Milky Way (MW) satellites reside within dark matter (DM) subhalos with a broad distribution of circular velocity profiles. This diversity is enhanced with the inclusion of ultra-faint satellites, which seemingly have very high DM densities, albeit with large systematic uncertainties. We argue that if confirmed, this large diversity in the MW satellite population poses a serious test for the structure formation theory with possible implications for the DM nature. For the Cold Dark Matter model, the diversity might be a signature of the combined effects of subhalo tidal disruption by the MW disk and strong supernova feedback. For models with a dwarf-scale cutoff in the power spectrum, the diversity is a consequence of the lower abundance of dwarf-scale halos. This diversity is most challenging for Self-Interacting Dark Matter (SIDM) models with cross sections $sigma/m_chigtrsim1~$cm$^2$g$^{-1}$ where subhalos have too low densities to explain the ultra-faint galaxies. We propose a novel solution to explain the diversity of MW satellites based on the gravothermal collapse of SIDM haloes. This solution requires a velocity-dependent cross section that predicts a bimodal distribution of cuspy dense (collapsed) subhaloes consistent with the ultra-faint satellites, and cored lower density subhaloes consistent with the brighter satellites.