In this paper we discuss two changes to the use of data which are driven by the recent rise of synthetic data. Adopting a framework informed by critical political economy and critical data studies, we argue that these technical changes have social and political implications for data-intensive production and the governance thereof. We draw on interviews with synthetic data producers and ethnographic study at a recent developer conference hosted by Nvidia, a leader in the emerging synthetic data sector. First, we discuss a shift from indiscriminate or “frameless” data collection (Andrejevic 2020) to what might be called the precision use of synthetic data to train bespoke models. This has economic implications for the labor processes of data-intensive production, but also political implications insofar as it reconfigures the possibilities of resistance to data-intensive capital, away from opposing surveillance to opposing the mechanisms of data synthesis. Second, we analyze a shift towards increased data portability. With data portability, we refer to how the promises of anonymization made by synthetic data providers suggest a conception of data as fluid and freely transferable, in opposition to emerging legislation such as the EU’s AI Act (Elfering 2019). Drawing on scholars’ work on the political re-use of data (Custers & Uršič 2016; Thylstrup et al. 2022), we aim to further characterize these tensions around data as a controllable resource for machine learning practices that synthetic data presents for industry, private and public sectors (Jacobsen 2023; Offert & Phan 2022; Savage 2023).