiple buyers and often multiple industry ecosystems, depicted by the triple yellow lines linking elements of the mobile handset MME to other ICT ecosystems. Even with these adjacent links taken into consideration, the depiction of a single product in Figure 3 only hints at the complexity and depth of the mobile handset and ICT MMEs. 22 Complexity at scale The rise of modularity in mobile handsets has enabled the industry to solve the problem of complexity at scale without any single firm leading the creation or orchestration of the overall system. The organizational challenges involved with the production of a modern mobile handset are immense. In 2019, the number of distinct components in a handset ranged from 2,530 in Huawei’s Mate 20X (5G) to 479 in Shenzhen-based Elong Mobile’s W45. The multiple technologies that reside in these components must work together to create a good user experience. Phones must work under a wide variety of real-world conditions (e.g., on an elevator moving through a high-rise building with a dozen or more similar devices operating within a few feet). The largest semiconductors (CPUs, core memory) may contain billions of micro-components. Nevertheless, mobile handsets are produced in huge volumes and have been evolving and improving very quicky in technological sophistication and performance. Scale The modular approach to system design and application software was hugely successful for participants in both Apple and Google’s ecosystems, and devastating for incumbents unwilling or unable to pivot to Android. As can be seen in the blue line in Figure 4, sales of smartphones increased rapidly after the introduction of Android in 2008. Between 2009 and 2015, by which time Android had become the dominant operating system along with iOS (see Figure 5), sales of mobile handsets increased at a compound annual rate of 33.7%, peaking at 1.6 billion units shipped in 2017. Smartphone penetration rates grew accordingly from about 50% in 2016 to 78% in 2020 (Ericsson Mobility Report, 2021). By contrast, annual shipments of feature phones peaked at 1.2 billion units in 2011, and fell year on year thereafter; only 286 million were shipped in 2020 (see the red line in Figure 4). 22 It almost goes without saying that the ICT sector is comprised of multiple industries, made up of infrastructure and product categories such as telecommunications networks, mobile handsets, cloud computing and digital services, personal computers, consumer electronics, industrial electronics, automotive electronics, military electronics, and so on. While exact iterations (e.g., part numbers) of complex components and software are rarely shared across end products – since some variation or even minor customization is characteristic of modular systems – the firms, technologies, and standards present in the ICT sector generally serve multiple ICT sector industries. There are also multiple input industries, including various types of semiconductors, displays, circuit boards, and software and process industries that serve nearly all of these, such as contract manufacturing (at the semiconductor, circuit board, product levels), contract design services, outsourced software development, and other types of technical consulting. While each of these industries tends to have distinct ecosystems, there is a loose hierarchical structure in which the output of each lower order industry typically flows into multiple higher order industries. Standards also form the basis for modular interfaces across multiple industries, more often than not. 20 MASSIVE MODULAR ECOSYSTEMS Mobile app availability data show a pattern of gowth even more dramatic than mobile handset shipments. As the network effects of Android gathered steam with the introduction of new third-party apps, the compound annual growth rate of new apps on GooglePlay spiked to 63.6% between 2013 and 2018, with available apps rising from 700,000 to 3,600,000 in the period (see the green line in Figure 4).23 Figure 4. Mobile handset shipments and available smartphone apps, 2007-2020 1.6 billion smartphones @ peak 1.2 billion feature phones @ peak 3.6 million apps @ peak Sources: Feature phones: Statista based on IDC and CCS Insight; Smartphones: Statista based on IDC and Gartner; Mobile apps: Statista based on data from Google, App Annie, and AppBrain, as published by AppBrain. Note: In summer of 2018 Google removed a large number of apps due to an update to the company’s Developer Policy. In short, within a few years, incumbent handset brands that were not running Google’s Android or Apple iOS, including perennial market leader Nokia, had all but disappeared from the market (see Figure 6).24 Of the top 5 firms prior to 2007, only Korean brands Samsung and LG were able to successfully make the transition to Android (Thun and Sturgeon 2019), though after struggling for years, LG left the market in 2021. 23 In 2021, there were 3.5 million apps available on GooglePlay and 2.2 million available on Apple’s App Store. (https://www.statista.com/statistics/276623/number-of-apps-available-in-leading-app-stores/) 24 KaiOS, from Hong Kong SAR, China was launched in 2017 with the intention of creating a platform, based on the Mozilla Foundation’ Linux-based browser Firefox platform and including a developer API and app store for keypad-based feature phones, thus implementing a business model Nokia and Symbian failed to achieve in the early 2010s. 21 MASSIVE MODULAR ECOSYSTEMS Figure 5. Mobile handset OS market share, January 2012 – April 2021 Others Other/Unknown < KaiOS (Hong Kong SAR, China) Blackberry (RIM - Canada) < Windows Mobile (Microsoft – USA) Samsung (Republic of Korea) Symbian (Nokia and others) Series 40 (Nokia - Finland) iOS (Apple - USA) Android (Google - USA) Source: Statista based on StatCounter, which calculates OS data based on more than 1.7 billion page views per month worldwide. StatCounter defines a mobile device as a pocket-sized computing device - tablets are not included. Product complexity Product complexity is associated with three features of mobile handsets: the depth of required capabilities (as performance requirements increase), the breadth of required capabilities (as the demand for variety by end users and consumers increase), and the speed of improvements and innovation. The follow data are based on a dataset scraped from PhoneDB that contains over 15,000 phone models. The steady increases in the depth of required capabilities is evident in the dramatic improvements in the processing speed and functional capabilities of mobile handsets that are summarized in Table 2. Central processing units (CPUs) run applications and manage the flow of information within the handset. CPU clockspeed, a measure of how often calculations are made, has increased from an average of 615 megahertz (MHz) in 2009 to 2,406 in 2020. Another approach to increasing the performance of CPUs is to include multiple processing cores within the main semiconductor. In 2009, all handsets had a single processing core, but by 2020, the average number was 6.8. Each core can perform processing functions independently, allowing users to simultaneously run multiple applications, and also support the flow of information in and out of the phones across multiple wireless frequency bands without noticeable degradation of performance. Accessing to and switching across multiple bands allows users to send and receive data simultaneously, conference multiple voice calls, and connect to networks operated by different carriers and in various countries on the fly. On average, handsets in the PhoneDB dataset had the capability to access 5.2 frequency bands in 2009 and 25.7 in 2020, while available data links increased from 6.9 to 19.0. In 2009, only 26% of 22 MASSIVE MODULAR ECOSYSTEMS handsets included graphics processing units (GPUs), but by 2020 99% did, although they are now less likely to appear in handsets as a discrete module and more likely to be encapsulated within the CPU. In addition, to cache working data and speed processing, handsets have been equipped with greater amounts of volatile memory (DRAM) as well, also increasingly encapsulated in the CPU. In general, encapsulating functions on the CPU module speeds processing, lowers costs, and shifts complexity from the design of the handset to the design of the CPU. As discussed below, encapsulation can also capture markets previously occupied by suppliers selling single function modules, driving consolidation in the supply base. To store all of the photos, videos, application software, and other data users create, the capacity of the main memory module (static random-access memory, or SRAM) has increased almost 40% per year, on average, from 4 megabytes in 2009 to 147 in 2020. To improve image quality for photos and video, specialized cameras (e.g., zoom, wide angle, selfie) have been added to higher-cost