CMIP5 models show a mean dry bias and a wide inter-model spread in their representation of Indian Monsoon precipitation. The origins of the bias and spread have not been well understood. Using moisture and energy budget analysis that exploits the weak temperature gradients in the tropics, I will show that they are linked to the models' treatment of convection over equatorial Indian Ocean. About half of the 20 models analyzed operate at the steep part of the normalized non-linear relationship between precipitable water and precipitation, where small differences in the former would lead to wide spread and preferentially to higher multi-model mean precipitation over equatorial Indian Ocean and consequently to a mean dry bias and a large spread in monsoon precipitation. The models at the flat part of the relationship tend to show less spread and are in a better agreement with observations. Furthermore, the models on the steep part of the relationship project 33% increase in Indian monsoon precipitation by the end of the century under RCP 8.5 forcing scenario. Those on the flat part on the other hand project about 15% increase on average, which suggests the multi-model mean increase of about 20% is likely an overestimate. The processes that control the non-linearity and implications for observing and modeling of tropical convection will be discussed.