The existence of fast moving, cold gas ubiquitously observed in galactic winds is theoretically puzzling, since the destruction time of cold gas is much smaller than its acceleration time. In Gronke & Oh (2018), we showed that cold gas can accelerate to wind speeds and grow in mass if the radiative cooling time of mixed gas is shorter than the cloud destruction time. Here, we study this process in much more detail, and find remarkably robust cloud acceleration and growth in a wide variety of scenarios. Radiative cooling, rather than the Kelvin-Helmholtz instability, enables self-sustaining entrainment of hot gas onto the cloud via cooling-induced pressure gradients. Indeed, growth peaks when the cloud is almost co-moving. The entrainment velocity is of order the cold gas sound speed, and growth is accompanied by cloud pulsations. Growth is also robust to the background wind and initial cloud geometry. In an adiabatic Chevalier-Clegg type wind, for instance, the mass growth rate is constant. Although growth rates are similar with magnetic fields, cloud morphology changes dramatically, with low density, magnetically supported filaments which have a small mass fraction but dominate by volume. This could bias absorption line observations. Cloud growth from entraining and cooling hot gas can potentially account for the cold gas content of the CGM. It can also fuel star formation in the disk as cold gas recycled in a galactic fountain accretes and cools halo gas. We speculate that galaxy-scale simulations should converge in cold gas mass once cloud column densities of N ∼ 10 18 cm −2 are resolved.1 Note that by Eq. (3), clouds which should survive by our criterion are nonetheless disrupted in the simulations of Armillotta et al. (2017). We attribute this to the difference between 2D and 3D simulations: disruption is easier in 2D and continues at cloud sizes which survive in 3D simulations (see § 5.5).