handsets and resolution has improved from about .24 million pixels to nearly 11, on average. In order for users to view the improved images on their handset, display resolution has increased accordingly, from about .19 million pixels in 2009 to nearly 2.5 in 2020. Battery life has increased as well, with average talk time increasing from 7.5 hours in 2009 to 30 in 2020 while the amount of energy storage improved from 4.7 watt-hours to 17.0. This is doubly impressive since the increased energy needs created by all the other performance improvements needed to be supported as well, entailing major advances in both battery capacity and power management (a main benefit of ARM’s CPU architecture). Table 2. Mobile handset performance and functional improvements (2009-2020) Function Measure 2009 2020 % Change CAGR Application & graphics processing (CPU) Clockspeed Avg. MHz 615 2,406 291% 13% Processing cores Avg. # of cores 1.0 7.9 686% 21% Cache memory (volatile - DRAM) Avg. gigabyte 0.2 6.8 3,114% 37% GPU Phones w GPU 26% 99% 281% 13% Main memory (non-volatile - SRAM) Avg. gigabyte 4.0 146.8 3,579% 39% Display Resolution Avg. # pixels 189,686 2,478,871 1,207% 26% Pixel Density Avg. density 188 387 106% 7% Scale Avg. # of colors 2,315,489 265,413,043 11,363% 54% Camera Video Resolution Avg. # pixels 237,492 10,997,856 4531% 42% Optical Zoom Avg zoom (x -times) 1.0 1.7 68% 5% Primary Camera Avg. # camera functions 1.4 14.9 964% 24% Secondary Camera (2013-20) Avg. # camera functions 2.0 9.5 375% 22% Battery Capacity Avg. watt-hours 4.7 17.0 262% 12% Talk Time (2010-20) Avg. hours 7.5 30.4 305% 15% Network Connectivity Frequency Bands Supported Avg # of bands 5.2 25.7 393% 16% Cellular Data Links Supported Avg. # of links 6.9 19.0 177% 10% Other wireless functions WIFI standards supported Avg. # of standards 2.1 5.2 150% 9% Navigation systems supported (GPS, etc.) Avg. # of systems TBD TBD TBD TBD Nearfield comm. (mobile payment, etc.) Share of phones 0% 62% NA NA Wireless Charging Share of phones 1% 24% 2,300% 33% Average CAGR: 20% Source: PhoneDB (https://phonedb.net/), a phone specifications database. N=15,544 phone models. In addition to improved performance, the breadth of capabilities in handsets has steadily increased as new functions have been added, including WIFI for internet connectivity, GPS for navigation, nearfield for mobile payment and similar applications, and wireless charging. A few of the most recent phones in the dataset (e.g., the iPhone 11 and newer) include “lidar” range detectors, mainly developed to allow facial recognition in the dark, 23 MASSIVE MODULAR ECOSYSTEMS but application developers have seized on this new capability to develop 3D scanning and imaging and other novel applications. In sum, the complexity, capabilities, and pace of performance improvements in the mobile handsets industry are nothing short of astonishing, with features and performance increasing at an annual rate of 20% from 2009 to 2020 across the board. At the same time, there has been little change in the size and form of touch screen smartphones, so additional functionality has been achieved with greater component and circuit density, requiring creative use of very limited space by handset design engineers.25 High-end mobile handsets are, in some cases, able to stand in for professional equipment (e.g., video cameras, navigation devices, and 3D scanners) that have previously been both bulky and cost prohibitive for average users. Because they can connect to the internet, both via wireless networks and WIFI, mobile handsets can run web applications remotely and link users to the world of data and media available on-line, including web search and social media. It is not an overstatement to say that internet-connected mobile handsets have changed how billions of people conduct their professional, political, and personal lives. Despite the complexity of mobile handsets, it is evidently possible to design and produce them in huge volumes of more than one billion units per year (see Figure 4), in large part due to extensive modularity in the multilayered component ecosystems underlying the industry. As we argue, the accommodation of complexity at scale is a signature feature of massive modular ecosystems. Market concentration and fragmentation The market structure of an MME combines both increasing concentration and fragmentation, in which the MME as a whole is segmented into a series of oligopolistic markets rather than one, the division of the overall system into specialized market segments works to diffuse the market power of even the most dominant actor, and the proliferation of platforms at multiple levels invites the participation of complementors. In this section, we illustrate how these trends are reflected in the handset and subsystem levels of the smartphone industry. Handsets At the level of the handset, the shift to a modular industry architecture that accompanied the emergence of the touchscreen smartphone led to high levels of market concentration in the premium segment, as Apple and Samsung gained dominant market shares and Huawei eventually made inroads. The pivot point in the industry centered on the period 2010-2011, about three years after the introduction of the iPhone and Android (see Figure 25 For example, as just mentioned, each type of wireless connection, and often each standard or frequency band, must have its own antenna embedded within the handset. Finding places for these that do not cause interference is a very difficult engineering challenge. This type of problem is multiplied across the handset in the layout of componentry, and while encapsulation of functions within the CPU (most CPUs have now encapsulated WIFI functions for example, eliminating the need for a dedicated chip) work in the direction of simplifying system integration, the accretion of new functions and pressure to accommodate larger batteries and more cameras keeps the pressure on. 24 MASSIVE MODULAR ECOSYSTEMS 6). In 2010, the dominant player, Sweden’s Nokia, along with Canada’s Blackberry and Taiwan, China’s HTC, still accounted for nearly 64% of keypad-based smartphone sales.26 Touchscreen smartphones running iOS and Android began to replace both these traditional smartphones as well as mass-market feature phones, or “flip phones” which dominated the consumer segment, as shown in Figure 4. By 2013, fixed key smartphones and feature phones were all but eliminated from the market. Obviously, the speed of this transition is remarkable. Apple and Samsung dominated sales of touchscreen smartphones based on the depth of product capabilities and, in the case of Apple, powerful network effects, but outside of the premium segment, modularity led to increasing fragmentation of handset market share. The modular architecture of smartphones, and the advent of highly functional systems-on-chip (SOC) CPUs and associated “reference design” resources carried over from the feature phone industry lowered barriers to entry and enabled an effervescence of low-cost smartphone handsets worldwide, but especially in China, where market growth was strongest. This can be proxied by the rise the “other major Chinese brands” and “others” categories in Figure 6, which rose to a peak of 1.46 billion units in 2017, representing nearly 46% of the market. In 2020, the latest year for which data is available, the smaller brands making up the “other category” still represented more than a third of smartphone shipments.27 Figure 6. “Smartphone” shipments by brand and geography of ownership, 2007 – 2020 Source: Adapted from Statista based on data from IDC and Gartner (2007) Note: this chart excludes feature phones. 26 Before the first iPhone was released in 2007, fixed key “smartphones” were a niche product aimed mainly at executives and engineers that needed a mobile computing platform powerful enough to run spreadsheets and other business applications while travelling. 27 Imai and Shiu (2010) and Brandt and Thun (2011) document the same pattern in China in the mid-2000s with the introduction of low cost “system-on-chip” CPUs and reference designs offered by Taiwan, China’s MediaTEK intended for feature phones. However, since 2014, several Chinese brands have consolidated their positions, mainly in the Chinese market, including Xiaomi, and the two brands from BBK (OPPO and Vivo). These brands all run Android and have taken market share mainly from Samsung. 