Machine learning interatomic potentials (MLIPs) trained on quantum chemistry data provide access to systems and time scales significantly larger than first principle molecular dynamics simulations. The deep potential (DP) model, an end-to-end deep neural network representation of a potential energy surface, allows reliable investigations of dynamical properties of deep eutectic solvents based on choline chloride and urea. Equivariant atomic representations reduce the computational cost since less expensive quantum chemistry data is needed to train the model. However, large system sizes are needed to obtain activity coefficients from molecular dynamics (MD) simulations.