
Four issues a year, free to read, since 2012
Network Law Review
The complex science of markets & digital laws
Article
Measuring Dynamic Competition in Digital Markets: Artificial Intelligence, Generative AI and the Evolution of Market Power
- a Japan Fair Trade Commission and Hitotsubashi University
Article info
- Published
- 18 November 2026
- Issue
- Fall 2026, pages 44–52
- Cite as
- Aoki, R. (2026, Fall). Measuring dynamic competition in digital markets: Artificial intelligence, generative AI and the evolution of market power. Network Law Review, 44–52. https://
www .networklawreview .org/ measuring-dynamic-competition/
Abstract
We first characterize digital markets as markets of heterogeneous goods and services, with complicated supply chains, using the newest and most dynamic digital markets, AI and GenAI, as a reference point. We identify two important factors, network effects and economies of scale, common in digital markets and at all levels of the AI stack such as data and foundation models and two important dynamic competition concerns, contestability and foreclosure/entry deterrence. Based on traditional static markup and foundation model dynamics, we propose a dashboard approach to assessing dynamic competition. The dashboard has indices: market power, entry and scaling, innovation, market mobility, switching and access to strategic inputs. Weights on the indices differ across technologies and over time.1 I am very grateful to Thibault Schrepel for his insights and comments. The author used ChatGPT to assist in preparing this article.
1. Introduction
Digital markets have transformed both the nature of competition and the problem of measuring it. Traditional competition analysis developed largely around markets in which products, prices and market boundaries were reasonably identifiable. Competition could consequently be assessed through market shares, concentration, prices, output and profitability.
Digital markets are more complicated. Firms frequently compete through technological improvement, product quality, data, platforms and ecosystems rather than price alone. The same firm may operate in several related markets and use technologies, data and infrastructure developed in one market to compete in another. Moreover, market positions can change as new technologies emerge.
The distinction between static and dynamic competition may be difficult to make. Static competition concerns the competitive conditions surrounding existing products at a given time. Dynamic competition concerns how firms compete for future market positions through research and development, investment, entry, technological improvement and innovation.
In markets characterized by technological change, a high current market share does not necessarily imply an absence of competition. A firm may dominate because it has developed a substantially better product, while simultaneously facing the possibility that a new technology will displace it. Conversely, the presence of several firms does not necessarily imply strong competition if entrants cannot acquire the resources necessary to expand or if incumbents can foreclose new competitors.
The important question is not simply how much market power a firm possesses today, but how durable that market power is and whether competitors can challenge it.
Neither high concentration nor high profitability is in itself evidence of anticompetitive conduct. Successful innovation should generate returns. Indeed, eliminating the possibility of earning such returns could weaken incentives for research and development. The competition concern arises when firms can protect market power from future competitive challenges through foreclosure, raising rivals’ costs, control over essential inputs or other mechanisms that reduce contestability.
GenAI makes this issue particularly important. AI can lower the costs of producing software, information and other services and thereby facilitate entry. At the same time, developing advanced AI may require substantial access to computing capacity, specialized semiconductors, data, energy, capital and distribution. AI can therefore simultaneously strengthen competition in some markets and increase barriers in others.
The remainder of the paper proceeds as follows. Section 2 describes the characteristics of digital markets, including the emerging GenAI stack. Section 3 examines contestability, foreclosure and entry deterrence. Section 4 considers conventional and dynamic measures of competition. Section 5 discusses implications for AI and GenAI, and Section 6 concludes.
2. Characteristics of Digital Markets
2.1 Diversity of goods, services, technologies and firms
The term “digital market” encompasses a very wide variety of goods and services. These include semiconductors and digital hardware, operating systems, software, search, online retail, digital advertising, cloud computing, communications services and, increasingly, artificial intelligence.
The relationships among these products are complicated.2 Crémer, Jacques, Yves-Alexandre de Montjoye, and Heike Schweitzer. 2019. Competition Policy for the Digital Era. Brussels: European Commission. Hardware and software are frequently complementary. Some firms provide only particular components, while others provide integrated combinations of hardware, software and services. Firms competing in one product market may therefore differ considerably in overall size and scope.
Consequently, even where a reasonably narrow product market can be defined, firms within that market may have very different competitive resources. A firm operating in several adjacent markets may possess data, distribution or technological assets unavailable to a specialized rival. These differences are relevant when assessing whether current market shares accurately describe competitive constraints.