25 MASSIVE MODULAR ECOSYSTEMS Sub-systems As the capabilities of mobile phones have increased, rising performance requirements in each major subsystem have narrowed the number of players able to compete successfully in each segment, and the most powerful CPU firms have steadily encapsulated adjacent functions, driving concentration further. The result is a general trend toward market concentration at the sub-system level of the MME, especially for the most technologically demanding and capital-intensive inputs. To demonstrate this, we show data in Figure 7 for the most important components and subsystems found in a basket of 468 phones included in the “teardown” reports published by IHS Markit. Measured by firm-level contributions to the cost of four key subsystems – displays, CPUs, memory, and telecom network connectivity – the data show a general trend toward higher market concentration. In three of these subsystems (displays, CPUs, and memory), the concentration levels found in the dataset have been consistently well above the U.S. Department of Justice’s Horizontal Mergers Guidelines, which considers 0.25 to be a ‘highly concentrated market’, as shown in the upper portion of Figure 9. Figure 7. Rising concentration at the mobile handset sub-system level, 2008-2019 Hirshman-Hirfindhal concentration index 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 U.S. Department of Justice Horizontal Mergers Guidelines considers 0.25 a “highly concentrated market” 0 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Number of suppliers 25 20 15 10 5 0 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Source: IHS Markit based on teardown reports of 456 handsets (average 38 reports per year) 26 MASSIVE MODULAR ECOSYSTEMS Displays, while not hugely complex from a circuit design point of view, are very exacting and capital-intensive to produce, and the small size and rising functionality of mobile handsets demand very high-quality displays that can, so far, only be produced by industry leaders. Samsung Display’s share of costs in this basket of phones, which is somewhat biased toward the high end, increased from 48% in 2010 to 79% in 2019, and the only other significant supplier in 2019 was another Korean firm, LG, with a 9% share. Similar to displays, memory modules do not have hugely complex circuitry, but require huge capital expenditures and have extreme requirements for precision in manufacturing. As in the case of displays, the market for mobile handset memory is dominated by two Korean firms, Samsung and SK Hynix, with 61% and 18% cost share in 2019, respectively. CPUs for mobile handsets have immensely complex circuitry, with upwards of 10 billion micro-components covering a growing range of functions, as discussed earlier. In addition, the requirements for power conservation are extreme since the CPU (along with the display) consume the vast majority of a handset’s battery power. Most mobile handset CPUs are designed by fabless chip design firms, a segment that is dominated by Qualcomm (USA), which had a 69% cost share in 2019. Also present in the data set are the CPUs designed in-house by Huawei’s HiSilicon division (13%), Samsung (12%), and Apple (4%). The design architecture in more than 90% of the world’s mobile phone CPUs relies on UK’s ARM and its ecosystem of complementors. CPUs are generally produced by chip “foundries” – contract manufacturers – mainly located in , China, especially for advanced smartphone chips. Foundries are able to amortize their manufacturing investments across a large number of fabless design firms (Sturgeon, 2002). The market for modules and components in the network connectivity category are somewhat less concentrated, as shown in Figure 7. Network connectivity is centered in a handset’s radio frequency (RF) module, often referred to as a “front-end module” in this industry since it is the first interface between the handset and the analog radio signals broadcast to and from the cell tower. RF modules have historically been assembled on exotic ceramic substrates that are notoriously difficult to manufacture and assemble, but since manufacturing processes do not require extremely high capital investments, this segment of the MME is less concentrated than CPUs, displays, or memory. Still, front- end module expertise for leading-edge phones resides in a handful of firms worldwide, most notably Japan’s Murata (8% of segment costs in 2019), and two U.S.-based firms, Qorvo (8%) and Broadcom (5%), although Samsung (6%) and Huawei (4%) also produce RF modules.28 28 Another reason that the market for RF modules may be less concentrated is because they contain several types of radio frequency devices and firms specialize in different types. In some cases, rival firms combine, or “package” their devices together into one module. For exam ple, one firm (e.g. Murata) might design the module and include its putative rivals’ devices in the package. This lack of concentration is partly a function of the complexity of analog because in addition to analog processing and adjacent digital processing, many specialized “filters” are needed to be used to distinguish signal from noise and these are produced by a variety of firms. 27 MASSIVE MODULAR ECOSYSTEMS Geographic clustering and dispersion Market concentration and fragmentation in the handset industry are expressed geographically as clustering and dispersion, respectively. In this section, we analyze geographic ownership patterns at several levels of the mobile handset MME: the system level (i.e. handsets and their assembly), sub-systems, OS development and compatible apps, and network standards. Handsets Figure 6 shows how the switch to a fully modular ecosystem with the advent of the touchscreen smartphone and app stores for iOS and Android, led to the rise of smartphone brands such as Apple (USA) and Samsung (Republic of Korea). When combined with the large “other” category of Chinese brands, we can see a certain amount of geographic diversity. However, final assembly is heavily concentrated in China. As Figure 10 shows, China has been a dominant player in the export assembly of mobile phones for many years, and this dominance has increased as European incumbents lost market share. By 2015, China’s share of global exports was close to 70%, a position that it has maintained through 2019. Such extreme geographic concentration creates risks – political, supply chain, and exchange rate – leading some firms reliant on production in China to adopt a strategy sometimes referred to as a “China +1”, meant to diversify risk and seek lower costs as labor rates began to rise in China. Most notable are Samsung’s massive investments in Vietnam (Sturgeon and Zylberberg, 2016), which drove up that country’s share of handset export value to 11% of the world total in 2015 and 16% by 2019, mostly at the expense of exports from the Republic of Korea. Figure 8. Top mobile handset export shares and values, 2007-2019, US$ billions Notes: China includes Hong Kong SAR, China-based on the assumption that most handsets exported from Hong Kong SAR, China are imported from the Mainland and re-exported, even though they are not reported as such. Export values are calculated by summing imports from all trade partners of each reporter. Source: UN Comtrade, HS 851712 (Telephones for cellular networks or for other wireless networks) Nevertheless, China’s dominance in mobile handset production is very difficult if not impossible to mitigate, at least over the short term. The operations of giant foreign contract manufacturers such as Foxconn (Taiwan, China) and Flex (USA) in China and the development of local ecosystems have nurtured a unique, highly efficient and effective “fast manufacturing” and “maker” ecosystem, especially in Southern China, that 28 MASSIVE MODULAR ECOSYSTEMS supports the production of ICT hardware in high volumes – with great responsiveness and product variety – for both export and domestic markets (Lindtner et al. 2015). Even when firms have developed alternatives – such as Samsung in Vietnam – they often remain heavily reliant on imported parts from China and direct investments from Chinese suppliers (Sturgeon and Zylberberg 2016). Sub-systems Overall, the pattern of geographic specialization at the sub-system level is also pronounced and rising, as summarized by Figure 9, which relies on the same dataset as Figure 7 with a focus on cost shares in handsets according to the headquarters country of major sub-systems suppliers. Figure 9. Mobile phone component cost share by sub-system and country, 2010 and 2019 100% 100% Source: IHS Markit based on teardown reports of 456 handsets (average 38 reports per year). Place percentages in labels here and in figure 7 The share of handset cost by sub-system and country in Figure 9 reflects the firm-level trends depicted in Figure 7, suggesting a strong connection between firm and country- level vertical specialization. Tellingly, many key suppliers to feature phones and early smartphones have apparently left the market, as far as can be judged from the list of suppliers for the basket of 468 handsets in the dataset. For example, Texas Instruments was a key CPU supplier to Nokia and several of the most important Chinese handset producers in the 2000s. The last year the company appears in this dataset as a CPU supplier is 2013, down from 18% of CPU value in 2009. Overall, the number of CPU suppliers in the dataset fell from 17 in 2010 to 6 in 2019. Korean firms have increased their share of value added in these sub-systems, again, for the basket of handsets included in the dataset, in both displays and memory, while U.S.