Cattle ranching and linked tropical deforestation remain intractable global sustainability challenges with large implications for climate security and rural wellbeing. Despite the large area devoted to cattle in the tropics, cattle numbers have hitherto been estimated by downscaling from course data or limited to sparse case studies. In this study, we estimate cattle numbers using deep learning and 30 cm resolution satellite imagery, focusing on the case of the Brazilian Amazon. Our cattle dataset covering a total of 12’100 km2 is integrated with land cover maps and property-level data to assess cattle density. The analysis reveals low cattle intensity of 0.5-0.9 animals per hectare across Amazonian states, with lower intensity on properties with higher recent deforestation. Promisingly, we find evidence that combined policies for conservation (zero-deforestation commitments) and intensification (low carbon agriculture incentives) are linked to cattle intensification. Our method opens new avenues for monitoring and researching cattle practices.