This paper presents a highly efficient formulation of charge equilibration (Qeq) for fourth-generation machine learning potentials (MLPs), which are used to describe atomic interactions with the accuracy of electronic structure methods at a fraction of the computational cost. The proposed method does not require explicit computation of the Coulomb matrix elements, achieving quasi-linear scaling with respect to the number of atoms. This makes it suitable for large systems where traditional Qeq methods are computationally expensive. The method also allows for the efficient calculation of energy derivatives, which consider the global structure dependence of atomic charges obtained from Qeq. The authors demonstrate the effectiveness of their approach by incorporating it into the RuNNer software, showing significant performance improvements over conventional methods. The method is applicable to various force fields and machine learning potentials, enabling more accurate simulations of complex systems.This paper presents a highly efficient formulation of charge equilibration (Qeq) for fourth-generation machine learning potentials (MLPs), which are used to describe atomic interactions with the accuracy of electronic structure methods at a fraction of the computational cost. The proposed method does not require explicit computation of the Coulomb matrix elements, achieving quasi-linear scaling with respect to the number of atoms. This makes it suitable for large systems where traditional Qeq methods are computationally expensive. The method also allows for the efficient calculation of energy derivatives, which consider the global structure dependence of atomic charges obtained from Qeq. The authors demonstrate the effectiveness of their approach by incorporating it into the RuNNer software, showing significant performance improvements over conventional methods. The method is applicable to various force fields and machine learning potentials, enabling more accurate simulations of complex systems.