-based fabless semiconductor design firms have held onto their dominant position in CPUs (about three fourths of value added), and increased their value share of (RF) modules for network connectivity from 49% to 69%. The falling share of U.S. suppliers’ value added to the shrinking “other wireless” category, from 48% to 25%, may reflect the encapsulation of these functions in the CPU.29 Since all of these sub-systems are very complex and technologically 29 To provide additional detail, the total value of the five sub-systems rose from $72.65 per handset in 2010 (on average for 38 models) to $157.81 in 2019 (on average for 40 models), a period in which average retail 29 MASSIVE MODULAR ECOSYSTEMS demanding, and therefore difficult to substitute, the figure speaks to how locked in the mobile handset industry is to a global supply chain that is at once geographically clustered according to each sub-system, yet dispersed overall. The paradox of geographic clustering and dispersal is even more evident in the case of the Android operating system. As discussed above, Google retains control over its official distributions of Android, but invites outside contributions to the development of new versions through the Android Open-Source Project (AOSP). Figure 10 shows the contributions of software code accepted by Google in AOSP from the beginning of Android in 2008 to 2020. The figure reveals a strong pattern of geographic clustering, with more than half of companies headquartered in the United States.30 European- headquartered companies have contributed only 10%, and countries in other regions have contributed a negligible number, with China contributing less than 1%. However, contributions made by individual engineers with non-corporate email addresses have also played a key role in advancing open-source software, contributing nearly 19% of the code commits over the life of Android, and these communities may well be populated by engineers from many countries (Shiu, forthcoming). Figure 10. Contributions (software code “commits”) to Google’s Android Open-Source Project (about 10 million since 2008) Note: ‘Virtual’ includes private individuals and individuals who contribute to open -source organizations, but who do not have a formal employment relationship with the organization, such as Linux contributors. Source: Courtesy of Jing-ming Shiu at National Cheng Kung University, , China. Data scraped from Android Open-Source Project website (https://source.android.com/) While U.S.-based firms, including Apple, which develops iOS in-house, and Google, dominate the development smartphone OS, the geographic dispersal of app development prices have also increased for high-end handsets. In addition, the mix of sub-system value added has changed, as shown by the percentages in the labels in Figure 9. The share of value coming from displays dropped from 39% in 2010 to 26% in 2019, while the share from core memory increased from 18% to 33%. The relative value added of network connectivity sub-systems also rose, from 8% to 13%, while the value from other wireless components fell from 7% to just 2%, a trend that is likely due to the encapsulation of these functions in the CPU. 30 Of course, this code may be written by software engineers located anywhere in the world, in either affiliates of the companies listed or by contractors. Firm affiliation, and by extension, parent country designation, was determined by email addresses scraped from AOSP folders. 30 MASSIVE MODULAR ECOSYSTEMS is striking. As Table 3 shows, only 33.5% of new app releases in 2017 – for both the Android and iOS platforms – were in the United States. China, which is nearly non- existent in the dataset on Android code commits, is second in app releases with 15.9%. Here we see demand from large countries and low barriers to entry opening the door to many thousands of app developers, large and small. Table 3. App releases by country, 2017 Releases in 2017 Share of releases United States 1,206,000 33.50% China 572,400 15.90% India 183,600 5.10% United Kingdom 118,800 3.30% Brazil 100,800 2.80% Germany 100,800 2.80% Japan 86,400 2.40% France 75,600 2.10% Russian Federation 75,600 2.10% Canada 68,400 1.90% Source: Statista based on data from AppFigures Telecom standard-setting A brief examination of how interconnect standards are created and evolve is key to understanding how the mobile handset industry has emerged as an MME. The mobile telecom industry consists of three primary bundles of products and services, mobile handsets (and other devices that connect to mobile telecom networks), equipment (network infrastructure), and mobile service provision (network operators). How these devices and systems interconnect is governed in large part by an evolving set of de jure radio interconnection standards established by the International Telecommunication Union (ITU), an intergovernmental organization founded by the United Nations. The ITU sets key performance requirements for each standard generation (i.e., 1G, 2G, 3G, 4G, and 5G). A consortia of industry actors creates various Standard Setting Organizations (SSOs) that develop the technical standards to meet these requirements. The standard requirements set by the ITU and worked out by SSOs cross many layers of the mobile telecom MME, from component firms like Qualcomm, to OEMs like Apple, to telecommunication equipment makers like Ericsson, to network service providers like Vodaphone. All of these categories of companies propose, refine and vote on specifications, and then once finalized, implement them. Since 1998, the dominant industry consortium has been the 3rd Generation Partnership Project (3GPP), which in 2021, had 719 member companies. As the name suggest, this SSO rose to prominence as the 3G interconnect standard was being created31 The process of creating an interconnect standard is complex: by one count, between 2005 and 2014, 3GPP considered and debated over 300,000 contributions proffered by 492 parent companies (not subsidiaries) from all over the world, which, after a long process 31 Technically, 3GPP does not create standards. It creates technical specifications which are then passed onto the seven telecom associations to turn into standards in their respective markets (Europe, US, China, India, Japan, and Republic of Korea). 31 MASSIVE MODULAR ECOSYSTEMS of evaluation, were turned into thousands of technical specifications (Baron and Gupta 2018). Despite the complexity, technical challenges, and potential for contention, the process has so far resulted in an orderly rollout of standards, as shown in Figure 11. The decreasing number of standards mentioned after each label in the figure reflects the shift from regional (e.g., CDMA for North America and GSM for Europe in the 2G era) to global coverage (two minor variants of LTE for 4G and, so far, a single global standard for 5G). Given the complexity of the technology and the need for global interoperability of mobile telecom networks, SSOs have so far proven to be a viable pathway to achieve standardization of network interconnection technologies. The strengths of SSOs are their collaborative environment and consensus-based rules, which foster an environment where engineers can directly analyze, test and debate the relative merits of different proposed engineering solutions among hundreds of competing companies, many of them intense rivals. The weakness of SSOs is their slow pace, a pace that is dictated by rules that force a high level of consensus (e.g., 3GPP rules require a 71% supermajority in order to approve a specification). Figure 11. Mobile subscriptions by network interconnect standard, 2011-2021, millions Source: Ericsson (2021) Participation in 3GPP standard setting is perhaps the most notable layer of the mobile telecom MME where China’s involvement is important and growing, showing that geographic dispersion can occur with sustained and concentrated effort. Figure 12 shows participation in mobile telecom interconnect standard setting by country from 2001 to 2020, with some of the most important companies indicated. Long dominated by European telecom equipment makers Ericsson and Nokia, China’s involvement in standard setting has gradually become more important both as a contributor of new work items (which if and when finalized become technical specifications), as well as serving as the chair of work item committees, which coordinate the procedures from proposal to finalization of work items. Thun and Sturgeon (2019) show how the Chinese state and 32 MASSIVE MODULAR ECOSYSTEMS Chinese firms tried