Observers proclaim artificial intelligence (AI) as ‘the most important general-purpose technology of our era' and will reshape our economy and society (Brynjolfsson and McAfee, 2017). With robots expected to replace humans in some professions, AI presents a new development prospect, through the provisions of data work. Large Silicon Valley technology firms rely on outsourcing of data work to different parts of the globe via a host of intermediary suppliers and labour platforms often based in the Global South regions (see Anwar and Graham, 2020; Tubaro et al. 2020). In other words, the poor and marginalised in the Global South form the hidden underbelly of the Silicon Valley, training some of their most advanced machines. AI-dependent systems such as ChatGPT and driverless cars are actually trained by human workers, many of whom are sourced from the Global South. With research on data work is beginning to highlight the risks to workers including the precarious employment relations (Muldoon et al., 2023; Anwar and Graham, 2022) and physical and psychological toll (Micelli and Posada, 2022), there is limited knowledge on the outsourced and subcontracting networks and the socio-political processes (e.g. regulation and policies) that enable this work in the Global South regions. Most importantly, data work's potential for wider labour market transformation (including segmentation and occupational mobility of workers) remains less studied. This paper addresses these gaps by drawing upon one of the largest surveys and in-depth interviews of East African data workers in Kenya and Uganda, two of the main hubs of data work for AI in Africa. This collaborative study is done in partnership with two ex-data workers who worked for one the major data enrichment firms in the East Africa. Overall, the study highlights not only the mental and psycho-social harms and risks of human labour behind the making of AI systems but also the ways in which the value chains of AI might be generating new power dependencies in the Global South labour markets and how AI is being resisted by the same people it aims to replace.