Wide-band-gap (WBG) semiconductors are the materials of choice for future high-power applications as they enable pushing operation limits in terms of switching speeds, operating voltages, currents, or temperatures. An outstanding candidate is silicon carbide (SiC) which has become the WBG semiconductor with the most mature technology. Still, applications of SiC in devices is still largely limited by the challenges associated to the production of high-quality wafers with reduced defect densities arising from micropipes, dislocations, stacking faults or polytypes. To sustain the improvements in efficiency and performance of SiC-based devices, research efforts need to be continued especially on the level of material physics. We present atomistic calculations of silicon carbide with a particular focus on crystallographic defects in different polytypes of SiC including 3C, 6H, 4H and 2H (Figure 1 left). We address point defects with density functional theory (DFT) including, in particular vacancies and dopants and elucidate their impact on polytype stability. Vacancies are key for understanding non-stoichiometry in crystals and their impact on the relative energetic alignment of polytypes in different growth conditions. Furthermore, N and Al are added as single dopants and also as dopant pairs to explore geometric and energetic modifications in the crystal. We show that N doping leads to a clear stabilization of the 3C polytype while Al doping causes only minor changes in the energetic ordering of polytypes (Figure 1, right). This effect is explained based on the differences in band-gap of the polytypes and the dopant-induced defect states in agreement with an explanation suggested previousely. [1] As a consequence, Al in proximity of N can entirely neutralize the donor and its stabilization effect. We demonstrate that charging of the defects can induce exactly the same trends providing a new view on polytype stability, which may lead to new approaches for adapting growth conditions to influence the occurrence or suppression of polytypes. To investigate extended defects such as dislocations and micropipes, straightforward application of DFT is not possible due to the related system sizes larger than 1000 atoms. Therefore, we generate machine-learned interatomic potentials based on the atomic cluster expansion, [3] which are trained on a large number of small DFT calculations where cell dimensions and atomistic arrangements are varied over a wide range to sample the potential energy landscape (Figure 2 left). Our ML potential reproduces polytype energetics from DFT in contrast to available interatomic potentials (Figure 2 right). This allows us to treat large atomistic systems at DFT accuracy. We demonstrate the approach by investigating threading screw dislocations (TSD)s, which play a decisive role for the crystal growth process. We provide a fully atomistic view into the dislocation core with its heavily distorted environment arising from the large burgers vectors typical of TSDs in 4H (Figure 3) or 6H. We investigate at which magnitudes of the burgers vector removal of atoms from the dislocation core becomes energetically favorable to form the open core of micropipes and relate our findings to experimental results from the literature. [3] Finally, we discuss how our results can help establishing guidelines for improving crystal growth quality with the physical vapor transport method. [1] V. Heine, C. Cheng, and R. J. Needs, J. Am. Ceram. Soc. 74, 2630, (1991). [2] R. Drautz, Phys. Rev. B 99 14104, (2019). [3] S. I. Maximenko, P. Pirouz, S. S. Tangali, Mater. Sci. Forum 527, 439, (2006).