The anoxic/aerobic granular sludge process is an effective, low-footprint alternative to conventional activated sludge wastewater treatment. The architecture of the microbial granules allows the co-existence of different functional groups, e.g. nitrifying and denitrifying communities, which permits compact reactor design. However, little is known about the factors influencing community assembly in granular sludge, such as the effects of reactor operation strategies and influent wastewater composition. Here we analyze the development of the microbiomes in parallel laboratory-scale granular sludge reactors operated at low (0.9 kg m-3 d-1), moderate (1.9 kg m-3 d-1) and high (3.7 kg m-3 d-1) organic loading rates and the same ammonium loading rate (0.2 kg NH4-N m-3 d-1) for 84 days. Complete removal of organic carbon and ammonium was achieved in all three reactors after start-up, while the nitrogen removal (denitrification) efficiency increased with the organic loading rate: 0% at low, 38% at moderate, and 66% at high loading rate. The bacterial communities at different loading rates diverged rapidly after start-up and showed less than 50% similarity after 6 days, and below 40% similarity after 84 days. The three reactor microbiomes were dominated by different genera (mainly Meganema, Thauera, Paracoccus and Zoogloea), but these genera have similar ecosystem functions of EPS production, denitrification and PHA storage. Many less abundant but persistent taxa were also detected within these functional groups. The bacterial communities were functionally redundant irrespective of the loading rate applied. At steady-state reactor operation, the identity of the core community members was rather stable, but their relative abundances changed considerably over time. Furthermore, nitrifying bacteria were low in relative abundance and diversity in all reactors, despite their large contribution to nitrogen turnover. The results suggest that the organic loading rate has considerable impact on the composition of the granular sludge communities, but also that the granule communities can be dynamic even at steady-state reactor operation due to high functional redundancy of several key guilds. Knowledge about microbial diversity with specific functional guilds under different operating conditions can be important for engineers to predict the stability of reactor functions during the start-up and continued reactor operation.