Observations of nearby galaxies have revealed a strong correlation between molecular gas mass and star formation rate (SFR), but a much weaker correlation between atomic gas and star formation. Indeed, outside of H_2-dominant regions, SFR correlates with stellar, not gas surface density. One possible explanation is that the rate of giant molecular cloud (GMC) formation is strongly influenced by the stellar potential well. Equilibrium models, such as those of Krumholz, McKee, & Tumlinson, have had some success at explaining the observed molecular gas fractions and star formation efficiencies in galaxies. However, they do not track the processes of GMC formation and destruction, so they implicitly assume that the bottleneck for star formation is in GMCs rather than the diffuse ISM. Using HI and CO observations of the Large Magellanic Cloud and other spiral galaxies, I discuss how the conditions for H_2 formation may be more complex than simple shielding models would indicate, requiring both high HI column density and relatively short dynamical timescales. New observations from the MAGMA CO survey of the LMC also call into question the assumption that the mass surface density and virial parameter of GMCs are nearly constant.