Derivative Data: Rethinking Market Definitions in the Age of Generative AI

Antitrust law fails to keep pace with the data-driven realities of the dig- ital economy, and foundation models further exacerbate the issue. Founda- tion models such as ChatGPT, Claude, and Gemini are trained on broad datasets across different domains. While traditional antitrust frameworks fo- cus on narrow market definitions and readily observable effects, these frameworks fail to capture the anti-competitive potential of derivative data— data that is derived by a business through its operations and exerts cross- market influence—thereby fueling new forms of dominance. Moreover, the dynamics of foundation model training data create a “Tragedy of the Data Commons,” where a few powerful actors benefit at the expense of accessible data resources. Without intervention, the control of foundation models will remain in the hands of tech conglomerates, limiting the open exchange of information and further entrenching their market power. This paper argues that antitrust law must expand the upper bounds of liability to encompass the broader ecosystems in which data operates, rec- ognizing that the “relevant market” for market entities that collect derivative data extends far beyond the immediate product or service the entity provides to consumers. By redefining these bounds, antitrust enforcement can address the evolving challenges of foundation models and ensure that their bottom- less potential does not translate into boundless dominance by a select few.