Physical activity (PA) has been shown to reduce diabetes mortality, but largely based on imprecise self-reported data, which may hinder the development of related recommendations. Here, we perform a prospective cohort study of 19,624 individuals with type 2 diabetes (T2D) from the UK Biobank with a median follow-up of 6.9 years. Duration and intensity of PA are measured by wrist-worn accelerometers over a 7-day period. We observe L-shaped associations of longer duration of PA, regardless of PA intensity, with risks of all-cause and cancer mortality, as well as a negatively linear association with cardiovascular disease mortality. 12.7%, 15.8%, and 22.3% of deaths are attributable to the lowest level of light-intensity, moderate-intensity PA, and vigorous-intensity PA, respectively. Collectively, our findings provide insights for clinical guidelines that should highlight the potential value of adherence to greater intensity and duration of PA for patients with T2D.