Leading power TMD factorization for physical and lattice observables is a well understood and well developed framework. However, it suffers from having fairly limited kinematical regions of applicability. To improve the situation, one may consider to go beyond the leading power approximation. In this talk, I will present the results for the TMD factorization theorems beyond leading power. I will do so by introducing the concept of physical TMD distributions of next-to-leading power, discussing their peculiar properties and how they emerge in the context of a next-to-leading power factorization theorem for an observable. I will motivate our results in the context of quasi-TMD distributions and the possibility of extracting the Collins-Soper kernel from lattice simulations.