Plasmids play an important role in shaping bacterial evolution and adaptation to heterogeneous environments. As modular genetic elements that are often conjugative, the selective pressures that act on plasmid-borne genes are distinct from those that act on the chromosome. Many bacteria are coinfected by multiple plasmids that impart niche-specific phenotypes. Thus, in addition to host-plasmid dynamics, interactions between co-infecting plasmids are likely to be important drivers of plasmid population dynamics, evolution and ecology. Agrobacterium tumefaciens is a facultative plant pathogen that commonly harbours two distinct megaplasmids. Virulence depends on the presence of the tumour-inducing (Ti) plasmid, with benefits that are primarily restricted to the disease environment. Here, we demonstrate that a second megaplasmid, the At plasmid, confers a competitive advantage in the rhizosphere. To assess the individual and interactive costs of these plasmids, we generated four isogenic derivatives: plasmidless, pAt only, pTi only and pAtpTi, and performed pairwise competitions under carbonlimiting conditions. These studies reveal a low cost to the virulence plasmid when outside of the disease environment, and a strikingly high cost to the At plasmid. In addition, the costs of pAt and pTi in the same host were significantly lower than predicted based on single plasmid costs, signifying the first demonstration of non-additivity between naturally occurring co-resident plasmids. Based on these empirically demonstrated costs and benefits, we developed a resource-consumer model to generate predictions about the frequencies of these genotypes in relevant environments, showing that non-additivity between co-residing plasmids allows for their stable coexistence across environments.