Additive manufacturing (AM) has recently been widely developed. The highly microstructural-dependent mechanical properties of the AM metallic materials are essential for their application, especially the fatigue performance during cyclic loadings. Traditional finite element method based on an incremental analysis makes it difficult to overcome expensive computations of cyclic responses for a polycrystalline alloy. However, applying direct methods to predict the stable cyclic response and the endurance limit of a polycrystalline alloy shows great computational efficiency. In this work, two direct methods, crystal plasticity enhanced direct cyclic algorithm (DCA) and von Mises yield criterion based classic shakedown theorem (CST) are applied to predict the endurance limit of laser powder bed fusion (PBF-LB/M) 316L stainless steel (SS). Fifteen statistically equivalent representative volume elements (SERVEs) were generated based on the features of the grains. The endurance limits solved by CST are 5.1–34.6% higher than those solved by DCA and the CST results show a greater scatter due to the weak stress localisation at the grain interfaces. Statistically, incorporating defects in SERVEs reduced the shakedown limits. The results indicate that incorporating grain-specific physical properties into fatigue modelling improves the reliability of local mechanical response predictions at the mesoscale.