Cooperative interactions may frequently be reinforced by "partner fidelity feedback," in which high-or low-quality partners drive positive feedbacks with high or low benefits for the host, respectively. Benefits of plant-animal mutualisms for plants have been quantified almost universally in terms of growth or reproduction, but these are only two of many sinks to which a host-plant allocates its resources. By investigating how partners to host-plants impact two fundamental plant resources, carbon and water, we can better characterize plant-partner fidelity and understand how plant-partner mutualisms may be modulated by resource dynamics. In Laikipia, Kenya, four ant species compete for Acacia drepanolobium host-plants. These ants differ in multiple traits, from nectar consumption to host-plant protection. Using a 5-year ant removal experiment, we compared carbon fixation, leaf water status, and stem non-structural carbohydrate concentrations for adult ant-plants with and without ant partners. Removal treatments showed that the ants differentially mediate tree carbon and/or water resources. All three ant species known to be aggressive against herbivores were linked to benefits for host-plant resources, but only the two species that defend but do not prune the host, Crematogaster mimosae and Tetraponera penzigi, increased tree carbon fixation. Of these two species, only the nectivore C. mimosae increased tree simple sugars. Crematogaster nigriceps, which defends the tree but also castrates flowers and prunes meristems, was linked only to lower tree water stress approximated by pre-dawn leaf water potential. In contrast to those defensive ants, Crematogaster sjostedti, a poor defender that displaces other ants, was linked to lower tree carbon fixation. Comparing the effects of the four ant species across control trees suggests that differential ant occupancy drives substantial differences in carbon and water supply among host trees. Our results highlight that ant partners can positively or negatively impact carbon and/or water relations for their host-plant, and we discuss the likelihood that carbon-and water-related partner fidelity feedback loops occur across ant-plant mutualisms.