Digital markets also encompass technologies of different vintages. The commercial internet and many associated technologies are now mature. Other technologies are being incrementally improved. AI, and GenAI in particular, remains in a period of unusually rapid development and adoption.3 Schrepel, Thibault, and Alex ‘Sandy’ Pentland. 2025. “Competition between AI Foundation Models: Dynamics and Policy Recommendations.” Industrial and Corporate Change 34 (5): 1085–1103. Competition analysis therefore needs to accommodate mature technologies and emerging technologies within the same broader digital ecosystem.
2.2 Technological change and dynamic competition
Technological change is central to competition in digital markets. Firms compete by developing new technologies, improving existing products and finding new combinations of technologies. R&D and investment may therefore be more important competitive variables than short-run price changes.
We use dynamic competition to describe competition in technologies and products requiring investment and innovation over time. This does not mean that price competition is irrelevant. Rather, price is only one dimension of competition, and in some digital markets it may be considerably less informative than technological progress.
AI and GenAI provide a prime example of dynamic competition. The relationship between levels of the AI stack is not homogeneous and each level has heterogeneous components.4 Schrepel, Thibault, and Alex ‘Sandy’ Pentland. 2025, supra Note 3. For instance, data can be public or proprietary, foundation models can be public or individualized, depending on users, and weights can be open or closed. Not to mention different relationships to the digital ecosystem, such as Gemini to search and Copilot to the operating system.
Dynamic competition can take several forms. An entrant can introduce a new product that substitutes for an incumbent product, as Llama and Mistral did. An existing firm can introduce a technology that creates an entirely new market as with the introduction of Claude. Alternatively, a new technology can change competition in an existing market by lowering production costs or allowing firms to offer new services as DeepSeek did.
3. Competition Concerns in Digital Markets
3.1 Contestability
The principal concern in rapidly changing digital markets should be contestability: whether existing market positions can be challenged.
For measurement purposes, dynamic competition should therefore encompass entry and exit, innovation, technological displacement, market-share mobility, switching and access to the resources required to compete.5 Katz, Michael L., and Carl Shapiro. 1985. “Network Externalities, Competition, and Compatibility.” American Economic Review 75 (3): 424–440. These variables are complements rather than substitutes for conventional concentration measures.
This approach also helps distinguish temporary technological leadership from durable market power. A firm may retain a high market share because it continuously innovates. That situation differs economically from one in which the firm retains the same share because rivals cannot enter or obtain necessary inputs.
This distinction is particularly important when there are increasing returns and network effects, as in GenAI.
Increasing returns are not themselves anticompetitive, but they matter for dynamic competition because advantages that initially arise from innovation can become increasingly difficult for rivals to challenge.6 Farrell, Joseph, and Garth Saloner. 1986. “Installed Base and Compatibility: Innovation, Product Preannouncements, and Predation.” American Economic Review 76 (5): 940–955.
Network effects provide a related mechanism. Where the value or quality of a service improves with adoption and use, successful firms may develop advantages that new entrants cannot readily reproduce. Network effects may benefit consumers while simultaneously increasing the minimum scale required for effective competition.7 Rochet, Jean-Charles, and Jean Tirole. 2003. “Platform Competition in Two-Sided Markets.” Journal of the European Economic Association 1 (4): 990–1029.
The competitive implications of data are similarly ambiguous.8 Shapiro, Carl, and Hal R. Varian. 1999. Information Rules: A Strategic Guide to the Network Economy. Boston: Harvard Business School Press. Data may improve model quality, but the amount of data is not necessarily determinative. Improvements in algorithms, training methods and the use of higher-quality datasets may allow smaller models or firms with fewer resources to compete effectively.
This observation has an important implication for competition measurement. A static comparison of the quantities of data or compute controlled by different firms may exaggerate or understate competitive advantages. What matters is whether those inputs are necessary for effective competition at the technological frontier and whether alternative combinations of inputs and innovation remain commercially viable.
Dynamic competition analysis should therefore investigate both increasing returns and countervailing technological change. The relevant question is not simply whether the leading firm has more users, data or compute, but whether those advantages make its position increasingly difficult to contest.
3.2 Foreclosure and entry deterrence
Foreclosure is particularly important because digital firms frequently participate in vertically related markets. A firm may simultaneously provide infrastructure to rivals and compete with those rivals downstream.
Vertical integration itself is not anticompetitive. It can reduce transaction costs, facilitate investment and improve coordination. The relevant concern is whether control of an upstream product or platform allows a firm to disadvantage competitors in a related market.
The underlying mechanism is raising rivals’ costs. An incumbent controlling an important input may worsen the price, quality or availability of that input to competitors. In AI markets, potentially important inputs include advanced semiconductors, computing capacity, cloud services, data and access to distribution.
