This paper develops a quantitative model of internal urban structure that incorporates both agglomeration economies and commuting congestion to study the effect of land use regulations on the spatial organization of cities and welfare. Using newly constructed spatially disaggregated data for Shanghai, we estimate and calibrate the model. In particular, we estimate the elasticities of bilateral commuting time with respect to residential and workplace employment densities respectively. We show that the model-predicted equilibrium outcomes can fit the current data better than the predictions from an alternative model without considering the congestion effect. We carry out counterfactual analyses to evaluate various land development strategies for Shanghai in the future. We note three sets of findings. First, without adding any new floor space, re-allocating the current floor space according to the values of local amenities can generate large welfare gains. Second, if we keep the allocation of current floor space unchanged and spread out the new floor space all over the city, the welfare gains are 9.58% relative to the initial level. By contrast, if we focus on developing certain localities of the city and build all the new floor space there, the welfare gains are around 12% or more relative to the initial level, much higher than the spread-out case. Thirdly, decentralized development works better than centralized ones for Shanghai. Among all the 10 km-wide belts along the radius from the city center, if we develop all the new floor space in the one that is nearest to the city fringe (i.e., the 50-60km belt), we can achieve welfare gains of 15.42%, relative to the initial level. Or if we decentralize the new floor space to the sub-centers, we can get a welfare gain of 14.12%, relative to the initial level. By contrast, allocating new floor space to either 0-10km or 10-20km belts works poorly. Alternatively, if we focus on developing the government-specified core urban area (i.e., the 35 streets near the city center), we can get a welfare gain of 12% relative to the initial level, which is still lower than the above more decentralized development pattern. In sum, our findings suggest that a more decentralized yet compact land development according to local amenities seem to be a sensible way to build up Shanghai in the future.