A dynamic gradient architecture generates brain activity states

•Latent fMRI BOLD activity dimensions represent dynamic spatial activity gradients.•The first latent dimension reflects a unipolar sensory-association gradient that 1) explains the most variance in the BOLD signal, 2) has a near-perfect correlation with the global signal, 3) shows the strongest spatial correlation with the principal component of gene expression, and 4) exerts the most influence on the dynamic activity of the other gradients.•Activity gradients appear to be intrinsic based on spatial stability across individuals and task states.•We use a dynamical systems model to show how activity gradients can interact via state-specific coupling parameters to create distinct brain activity and connectivity patterns in different individuals and during different tasks.