The distribution of fire on Earth has been monitored from space for several decades, yet the geography of global fire regimes has proven difficult to reproduce from interactions of climate, vegetation, terrain, land use and other human activities by empirical and process-based fire models. Here, we propose a simple, yet robust, model for the global distribution of fire potential based on fundamental biophysical constraints controlling fire activity in all biomes. In our ‘top-down’ approach we ignored the dynamics of individual fires and focus on capturing hydroclimatic constraints on the production and (seasonal) desiccation of fuels to predict the potential mean annual fractional burned area at 0.25° spatial resolution, here estimated by the 0.99 quantile of the observed mean annual fractional burned area () over the 1995-2016 period of the Global Fire Emissions Database (GFED4). We show that 80% of the global variation in can be explained from a combination of mean annual precipitation and potential evapotranspiration. The proposed hydroclimatic model reproduced observed 0.99 quantile fire activity levels equally well across all biomes and provided the first objective underpinning for the dichotomy of global fire regimes in two domains characterised by either fuel production limitations on fire or fuel dryness limitations on fire. A sharp transition between the two climate-fire domains was found to occur at a mean annual aridity index of 1.9 (1.94 ± 0.02). Our model provides a simple but comprehensive basis for predicting fire potential under current and future climates, as well as an overarching framework for estimating effects of human activity via ignition regimes and manipulation of vegetation.