Spectral submanifolds (SSMs) have emerged as accurate and predictive model reduction tools for dynamical systems. Recent advances have enabled direct computation of SSMs and their reduced dynamics for high-dimensional finite-element (FE) models by solving the associated invariance equations in physical coordinates using only the eigenvectors associated with the master modal subspace. However, FE models in commercial solvers do not generally provide information on the nonlinearities required for the analytical construction of SSMs. In this talk, we show how to overcome this limitation using a nonintrusive construction of SSM-based ROMs from a small number of unforced FE simulations and use these ROMs to predict the forced response of the FE model without performing any costly forced simulation. Our examples range from simple structures, such as beams and shells, to more complex geometries, such as a micro-resonator model containing more than a million degrees of freedom.