The phytoplankton–light feedback (PLF) describes the interaction between phytoplankton biomass and the downwelling shortwave radiation entering the ocean. The PLF allows the simulation of differential heating across the ocean water column as a function of phytoplankton concentration. Only one third of the Earth system models contributing to the 6th phase of the Coupled Model Intercomparison Project (CMIP6) include a complete representation of the PLF. In other models, the PLF is either approximated by a prescribed climatology of chlorophyll or not represented at all. Consequences of an incomplete representation of the PLF on the modelled biogeochemical state have not yet been fully assessed and remain a source of multi-model uncertainty in future projection. Here, we evaluate within a coherent modelling framework how representations of the PLF of varying complexity impact ocean physics and ultimately marine production of nitrous oxide (N2O), a major greenhouse gas. We exploit global sensitivity simulations at 1∘ horizontal resolution over the last 2 decades (1999–2018), coupling ocean, sea ice and marine biogeochemistry. The representation of the PLF impacts ocean heat uptake and temperature of the first 300 m of the tropical ocean. Temperature anomalies due to an incomplete PLF representation drive perturbations of ocean stratification, dynamics and oxygen concentration. These perturbations translate into different projection pathways for N2O production depending on the choice of the PLF representation. The oxygen concentration in the North Pacific oxygen-minimum zone is overestimated in model runs with an incomplete representation of the PLF, which results in an underestimation of local N2O production. This leads to important regional differences of sea-to-air N2O fluxes: fluxes are enhanced by up to 24 % in the South Pacific and South Atlantic subtropical gyres but reduced by up to 12 % in oxygen-minimum zones of the Northern Hemisphere. Our results, based on a global ocean–biogeochemical model at CMIP6 state-of-the-art level, shed light on current uncertainties in modelled marine nitrous oxide budgets in climate models.