We present the design and the implementation of a new expansion type algorithm to solve large-scale bundle adjustment problems. Our approach -- called Power Bundle Adjustment -- is based on the power series expansion of the inverse Schur complement. This initiates a new family of solvers that we call inverse expansion methods. We show with the real-world BAL dataset that the proposed solver challenges the traditional direct and iterative methods. The solution of the normal equation is significantly accelerated, even for reaching a very high accuracy. Last but not least, our solver can also complement a recently presented distributed bundle adjustment framework. We demonstrate that employing the proposed Power Bundle Adjustment as a sub-problem solver greatly improves speed and accuracy of the distributed optimization.