Estimates of infants' language exposure are necessary for computational studies that attempt to model and learn from infant language experiences. However, there are no well-established input estimates usable for this purpose. This paper explores empirical data on infant language exposure across various cultural settings to derive plausible limits on the speech exposure infants might receive during their first years of life. First, we argue that several assumptions lack unanimous agreement and demonstrate that existing data are problematic in multiple ways. Integrating these uncertainties and published information, we find estimates that range from 1 to 3,300 hours per year. We end by discussing how such a large possible range may impact evaluation of the plausibility and benchmarking of computational models.