to set a Chinese mobile telecom standard during the 3G era (TD- SCDMA), and argue that while this effort was unsuccessful in creating an exclusive standard for the country (the standard never captured a majority of the Chinese market), it did teach Chinese companies and their engineers how to work effectively within 3GPP in the 4G and 5G eras. From the data shown in Figure 12, it appears that Huawei, in particular, has made major strides as both a contributor of work items and an organizer of work groups, something that has become a point of contention with the intensification of U.S. government sanctions against Huawei beginning in 2018 (Gillis, 2019). Figure 12. Participation in mobile telecom interconnect standard setting by country (3GPP), 2001-2020 Source: Data is courtesy of Jing-ming Shiu of National Cheng Kung University, . Data scraped from 3rd Generation Partnership Project website (https://www.3gpp.org/) 5 The policy paradox The geographic dispersion of business functions across GVCs has long been a critical concern of policymakers. Developing countries have sought to attract and take advantage of GVCs to create opportunities for industrial development and upgrading (Giuliani et al., 2005; Gereffi, 1999, 2019; World Bank, 2015), while developed economies have worried about the loss of core capabilities and the economic and social dislocation created by “offshoring” (Autor et al, 2016). While the challenges posed by meeting the scale and capability requirements of MMEs have long proved to be daunting, and outcomes associated with GVC participation have proved to be mixed (Phal and Timmer, 2019), it is the geopolitical pressures they exert that have come to the fore most forcefully in recent years. These stem from the interplay between the centripetal and centrifugal forces within MMEs such as mobile telecom (Baldwin 2021). The centripetal force is associated with interdependencies within the system that cannot easily be codified. These activities are internalized within firms and clusters of firms and can even be coordinated at a distance via relational ties. On the other hand, centrifugal forces result from the need to access knowledge and capabilities that are widely dispersed, both organizationally and geographically, and the complementors to be added at the margin of modular systems. The result is a large set of interdependencies that are, by most practical measures, insurmountable: the clusters that are crucial for the critical activities and key sub-systems within an MME are by definition spatial units that reside within the jurisdictions of nation-states, but even the most 33 MASSIVE MODULAR ECOSYSTEMS dominant nations within the MME will find that not all the required clusters are within their own borders. The tension that results from participating in an MME leads to several key policy challenges for nation-states. The first is the problem of replication or substitution. The scale and complexity of MMEs means that any country or firm that seeks to control or capture an entire complex, geographically distributed industry within its borders is likely to fail.32 Self-sufficiency would require extraordinary levels of capital investment and likely entail significant additional costs from efficiency losses.33 The capital and knowledge resources needed simply outstrip what can be expected to reside in any one country. Furthermore, the specificity of industrial clusters means that they emerge only after long periods of time, and are therefore difficult to replicate, especially over the short term. Even if states opt for a more targeted approach, focusing on a few “strategic nodes,” “chokepoints,” or “bottlenecks,” the multi-layered character of MMEs, rife with inter-dependencies, means that policymakers will be hard pressed to know where, exactly, to focus their attention.34 The second is the problem of innovation. Withdrawing from the distributed and specialized capabilities that buoy innovation in any complex and knowledge-intensive global industry will almost certainly result in products and services that are several generations behind the frontier, not least because of the time it would take to build up anything close to an autarkic industry. If national firms were to be isolated from MMEs, they would risk being partitioned away from the intricate dance of innovation that takes 32 For example, China’s efforts to develop indigenous technologies are long -standing, but these efforts accelerated in the mid-2000s under the rubric of ‘indigenous innovation’ and in a new policy initiative called the Medium to Long-term Plan. The policy of import substitution in high-tech sectors became even more explicit and included substantially more funding after the global financial crisis and creation of Strategic Emerging Industries Initiative, which fed into the 12th five-year plan. By the mid-2010s, industrial policy became supercharged and combined with skyrocketing funding (through government guidance funds) for individual ICT sectors (integrated circuits, artificial intelligence), along with advanced manufacturing, most prominently, Made in China 2025, which cribbed from German ideas surrounding Industry 4.0, and targeted key manufacturing industries for upgrading through indigenous innovation.. The policy set precise targets for the replacement of foreign firms in “basic core components and important basic materials,” and publicized these initiatives to the world (Wubbeke et al. 2016). These efforts were redoubled in the wake of the (albeit temporary) shutdown of ZTE ’s production by U.S. Department of Commerce sanctions in 2018 and crippling sanctions levied by the U.S. government against Huawei in 2019. The effort to create an independent set of domestic ecosystems to support the Chinese mobile telecom and semiconductor industries, in particular, was deemed to be the equivalent to a wartime effort by policy-makers in China. They identified the key foreign firms upon which they were reliant, selected the domestic firms that were most likely to be able to provide substitutes, and channeled massive amount of investment and incentives towards these firms. 33 According to one estimate, developing self-sufficiency in semiconductors alone would require an upfront investment of US$ 0.9 trillion to 1.2 trillion, and a loss of US$ 45 billion to 125 billion in annual cost efficiencies (Varas et al. 2021) 34 Semiconductor manufacturing plants (foundries) are widely viewed as highly strategic, for example, and China, the United States, the European Union, the Republic of Korea, India, Taiwan, China, and Japan are all poised to spend tens of billions to subsidize private investments in manufacturing and R&D, but a foundry has its own complex ecosystem of inputs and complements, and re-shoring new foundries may only generate a new set of vulnerabilities in upstream and adjacent ecosystems (see Ting-Fang and Li, 2022 for details). 34 MASSIVE MODULAR ECOSYSTEMS place independently in each node, and then connected up via modular linkages to create extraordinarily complex and highly functional products and services such as mobile telecom. Even more profoundly, the entire sectoral innovation system could be disrupted and degraded if key actors and institutions are excluded or withdraw. The third is the problem of unpredictable and rapid change. The immense geographic and functional complexity of MMEs – as compared to more discrete structures in the global economy such as MNEs and GVCs – means that disruptions, such as outcomes- based trade prohibitions and exogenous geopolitical or environmental shocks to supply chains, can come suddenly and from surprising places. Additional challenges can also originate from the encapsulation of functions by adjacent sub-systems, which may or may not have a geographical dimension. If a dominant firm successfully incorporates functionality (encapsulates) controlled by one or a few adjacent, downstream, or upstream players in its own products, markets can disappear with breathtaking rapidity, as we have seen with the rise of Google’s Android OS. Policymakers may find it difficult to react with the required wisdom and speed, since the outcomes of these competitive battles can only be observed in hindsight. In this paper we have described MMEs as an ecosystem of ecosystems, significantly governed by modular business linkages enabled by industry-specific standards arising from a variety of de facto and de jure mechanisms, and have pointed to three paradoxes embodied by MMEs: complexity at scale, market concentration and fragmentation, and geographic clustering and dispersal. Of course, these three facets of MMEs are not independent phenomenon. Complexity and scale at the leading edge of systems, sub- systems, software, and componentry favors firms with high competencies that are difficult to replicate or substitute for, leading to multiple MME