Manufacturing processes such as welding often lead to the formation of a martensitic microstructure in steel structures, along with macroscopic defects that can contribute to fatigue crack nucleation under cyclic loading. To ensure structural safety, it is crucial to predict the crack nucleation life (N_n) and its associated variability for these defects. This study introduces a two-scale approach to evaluate N_n for macroscopic defects in the vicinity of martensitic microstructure. At the macroscale, the structure with a defect is modeled using a continuum elastoplastic finite element model (FEM), while the mesoscale is represented by a crystal plasticity FEM of martensitic microstructure near the defect. The boundary conditions of the mesoscale model are governed by the macroscale simulation. Multiple microstructure instantiations are used in the mesoscale model to address the microstructural variability and a fatigue indicator parameter based on crystallographic slip is used to determine N_n. This approach is applied to evaluate the combined effect of defect shape and microstructural variability on N_n.