Non-native plants are invading terrestrial ecosystems across the globe, yet little is known about how invasions impact carbon (C) cycling or how these impacts will be influenced by climate change. We quantified the effect of a non-native C4 grass invasion on soil C pools and fluxes in a Hawaiian tropical dry forest over 2 years in which annual precipitation was average (Year 1) and ∼60% higher than average (Year 2). Work was conducted in a series of forested plots where the grass understory was completely removed (removal plots) or left intact (grass plots) for 3 years before experiment initiation. We hypothesized that grass invasion would: (i) not change total soil C pools, (ii) increase the flux of C into and out of soils, and (iii) increase the sensitivity of soil C flux to variability in precipitation. In grass plots, grasses accounted for 25–34% of litter layer C and ∼70% of fine root C. However, no differences were observed between treatments in the size of any soil C pools. Moreover, grass-derived C constituted a negligible fraction of the large mineral soil C pool (< 3%) despite being present in the system for ≥50 years. Tree litterfall was ∼45% lower in grass plots, but grass-derived litterfall more than compensated for this reduction in both years. Annual cumulative soil-surface CO2 efflux (Rsoil) was ∼40% higher in grass plots in both years, and increased in both treatments by ∼36% in the wetter Year 2. Despite minimal grass-derived mineral soil C, > 75% of Rsoil in grass plots was of C4 (i.e. grass) origin. These results demonstrate that grass invasion in forest ecosystems can increase the flux of C into and out of soils without changing total C pools, at least over the short term and as long as the native tree canopy remains intact, and that invasion-mediated changes in belowground C cycling are sensitive to precipitation.