nodes with high market concentration, even as modularity opens the door for complementors and a more open market structure at an ecosystem’s margin. Since innovation at the leading edge often relies on relational linkages (e.g. the build-up and exchange of tacit knowledge through face-to-face contact and long-term business relationships) and the support of regionally- embedded institutions (e.g., universities, specialized labor markets and support industries, and standard setting activities), market concentration is often expressed as geographic concentration in one or a few industry clusters focused on a specific technology and market areas, even as these specializations are linked together to create complex systems, leading to geographic dispersion in the MME as a whole, and also, in some cases, in a set of complementors working at the sub-system level. When placed under stress, the fact of geographic vertical specialization in MMEs can exacerbate or even generate geopolitical tension. Given that geographic concentration, as measured by the headquarter location of firms, will usually accentuate the level of concentration in an MME layer or node – meaning that a subsystem with an oligopolistic market structure will be a monopoly at the country level when all of the leading firms are located in the same country – it is a relatively short step to interrupt or even “weaponize” critical inputs as an instrument of industrial or national security policy. Such moves, however, create strong incentives for trade partners to make countermoves to reduce vulnerabilities. 35 MASSIVE MODULAR ECOSYSTEMS While MMEs generate pressure for decoupling, there is an opposite set of pressures that arise from this same set of circumstances. Because substituting for technology-intensive products, sub-systems, or components, or the industry clusters where they arise, is very likely to be a very long-term proposition, prohibitively expensive, a source of additional, unintended vulnerabilities, or in many cases, impossible to achieve, MMEs create pressure for geopolitical accommodation and co-operation. This, we argue, is the fourth paradox embodied by MMEs, a “policy paradox”, as depicted in Figure 13: it simultaneously creates pressure for decoupling and for accommodation and co-operation, especially when the systems and products involved are essential to the functioning of modern society.35 Figure 13. Three paradoxes found in MME associated with relational (centripetal tendencies) and modular (centrifugal tendencies) coordination, leading to a policy paradox Relational coordination – centripetal tendencies Complexity Market Concentration Geographic clustering Decoupling POLICY PARADOX Scale Market fragmentation Geographic dispersal Accomodation Modular coordination – centrifugal tendencies 6 Conclusions In this paper, we have sought to make three theoretical contributions. The first is the development of a framework that captures the complexity of global industries in the digital age. To accomplish this, we combine the strengths of literatures on industrial organization, GVCs, and innovation and technology management. We shift the analysis of global industries from a “chain” metaphor, which is the basis of theory building in the GVC literature and policy-making related to global supply chains, to an “ecosystem” metaphor, based on insights from scholarship on modular and platform ecosystems. The chain metaphor identifies and categorizes the flow of value addition across borders for specific “lead” firms or products, and while this has proven to be useful for explaining and predicting the geographic structure of specific industries, it has difficulty capturing or explaining the complexities found in modern, digitally-enabled industries. 35 While MMEs are not the only way that geographic vulnerabilities can come about (the supply of critical minerals such as cobalt are geographically fixed as a result of geological accident, for example), and not all industries are characterized by MMEs, it is clear enough that the colonization of multiple industries by the ICT MME, and the standards that govern it, is a process that is well underway, and this raises their importance for both strategy and policy. 36 MASSIVE MODULAR ECOSYSTEMS Second, while we draw key insights from the literature on modular and platform ecosystems on their implications for competition, innovation, and value capture, we point out that these theory-building efforts have tended to treat modular ecosystems as discrete entities, when in fact they are linked, in an operational sense, to upstream, downstream, and adjacent ecosystems and therefore embody dynamic structures that are far larger and more complex than most of the literature has grappled with so far. Third, unlike the literature on GVCs, the ecosystem literature has so far largely ignored geography, especially the international character of modular industry ecosystems. It is assumed that firms make decisions about how to set up their platforms, foster network effects, set rules for complementors, and organize supply chains within a frictionless information space without the potential for sudden force majeure disruptions, policy interventions, or due consideration of the uneven geographic distribution of costs, skills or industrial capabilities in the global economy. The parsimony of this approach has become increasingly problematic as supply-chain disruptions and geopolitical tensions have mounted. It is important to address several points about the generalizations we expect to be drawn from this paper. First, modularity can reduce the density of relational linkages and therefore the geographic and organizational “lumpiness” of industries, but only to a degree. There will always be significant benefits to be had from vertical (re)integration, alliances, and spatial and social proximity in any industry. Because of this, we should not expect modular linkages to be the exclusive form of governance in MMEs. Relational linkages, both within and between firms, are essential, particularly for fostering innovation. Platform owners, dominant sub-system producers, and powerful service providers (e.g., of manufacturing services) regularly and systematically develop close relationships with other powerful actors in MMEs to drive the state of the art forward and provide a host of other benefits.36 Second, while this paper is meant to take a step toward framing a broader picture of modular ecosystems, our frame of reference and case study are both drawn from the ICT sector. One possible conclusion is that MMEs are an artifact of the ICT sector. This would be a reasonable conclusion given how deeply modularity has been “baked” into the sector early on, in hardware since the IBM 360, but especially in software. Another conclusion, which is not in contradiction to the first, is that, because we are seeing digitization of business processes across the broad economy, the implications of trends in the ICT sector are becoming increasingly relevant for both business strategy and industrial policy across a large and expanding number of industries.37 In addition to contributing to the theoretical work on industrial organization, GVCs, and innovation and technology management, we want to offer the analysis in the paper as a guide for research methodology. Given the intense policy focus on resolving the political 36 For example, for suppliers, relational ties help lock in important customers and, for buyers, they can speed time to market. 37 Recent research has examined the rise of platform intermediaries in metal machining and apparel, to provide just two examples (López et al. 2022, Schneidemesser and Butollo 2022). 