This study aimed to analyse and predict the state of poverty in South Africa using remotely sensed data. In the developing world, trustworthy knowledge is still hard to come by for economic survival. One of the 17 sustainable development goals listed by the United Nations is to eradicate poverty. Ground‐level survey data, such as household consumption and wealth, were historically the primary sources for evaluating poverty. As technology has advanced, one method that is now used by both wealthy and developing nations is to estimate the poverty index of a region using remote sensing data, such as satellite images, and a machine learning algorithm. This method uses a highly structured, reasonably priced, and easily accessible data source. This machine learning method calculates the poverty index for a given year and determines how the index relates to other years. Inception NetV3, ResNet50 and convolutional neural network are trained beforehand to estimate the intensity of the night‐time illumination matching to the input daytime satellite images. The suggested model also forecasted the cluster wealth score and created a link between wealth scores derived from satellite images and data from the Demographic and Health Survey (DHS). The study investigates a possibly more affordable alternative technique for estimating poverty. The findings imply that satellite data and machine learning can be useful techniques for estimating poverty. By analysing the resilience of the generated estimates to user‐specified algorithmic parameters and model specifications, the work adds to the body of the current literature.