This article is part of the special issue New Measures of Competition in the Digital Economy, prepared in collaboration with the International Center for Law & Economics (ICLE).
1. Introduction
Competition law relies on measurement. A market share above a threshold can make a firm dominant, and a concentration index can decide whether a merger receives a second look. But these measures are often less relevant in digital markets. Marginal costs often sit close to zero and many services are nominally free for users. Products also evolve every few months, so that today’s market share may say little about next year.
The debate on digital markets has often moved to conclusions before the evidence was in. This special issue, prepared with the International Center for Law & Economics (ICLE), starts one step earlier. We asked lawyers and economists what competition authorities and courts should measure in digital markets, and how.
2. Agree on what to measure, then measure
Agreeing on what to measure is where the debate becomes productive. A choice of metric is a choice of theory. Counting tokens rather than revenue, or self-supply rather than sales, already says what one believes competition to be. In our experience, these conversations about variables produce the most interesting exchanges between scholars and enforcers, often more interesting than the debates about outcomes.
Once the variables are settled, the data can be collected. Researchers and enforcers can then test their theories against the evidence. This is the bedrock of the scientific method, and antitrust should be no exception. Aleksandra Boutin, Xavier Boutin and David Teece, whose contribution closes the issue, ask analysts to specify how competition is supposed to work before selecting the metrics. The contributions that follow take that first step.
3. Market shares under strain
Herbert Hovenkamp opens the issue with the measure antitrust uses most. He shows why price-cost margins are hard to apply to digital output, whose marginal cost is often close to zero, and why market shares are less reliable when zero pricing and multi-homing let usage shift within months. The rise of TikTok in social media and of DeepSeek in AI illustrates the point. He proposes that courts demand higher shares before inferring power and look for evidence that those shares are durable.
Stefan Thomas tests four candidate metrics for markets in AI models, namely turnover, token volumes, model downloads and web traffic. Each covers only part of the market, and they cannot be converted into one another. Where shares cannot be quantified, he argues, the usual sequence from market definition to market power collapses into a single analysis of competitive effects. He calls for user surveys to fill the gaps.
Reiko Aoki proposes a dashboard rather than a single index. It combines markups estimated from production data with indicators of entry and scaling, innovation, the persistence of market leadership, switching and access to strategic inputs. The weight of each indicator, she argues, should vary with the technology and its stage of development.
4. Where competition hides
Brian Albrecht argues that AI competes mostly in the markets where it is used rather than those where it is sold. A general-purpose model lets buyers produce in-house what they used to buy from specialists. Drawing on the review of the Getty Images and Shutterstock merger by the UK Competition and Markets Authority, he shows that customer surveys record this self-supply poorly. He proposes to count it separately and to survey past customers as well as current ones.
Georgiana Mirza introduces the innovation access lag, the time rivals wait for a new version or feature once the platform owner or its favored partners already have it. Competition law meets questions of access timing in Android Auto and under Article 6(7) of the Digital Markets Act, yet it has no measure of how long rivals wait. She sets out how the starting point of that measure should follow from the legal question being asked.
Mariateresa Maggiolino discusses the Commission’s April 2026 draft merger guidelines’ focus on innovation capabilities. Using the NBA salary cap as a foil, she concludes that merger control should not seek a more equal distribution of capabilities. Their concentration matters only when it weakens the competitive process, which must be shown transaction by transaction.
Alain Marciano draws on James Buchanan’s analysis of the atrophy of property rights and on Yoram Barzel’s distinction between legal and economic rights. Platform users may hold rights that have become too costly to exercise, and legal incompleteness raises that cost. He proposes to measure enforcement costs, informational asymmetry, asset portability and institutional dependency.
5. Doctrine that needs less measurement
Some contributions ask when the law can do without measuring competition at all.
Daniel Crane argues that an agreement whose sole function is to restrain competition should be illegal per se even when the parties are not horizontal competitors. Firms in digital ecosystems may compete for the value captured at each layer rather than for the same customers. Recognizing that form of competition would spare courts the need to measure effects for such agreements.
Thomas Nachbar moves the question from measuring competition to establishing causation between conduct and its effects. Antitrust lacks a unified theory of harm. In Illumina/GRAIL, the Fifth Circuit deferred to the Federal Trade Commission’s findings on causation, which shows how uncertain causal stories can move authority from courts to the agency. He asks courts to develop doctrine on causation.
Erik Hovenkamp examines defendants who argue that their conduct harmed competition no more than a lawful alternative they would otherwise have adopted. Building on Carl Shapiro’s analysis of the Google Search case, he shows that courts must ask why the alternative would be lawful. Relying on a practice that escapes liability only because legal rules are imperfect would turn a hypothetical judicial error into a real one.
6. Regulation and its evidence
Kevin Laurent asks what designating Amazon Web Services and Microsoft Azure under the Digital Markets Act would add to the Data Act, whose abolition of switching charges takes effect on 12 January 2027. The monitoring planned so far tracks the price of switching, not switching itself. He specifies metrics, weighted by customer spend, that would show whether the remaining frictions are common to all providers or concentrated in a few, and which instrument should then follow.
Alba Ribera Martínez reads the consumer profiling reports that gatekeepers file under Article 15 of the Digital Markets Act. The provision rests on the premise that transparency will reduce deep profiling. The reports point in opposite directions. Microsoft stopped profiling consumers in Windows, while ByteDance expanded its techniques. Without a well-defined objective, she concludes, the provision can neither fail nor succeed.
7. Competition as a process
Aleksandra Boutin, Xavier Boutin and David Teece close the issue. They argue that competition in the digital economy should be assessed as a process over time rather than inferred from a snapshot of market shares. In their view, a firm holds durable market power when it can keep earning high returns after it stops innovating. They ask enforcers to measure the capabilities that incumbents, adjacent firms and potential entrants hold, along with each firm’s position within its ecosystem and its control of critical assets. They then apply this framework to AI markets, where the question is whether rivals are closing performance gaps before they gain share, and to merger review, where the immediate effects of a transaction must be weighed against its longer-term effects on innovation.
8. What comes next
The contributions do not agree on a single metric. Several conclude that market shares carry less weight in digital markets and that other evidence must fill the gap. Others show that some legal questions can be settled without measuring competition at all.
But maybe a single metric was never the goal. Together, these pieces give agencies and courts a list of variables worth collecting. The next step is to collect them, then to meet again and discuss what the data show. The Network Law Review will welcome that work.