37 MASSIVE MODULAR ECOSYSTEMS tensions (geo- and otherwise) arising from MMEs, analysis of the geographic and ownership features of MMEs in a range of “critical” industries is urgently needed. Will the shift to an outcomes-based global system lead to a decline, fragmentation, or re- alignment of MMEs, or will the dispersed yet concentrated organizational and geographic structures we see today prove to be more durable? It is too early to tell, but in any case, industrial and national security policymakers will need to grapple with the entirety of MMEs to avoid the unintended consequences that might come from narrow interventions based on incomplete conceptions of how complex global industries function (Gereffi et al, 2021). Industrial and trade policy solutions – import substitution, re-shoring, near- shoring, or friend-shoring – ignore the reality of MMEs at their peril. References Adler, P. S. (2001). "Market, hierarchy, and trust: The knowledge economy and the future of capitalism." Organization science 12(2): 215-234. Adner, R. (2017). "Ecosystem as structure: an actionable construct for strategy." Journal of Management 43(1): 39-58. Autor, D.H., Dorn, D. and Hanson, G.H., 2016. “The China shock: Learning from labor- market adjustment to large changes in trade.” Annual Review of Economics, 8, 205-240. Baldwin, C. Y. (2008). Where do transactions come from? Modularity, transactions, and the boundaries of firms. Industrial and corporate change, 17(1), 155-195. Baldwin, C. Y. (2007). "Where do transactions come from? Modularity, transactions, and the boundaries of firms." Industrial and Corporate Change 17(1): 155-195. Baldwin, C. Y. (2020). Ecosystems and Complementarity. Harvard Business School Working Paper. Boston, Harvard Business School. Baldwin, C. Y. (2021). Platform Systems vs. Step Processes—The Value of Options and the Power of Modularity. HBS Working Paper. Boston, Harvard Business School. Baldwin, C. Y. and K. B. Clark (2000). Design rules: The power of modularity. Boston, MIT Press. Berger, S. (2006). How we compete: What companies around the world are doing to make it in today's global economy. New York, Currency Doubleday. Biddle, B., et al. (2010) How Many Standards in a Laptop? (And Other Empirical Questions). Bodrožić, Z. and P. S. Adler (2017). "The Evolution of Management Models: A Neo- Schumpeterian Theory." Administrative Science Quarterly 63(1): 85-129. Bogers, M., et al. (2019). "What is an ecosystem? Incorporating 25 years of ecosystem research." Academy of Management Proceedings 1. Brandt, L. and E. Thun (2011). "Going mobile in China: shifting value chains and upgrading in the mobile telecom sector." International Journal of Technological Learning, Innovation, and Development 4(1/2/3): 148-180. Brusoni, S. (2005). "The limits to specialization: problem solving and coordination in ‘modular networks’." Organization studies 26(12): 1885-1907. Casaccia, L. (2017). "Demystifying 3GPP – An insider’s perspective to how 4G and 5G standards get created." 38 MASSIVE MODULAR ECOSYSTEMS https://www.qualcomm.com/news/onq/2017/08/demystifying-3gpp-insiders- perspective-how-4g-and-5g-standards-get-created 2022. Coase, R. H. (1937). "The nature of the firm." Economica 4(16): 386-405. Cusumano, M. A., et al. (2019). The Business of Platforms: Strategy in the Age of Digital Competition, Innovation, and Power. New York, HarperBusiness. Dallas, M. P., et al. (2019). "Power in global value chains." Review of International Political Economy, 26(4): 666-694. Dossani, R. and M. Kenney (2003). "“Lift and Shift”: Moving the back office to India." Information Technologies and International Development 1(2): 21-37. Ericsson (2021). Ericsson Mobility Report Farrell, J. and T. Simcoe (2012). Four Paths to Compatibility,” in, M. Peitz and J. Waldfogel (eds). The Oxford Handbook of the Digital Economy. Oxford, Oxford University Press. Furr, N. F., et al. (2022). "What is digital transformation? Core tensions facing established companies on the global stage." Global Strategy Journal: 1-24. Gawer, A. (2014). "Bridging differing perspectives on technological platforms: Toward an integrative framework." Research Policy 43: 1239-1249. Gawer, A. and M. A. Cusumano (2008). "How Companies Become Platform Leaders." MIT Sloan Management Review 49(2): 28-35. Gawer, A. and M. A. Cusumano (2014). "Industry platforms and ecosystem innovation." Journal of Product Innovation Management 31(3): 417-433. George Mason University Law and Economics Research Paper Series. Georrge Mason University: 1-30. Gereffi, G., 1999. International trade and industrial upgrading in the apparel commodity chain. Journal of international economics, 48:1, 37-70. Gereffi, G., 2019. Economic upgrading in global value chains. Chapter 14 in: Handbook on global value chains, UK: Edward Elgar. Gereffi, G., et al. (2005). "The governance of global value chains’." Review of International Political Economy 12(1): 78-104. Gillis, Chris 2019. “BIS advisory cautions tech firms in standards-setting bodies” American Shipper, August 21. Giuliani, E., Pietrobelli, C. and Rabellotti, R., 2005. Upgrading in global value chains: lessons from Latin American clusters. World development, 33(4), pp.549-573. Granovetter, M. (1985). "Economic Action and Social Structure: the Problem of Embeddedness." American Journal of Sociology 91: 481-510. Hazlett, T., et al. (2011). Walled garden rivalry: The creation of mobile network ecosystems. Hobday, M. (1998). "Product complexity, innovation and industrial organisation." Research Policy 26: 689-710. Imai, K. and J. M. Shiu (2010). Value chain creation and reorganization: The growth path of China's mobile phone handset industry. The dynamics of local learning in global value chains: Experiences from East Asia. M. Kawakami and T. J. Sturgeon. Houndmills, UK, Palgrave Macmillan. 39 MASSIVE MODULAR ECOSYSTEMS Jacobides, M. G., et al. (2006). "Benefiting from innovation: Value creation, value appropriation and the role of industry architectures." Research Policy 35: 1200- 1221. Jacobides, M. G., et al. (2018). "Towards a theory of ecosystems." Strategic management journal 39: 2255–2276. Kenney, M. and J. Zysman (2016). "Rise of the platfform economy." Issues in Science and Technology in Society 32(3): 61-69. Kenney, Martin and Zysman, John, 2020. “The platform economy: restructuring the space of capitalist accumulation,” Cambridge Journal of Regions, Economy and Society, 13: 1, March, 55–76. Kretschmer, T., et al. (2020). "Platform ecosystems as meta-organizations: Implications for platform strategies." Strategic Management Journal: 1-20. Lamberg, J.-A., et al. (2021). "The curse of agility: The Nokia Corporation and the loss of market dominance in mobile phones, 2003–2013." Business History 63(4): 574-605. Langlois, R. N. (2006). The secret life of mundane transaction costs. Organization Studies, 27(9), 1389-1410. Leiponen, A. E. (2008). "Competing through cooperation: The organization of standard setting in wireless telecommunications." Management Science 54(11): 1904- 1919. Lindtner, S., et al. (2015). Designed in Shenzhen: Shanzhai Manufacturing and Maker Entrepreneurs. Aarhus Series on Human Centered Computing. Aarhus. Locke, R. M. (1997). Remaking the Italian Economy. Ithace, NY, Cornell University Press. López, T., et al. (2022). Digital Value Chain Restructuring and Labour Process Transformations in the Fast-Fashion Sector: Evidence from the Value Chains of Zara & H&M. Society of Socio-economics Annual Conference. Amsterdam. Moore, J. F. (1996). The Death of Competition: Leadership & Strategy in the Age of Business Ecosystems. New York HarperBusiness. Mudambi, R. (2008). "Location, control and innovation in knowledge intensive industries." Journal of Economic Geography 8: 699-725. Murmann, J. P. and K. Frenken (2006). "Toward a systematic framework for research on dominant designs, technological innovations, and industrial change." Research Policy 35: 925–952. Nadvi, K. (2008). "Global standards, global governance and the organization of global value chains." Journal of Economic Geography 8(3): 323-343. Pahl, S. and Timmer, M. (2019). ‘Do Global Value Chains Enhance Economic Upgrading? A Long View,’ Journal of Development Studies, Published online Dec 26. Park, Y. and K. Ogawa (2009). "The impact of product architecture on supply chain integration: a case study of Nokia and Texas Instruments." International Journal Services and Operations Management 5(6): 787-798. Ponte, S. and P. Gibbon (2005). "Quality Standards, Conventions and the Governance of Global Value Chains." Economy and Society 34(1): 1-31. 40 MASSIVE MODULAR ECOSYSTEMS Ponte, S. and T. Sturgeon (2013). "Explaining Governance in Global Value Chains: Towards a Broader Analytical Framework." Review of International Political Economy 21(1): 195-223. Powell, W. (1990). "Neither Market Nor Hierarchy: Network Forms of Organization Research." Organizational Behavior 12: 295-336. Prahalad, C. K. and G. Hamel (1999). The Core Competence of the Corporation. Knowledge and Strategy. M. H. Zack, Routledge. Prencipe, A., et al. (2003). The Business of Systems Integration. Oxford, Oxford University Press. Rodrik, D. and S. Walt (2021). How to construct a new global order. HKS Faculty Research Working Paper Series, Harvard Kennedy School: 1-37. Sanchez, R. and J. T. Mahoney (1996). "Modularity, flexibility, and knowledge management in product and organization design." Strategic management journal 17(S2): 63-76. Saxenian, A. (1996). Regional Advantage: Culture and Competition in Silicon Valley and Route 128. Cambridge, MA, Harvard University Press. Schneidemesser, L. and F. Butollo (2022). Distribution-Centred Digital Transformation – New Business Models and Spatial Reorganization of Production and Work in the Mechanical Component Industry. Society of Socio-economics Annual Conference. Amsterdam. Scott, A. J. (1988). Metropolis: From the Division of Labor to Urban Form. Berkeley, University of California Press. Shaikha, M. and N. Levina (2019). "Selecting an open innovation community as an alliance partner: Looking for