The appropriate measure is therefore not simply how much of an input the incumbent owns. Analysis should compare the cost and availability of the input to incumbents and rivals and ask whether an efficient rival can obtain a commercially viable alternative. Availability of which Nvidia chip motivated DeepSeek.
Entry deterrence raises similar issues. Digital technologies frequently require substantial fixed investments. Large expenditures on data centers, R&D, data acquisition or network development can make entry difficult. But fixed costs themselves are not evidence of anticompetitive conduct. They may simply reflect the underlying technology.
Competition analysis needs to distinguish technological entry barriers from barriers strategically created or reinforced by incumbent conduct.
4. Measuring Competition and Identifying Concerns
4.1 Traditional Measures
CR4, HHI and market shares
Traditional concentration measures remain useful starting points. The four-firm concentration ratio and HHI summarize the distribution of market shares and can identify markets in which a small number of firms account for most activity.
Their limitations in digital markets are largely extensions of familiar limitations in conventional markets. Two firms can compete aggressively with one another, while a market containing more firms can exhibit weak competition. High concentration therefore does not establish anticompetitive conduct, just as low concentration does not rule it out.
There is an additional practical problem: calculating concentration requires both a definition of the relevant market and an appropriate measure of market activity. In some digital markets, sales provide a reasonable denominator. In others, revenue may be misleading because services are provided without a monetary charge or because revenues arise on another side of a platform.
Possible measures include users, transactions, searches, clicks, views or time spent. For AI, possible measures include downloads, API calls, tokens, computing consumption or adoption by developers. None should automatically be regarded as the complete or correct measure. Available data usually do not cover the whole market, even when “market” is well-defined.9 Thomas, Stefan, 2026. “Market Share Analysis for Closed-Weight and Open-Weight AI Models”, this volume. The choice should follow from the economic activity and competitive mechanism under investigation.
Markups, Production Functions and Market Power
Market shares and concentration describe market structure but do not directly measure the ability of firms to price above marginal cost. Markups therefore provide an important complementary indicator of market power. The markup of firm at period is:
where is price and is marginal cost.
A markup greater than one indicates that price exceeds marginal cost. The practical difficulty is that marginal cost is rarely observed in firm-level accounting data. This is particularly important when competition is examined across large numbers of firms or industries, where the detailed price, quantity and cost information required for conventional demand-based estimation is generally unavailable.
One way to overcome this problem is to estimate markups from firms’ production decisions rather than directly observing marginal costs.10 De Loecker, Jan, Jan Eeckhout, and Gabriel Unger. 2020. “The Rise of Market Power and the Macroeconomic Implications.” Quarterly Journal of Economics 135 (2): 561–644 Under the firm’s cost-minimization condition for a flexible input (V), the firm-level markup can be expressed as:
where is the output elasticity of the variable input and is the expenditure on that input as a share of the firm’s revenue. The output elasticity is estimated from the production function, while the expenditure share can be calculated from firm-level accounting data. Thus, marginal cost does not have to be directly observed.
The intuition is straightforward. Under the assumptions required by the method, a cost-minimizing firm facing a flexible input chooses that input such that its contribution to production is related to its expenditure share. When the firm’s output price exceeds marginal cost, a wedge appears between the estimated output elasticity of the flexible input and its revenue expenditure share. That wedge provides the basis for estimating the markup.
The methodology is particularly useful for the present analysis because digital markets often contain heterogeneous firms operating across multiple products and industries. Production-based markup estimates can provide information about the evolution and distribution of market power even when detailed product-level marginal costs cannot be observed.
Nevertheless, estimated markups must be interpreted cautiously. They depend on assumptions about the production function, the treatment of variable inputs and the relationship between observed accounting expenditures and underlying production quantities. More recent industrial-organization research has also emphasized that estimates of aggregate markup trends can be sensitive to methodological and sample choices. Thus, a markup estimate should be treated as an indicator of market power rather than as direct evidence of anticompetitive conduct.
This distinction is especially important in innovative digital markets. A high markup may arise because a firm has introduced a superior technology, reduced its marginal costs or developed a product for which consumers have a high willingness to pay. Such innovation rents can provide incentives for R&D and investment. High markups alone therefore do not establish that competition has weakened.
Measuring market power separately from anticompetitive conduct
An important distinction is needed between evidence of market power and evidence of anticompetitive behavior. Concentration, markups and profitability may provide evidence about market power. They do not necessarily identify the mechanism through which competition is being harmed.
Certain conduct becomes particularly consequential when undertaken by a firm with substantial market power. Examples include foreclosure, denial of access to strategically important inputs and strategies that raise rivals’ costs. Measuring these concerns requires identifying both the rival and the competitive resource being affected.
