Why do some firms have persistently much higher profits relative to their assets than others? Do they have more productive technologies (i.e., which allows them to produce more for a given amount of capital) or are their technologies more scalable (i.e., marginal return to capital diminishes more slowly for them), or both? Are wealthy households more likely to invest in more productive or more scalable technologies? We use Canadian administrative data on firms and their owners to investigate these questions. In particular, we employ the non-parametric production function estimation method developed by Gandhi et al. (2020) to obtain the joint distribution of productivity and returns to scale. By linking firms to their owners, we then investigate whether wealthier households own firms with more productive or more scalable technologies. Finally, we develop and estimate a quantitative structural model of entrepreneurs adapting heterogeneous production functions, which we use to quantify the importance of returns to scale heterogeneity in wealth inequality and to study the optimal taxation of wealth and capital income.