healthy communities and ecosystems." Research Policy 48. Shiu, J.M., M.P. Dallas and H.H. Huang (2022) “A Friend of a Friend? Social Capital and Networks in Telecommunications Standard-setting Organizations” (Under Review). Shiu, J.M., M.P. Dallas and Z.R. Wang (2022) “Varieties of Generational Technology Transitions: The Android Platform, Complementors and Sequential Ambidexterity in Organizational Learning” (Under Review) Simon, H. A. (1962). "The Architecture of Complexity." Proceedings of the American Philosophical Society 106(6): 467-482. Storper, M. (1995). "The Resurgence of Regional Economies, Ten Years Later: The Region as a Nexus of Untraded Interdependencies." European Urban and Regional Studies 2(3): 191-221. Sturgeon, T. and E. Zylberberg (2016). The Global Information and Communications Technology Industry: Where Vietnam Fits in Global Value Chains. Policy Research Working Paper. Washington, DC, World Bank. Sturgeon, T. and Lester, R. (2004) ‘The new global supply-base: New challenges for local suppliers in East Asia’, in: S. Yusuf, A. Altaf and K. Nabeshima (eds.) Global Production Networking and Technological Change in East Asia, Oxford: Oxford University Press. Sturgeon, T. J. (2002). "Modular production networks: a new American model of industrial organization." Industrial and Corporate Change 11(3): 451-496. 41 MASSIVE MODULAR ECOSYSTEMS Sturgeon, T. J. (2008). From commodity chains to value chains: Interdisciplinary theory building in an age of globalization. Frontiers of commodity chain research. J. Bair. Stanford, Stanford University Press. Teece, D. J. (2018). "Profiting from innovation in the digital economy: Enabling technologies, standards, and licensing models in the wireless world." Research Policy 47:18, 1367-1387. Thun, E. (2006). Changing lanes in China: Foreign direct investment, local governments, and auto sector development. New York, Cambridge University Press. Thun, E. and T. J. Sturgeon (2019). “When global technology meets local standards: Reassessing China’s communications policy in the age of platform innovation,” in L. Brandt and T. G. Rawski. (eds), Policy, regulation and innovation in China’s electricity and telecom industries, Cambridge, Cambridge University Press. Timmer, Marcel; Miroudot, Sébastien; and de Vries, Gaaitzen, 2019. “Functional specialisation in trade,” Journal of Economic Geography, 19:1, 1–30. Ting-Fang, Cheng and Li, Lauly, 2022. “The resilience myth: Fatal flaws in the push to secure chip supply chains.” The Big Story, Nikki-Asia, July 27. Turkina, Ekaterina and Van Assche, Ari, 2018, “Global connectedness and local innovation in industrial clusters.” Journal of International Business Studies 49, 706–728. Turkina, Ekaterina; Van Assche, Ari; Kali, Raja, 2016. “Structure and evolution of global cluster networks: evidence from the aerospace industry,” Journal of Economic Geography, 16:6, 1211–1234 Tushman, M. L. and J. P. Murmann (1998). "Dominant Designs, Technology Cycles, and Organization Outcomes." Academy of Management Proceedings 1998(1). Van Alstyne, M. W., et al. (2016). "Pipelines, Platforms, and the New Rules of Strategy." Varas, A., et al. (2021). Strengthening the Global Semiconductor Supply Chain in an Uncertain Era, Boston Consulting Group and Semiconductor Industry Association. World Bank (2015), “Global Value Chains,” Brief, IBRD/IDA, July 16. Wübbeke, J., Meissner, M., Zenglein, M.J., Ives, J. and Conrad, B., 2016. Made in China 2025,” Mercator Institute for China Studies. Papers on China, 2, 74. 42 MASSIVE MODULAR ECOSYSTEMS Appendix – Data sources The IHS Markit product portfolio includes assessments of all electronic, electro- mechanical and mechanical components inside each mobile device, i.e. the teardowns. IHS purposively samples for flagship brands or product families (for e.g., Apple iPhones) as well as handsets produced by new or market gaining manufacturers, and handsets requested by clients. This data includes the BOM and a pricing model developed by IHS industry experts used to calculate component unit costs as well as the total manufacturing costs (hand assembly, auto insertion, test costs) of each device. Components are categorized by function (e.g. analog baseband, apps processing, battery and power management, etc.), by which sub-assembly it appears (e.g. printed circuit board, camera module, display module), and by component family and type (accessory, assembly, battery, electro-mechanical component, acoustics, etc). For each component, IHS reports the dimensions of the the package and - whenever possible - the name of component vendor. On average, vendor information is available for 21% of the components in each model, including all but a few makers of higher-value components. The list of 1108 unique vendors was enriched with the country of vendor headquarters and - when relevant - the headquarters country of the parent company. To truly capture the time dimension and evolution of manufacturer-supplier relationships, a better understanding of joint ventures and of mergers and acquisitions over time is needed. PhoneDB is an online data source that contains information on individual phone model specifications and contains 15,544 phone models with information on the (approximate, since price can vary by geography and time of sale) retail price, release date of the model, network technology and many other features not exploited in the current analysis. We were able to match 516 devices in the IHS teardown dataset to this data set. The Android Open-Source Project contains every open-source commit (software contribution) submitted and approved to be included into AOSP, which serves as the foundation of the Android operating system. Our dataset includes important metadata on every commit (e.g. commit date, email and organization of contributor, Android release) and the ‘location’ of the commit within the nested folders of AOSP, which indicates the purpose of the commit. Additional information also indicates some details about the substance of the commit itself. Our dataset included 11,432,275 commits, mostly from 2008 to 2020, of which we were able to identify the headquarter firm for 86% of the commits. 3GPP is the primary standard-setting organization for the creation and finalization of technical specifications for the mobile telecommunications industry. After finalization, technical specifications are turned into formal standards by the major telecom business associations in China, Europe, India, Japan, Republic of Korea and North America. For transparency, 3GPP publishes a wide range of information on conference meetings, personnel, and each stage of development of the specification, from initial proposal to final publication, including the individuals and organizations involved. Our dataset contained 16,189 work items between 2001 and 2020, which contributed to the finalization of nearly 2,500 3GPP technical specifications. 43 MASSIVE MODULAR ECOSYSTEMS Thus, the descriptive statistics presented in this paper cover a wide range of phone models. A dataset complete with all collected variables is available for a subset of 602 models released between 2002 and 2020. This baseline dataset includes the bill of materials, with data on the various components used in each device and their unit costs, total direct material costs, manufacturing costs and labor wage assumptions - as well as device-level specifications like the Operating System (OS), the retail price, and the headquarter location of the producer of each smartphone and of each branded component. The time dimension in the data-set is constructed by using information on the release date of individual handsets. This choice is informed by how the industry works. Handset makers cannot swap out complex parts in a handset without major redesign work. There are too many interdependencies. A redesign to swap out a CPU would take about 6 months. Given this, it is unlikely that a phone model will be redesigned to use different components over the lifespan of the product. It is more likely that a company would simply design a new handset, which is often done once a year for most major brands. It is not only design engineering that makes it hard to swap components; testing is also a costly and time-consuming activity that generate stickiness to the initial choice of design and inputs of both phones and individual complex components.38 38 Test engineers throw every imaginable use case at a device/system to make sure it works under all conditions and rarely occurring scenarios (“corner cases”) might take a long time to track down and adjust for. This includes connections to various external devices and systems, as well as various hardware and software “calls” on the device to do this or that. This is especially true for high -value brands like Apple that do not want to have their flagship phones fail in the field for any reason. 44