users generate more data. This feedback loop makes it harder for new entrants to catch up once a leader has emerged. Moreover, users may find it costly to switch between providers. It is difficult for a business that has built workflows around one AI provider to move to another or for individuals to move their data and history between platforms. Other markets, such as those for operating systems and search engines, are characterized by similar tendencies for data advantages that compound concentration. Data concentration is harder to address through standard competition tools than concentration in other inputs because of data externalities; that is, a single user’s data are nearly redundant for predicting that user’s own behavior, but useful for predicting the behavior of similar users.42 This pattern has significant policy implications. Giving individuals property rights over their own data does little to prevent the concentration of AI’s predictive capability. Even if many users withheld their data, firms holding the largest existing data sets would retain their advantage. Data externalities of this kind require collective governance arrangements rather than specific provisions of individual contracts between AI providers and users alone. Beyond data concentration, individuals in many developing countries face significant challenges in understanding what happens to their data because minimal safeguards are in place. As of 2026, only 73 percent of developing countries have enacted data protection legislation, compared with 98 ­percent of advanced economies.43 Even where such laws exist, they are rarely enforced. This concern has become increasingly important because data collected for one purpose are often repurposed to train models for entirely different applications, making it difficult to trace where the training data come from. Furthermore, documentation regarding the origin of training data, whether user consent was obtained, and the authenticity of the data remains insufficient.44 212 Several policy approaches are emerging to address these market failures. India’s digital public infrastructure, built before the current AI wave, shows how public alternatives can prevent the concentration that AI is likely to intensify. The Unified Payments Interface is a government-provided payment system accounting for about 70 percent of India’s digital payment volume as of fiscal year 2023–24. It processed more than 12 billion transactions in the single month of December 2023.45 The interface is open and works with many systems, allowing any private company to build on it. This feature prevents the platform lock-in that often happens with payment systems in other countries. Meanwhile, India’s Data Empowerment and Protection Architecture expands this model by giving individuals control over their data through consent managers and by enabling data p ­ ortability.46 While implementation challenges remain, the model has attracted interest from several countries. Data cooperatives offer an alternative approach in which individuals pool their data to gain collective bargaining power that they lack individually, aggregating decisions that individual users cannot make effectively on their own and thus addressing the externality problem. Small-scale pilots in India and Kenya are testing this approach for gig workers and others in the informal sector, though evidence on their effectiveness at scale is limited. These asymmetries are not new to AI; they characterize the platform economy more broadly, in which users generate the data that firms convert into commercial advantage. What AI changes is their intensity. When user interactions train the models that are a firm’s main asset, data become more valuable to capture, one user’s data reveal more about others, and switching providers gets harder. Individuals are left with limited means to bargain over how their data are used or to share in the returns they generate. Whether digital public infrastructure, open-weight models, data cooperatives, and competition policy enforcement can achieve sufficient coordination to shift World Development Report 2026