This creates a connection between conduct analysis and market definition. To determine whether a firm’s behavior raises rivals’ costs, it is necessary to know which firms constrain it, what customers demand and what alternatives are available. Market definition therefore remains relevant even when conventional market shares are difficult to calculate.
4.2 Developing Measures of Dynamic Competition
The principal measurement problem is to complement static measures with indicators showing whether market leadership can change over time.
One relatively simple measure is market-share persistence:
which measures the probability that a firm among the five largest firms today remains among the five largest after k years. The measure could equally be constructed for other rankings or market-share thresholds.
Persistence should not be interpreted mechanically. High persistence may indicate barriers to entry, but it may also indicate that incumbent firms continue to innovate. It should therefore be compared with innovation, entry and product-quality measures.
The combination of markup and persistence measures is more informative than either measure alone. A high-markup firm that is rapidly challenged by entrants presents a different competitive situation from a high-markup firm whose position remains unchanged for many years while entry and expansion decline. Similarly, persistent leadership may reflect continued superior innovation rather than exclusion and should therefore be considered together with measures of technological improvement.
The empirical framework can consequently distinguish three related but different concepts: concentration measures market structure; markups provide evidence concerning market power; and persistence helps measure the durability of that market power. Evidence of anticompetitive conduct requires an additional step—identifying mechanisms such as foreclosure, raising rivals’ costs or restrictions on access to strategic inputs that can explain why competitive constraints have weakened.
A second important indicator is whether entrants can scale. Counting entry alone may be misleading. A market could generate numerous startups while remaining effectively protected if entrants cannot acquire sufficient customers, data, compute, investment or distribution to challenge incumbents. Competition authorities should therefore observe not merely the number of new models but the rate at which entrants obtain investment, improve performance, attract users and become sustainable competitors.
Third, innovation should be measured directly where possible. R&D expenditure and patents provide useful information but are incomplete measures of digital innovation. For foundation models, changes in model performance, inference costs, speed, reliability and capabilities may provide more direct evidence of technological competition.
Fourth, switching and interoperability should be considered. High switching costs can make market shares more durable. Relevant costs include not only monetary costs but also loss of data, accumulated preferences, compatibility and integration with complementary services. In AI markets, the ability of developers and business users to move between models, use several models, or build applications that are not dependent on a single provider may provide particularly useful evidence about competitive constraints.
The dynamic competition framework proposed here therefore combines market power, entry and scaling, innovation, market mobility, switching and access to strategic inputs. It is deliberately a dashboard rather than a single numerical index. The relative importance of individual indicators will differ across technologies and over time. For instance, in a new and dynamic area such as GenAI where there are many potential players, entry and scaling and access to strategic inputs would be of concern. In a mature industry, such as search, innovation and switching would be of concern.
5. Conclusion
Digital markets require a broader approach to measuring competition. CR4, HHI, market shares, markups and profitability remain useful, but each captures only one dimension of the competitive process.
Traditional measures of market power direct attention toward the distribution and persistence of market power and the increasing economic importance of high-markup firms.11 De Loecker, Jan, Jan Eeckhout, and Gabriel Unger. 2020. Supra note 10. Careful inspection of the GenAI stack, by contrast, provides a particularly useful perspective on competition within an emerging technological market.12 Schrepel, Thibault, and Alex ‘Sandy’ Pentland. 2025. Supra note 3 The analysis emphasizes that foundation-model competition is currently dynamic, heterogeneous and innovation-driven, while also identifying increasing returns and access to important inputs as forces that could alter future market structure.
Taken together, these perspectives illustrate why static measures cannot be abandoned but also cannot be sufficient. High markups and concentration may identify where market power exists. Dynamic measures are needed to determine whether that power is being challenged.
The central competition concern is therefore not whether dominant firms exist, but whether their positions remain contestable.
This requires measuring entry and the ability of entrants to scale; changes in market leadership; innovation and technological displacement; switching and interoperability; and access to strategic inputs.
GenAI reinforces the need for analysis across markets. AI may reduce entry barriers in downstream industries while simultaneously generating increasing returns in foundation models or strengthening bottlenecks in compute, semiconductors and infrastructure. Competition can consequently increase at one layer of the digital economy while becoming weaker at another.
The appropriate objective for competition policy is therefore neither to prevent concentration nor to protect incumbent firms’ innovation rents. It is to preserve the conditions under which innovation can challenge existing market power. Static measures tell us who is large today. Dynamic competition analysis should determine whether an innovative and efficient rival can become large tomorrow.