This project is an ethnographic study of the oft-overlooked, labour-intensive work of producing datasets for AI training, fine-tuning, and evaluations. Based on fieldwork conducted in India, following data work at outsourcing centres and a civic tech organisation, this project examines how data work is located within the globalised production of AI. It highlights how prevailing modes of dataset production, as well as attempts to reorient it towards more responsible practices, are caught within and shaped by the logics of supply chain capitalism.