We investigate the impact of within-country spatial concentration of economic activity on country-level growth, using cross-section OLS and dynamic panel GMM estimation. Agglomeration is measured alternatively through urbanization shares and through indices of spatial concentration based on data for sub-national regions. Across estimation techniques, data sets and variable definitions, we find evidence that supports the "Williamson hypothesis": agglomeration boosts GDP growth only up to a certain level of economic development. The critical level is estimated at some USD 10,000, corresponding roughly to the current per-capita income level of Brazil or Bulgaria. Hence, the tradeoff between national growth and inter-regional equality may gradually lose its relevance. Our results also imply that, in terms of foregone growth, the cost of policies that inhibit economic agglomeration is highest in the poorest countries.