Cloud-resolving numerical experiments with prescribed large-scale vertical motion show that the spatial organization of deep convective clouds known as self-aggregation is sensitive to large-scale circulation. In the configuration we use, our model does not simulate self-aggregation in radiative-convective equilibrium: deep convection occurs disorganized cumulonimbus. If subsidence is prescribed, this disorganized stationary state becomes drier and even unstable if the subsidence is intense and the simulation drifts towards a self-aggregated stationary state. This self-aggregated stationary state is actually stable for a large range of subsidence, meaning that there is an interval of subsidence for which we have multiple stationary states. Subsdience favors self-aggregation because in this state the atmosphere is drier than in the disorganized stationary state, and this decreases the export of humidity out of the simulated domain by subsidence, limiting its damping impact on deep convection. We will detail the mechanisms at play to a maintain the two stationary states.