The giant impact hypothesis suggests that the Moon formed out of a partially vaporized disk created by a collision between an impactor and the proto-Earth. Three major models exist for this hypothesis: (a) standard model: a Mars-sized impactor hit Earth, (b) fast-spinning Earth model: a small impactor hit the rapidly-rotating Earth, (c) sub-Earths model: two half Earth-sized objects collided. Some of these models can explain several observed features, including the nearly identical isotopic ratios between Earth and Moon. However, it has not been clear if these models can explain other geochemical constraints, such as (1) the giant impact did not mix the Earth’s mantle, (2) Moon did not lose significant amount of water (hydrogen) during its formation. In this talk, I show results from giant impact simulations and investigate whether the suggested models are consistent with these geochemical constraints. I show that the standard model is more consistent with the survival of the mantle heterogeneity than the other models. I also find that water loss from the Moon-forming disk is minor in all models. Therefore, the giant impact hypothesis is consistent with the measured lunar water abundance. Finally, I will discuss implications of our model for the formation of exomoons. Our numerical simulations indicate that whether a planet can form an impact-induced moon depends on the planetary mass and composition.