We propose a first attempt at quantifying risk transmission between geographically close firms, as the economic environment is assumed to have an impact on the growth and survival rates of Small and Medium-sized Enterprises (SMEs). We assume that individual Probability of Default is driven by two risk factors: the first is linked to firm fundamentals, the second external and transmitted by neighbor-to-neighbor contagion. We test this hypothesis on the whole French SME population. We first apply modern machine learning techniques to estimate the internal risk of each firm. We then propose to use graphs and Message Passing Neural Networks (MPNN) to simulate risk contagion between neighboring firms. We apply explanatory methods to extract information from the unique structure of MPNN. We reveal statistical evidence for several candidate phenomena discussed in the literature: agglomeration economies, detrimental local competition, impact of multi-company leadership. We confirm that risk transmission is facilitated between companies sharing common characteristics such as size, sector and representatives.