Aim Biological invasions facilitate ecosystem transformation by altering the structure and function, diversity, dominance and disturbance regimes. A classic case is the grass–fire cycle in which grass invasion increases the frequency, scale and/or intensity of wildfires and promotes the continued invasion of invasive grasses. Despite wide acceptance of the grass–fire cycle, questions linger about the relative roles that interspecific plant competition and fire play in ecosystem transformations. Location Sonoran Desert Arizona Upland of the Santa Catalina Mountains, Arizona, USA. Methods We measured species cover, density and saguaro (Carnegiea gigantea) size structure along gradients of Pennisetum ciliare invasion at 10 unburned/ungrazed P. ciliare patches. Regression models quantified differences in diversity, cover and density with respect to P. ciliare cover, and residence time and a Fisher's exact test detected demographic changes in saguaro populations. Because P. ciliare may have initially invaded locations that were both more invasible and less diverse, we ran analyses with and without the plots in which initial infestations were located. Results Richness and diversity decreased with P. ciliare cover as did cover and density of most dominant species. Richness and diversity declined with increasing time since invasion, suggesting an ongoing transformation. The proportion of old-to-young Carnegiea gigantea was significantly lower in plots with dominant P. ciliare cover. Main conclusions Rich desert scrub (15–25 species per plot) was transformed into depauperate grassland (2–5 species per plot) within 20 years following P. ciliare invasion without changes to the fire regime. While the onset of a grass–fire cycle may drive ecosystem change in the later stages and larger scales of grass invasions of arid lands, competition by P. ciliare can drive small-scale transformations earlier in the invasion. Linking competition-induced transformation rates with spatially explicit models of spread may be necessary for predicting landscape-level impacts on ecosystem processes in advance of a grass–fire cycle.