Many models predict that the diversification and efficiency of financial intermediaries ("banks") increases with their size, so that a relatively unrestricted banking sector will settle into an equilibrium with several large, well-diversified, and competitive banks. However, this prediction is at odds with the actual pattern of unrestricted banking sector evolution in many countries. I develop a model that motivates this actual pattern and examine the model's implications for regulatory policy. I show that an investor's return from a bank depends on the number of investors using that bank; this adoption externality makes investor beliefs about other investors' actions critical for bank competition. In a young banking system with free entry, coordination problems lead to excessive fragmentation, and debt overhang makes it difficult for small banks to capture additional market share. As the system matures, many banks fail, and the survivors become the focus of investor beliefs; these incumbents gain a strong advantage over entrants, facilitating collusion. Entry restrictions reduce fragmentation but aid collusion, while government insurance for investors reduces incumbency advantage and collusion but may cause excessive fragmentation. Thus, regulators may wish to impose temporary entry restrictions, along with partial insurance. These results are consistent with historical evidence from several countries.Journal of Economic LiteratureClassification Numbers: G21, G22, L13.