The marine nitrogen (N) cycle is essential for regulating ocean productivity, yet over the course of the Anthropocene, it has been substantially perturbed by both anthropogenic and natural drivers. In order to understand how the marine nitrogen cycle has responded to these perturbations, we used the ocean component of the Community Earth System Model (CESM) coupled with the Biogeochemical Elemental Cycling (BEC) module to simulate the temporal progression of the marine N cycle from 1960 through 2010 under observed climatic and biogeochemical forcings. Here, we focus on the seasonal to interannual dynamics, a scale that has remained underexplored. The model simulated long-term mean state compares well with observational constraints, with nitrogen fixation adding 135 Tg N yr^-1 and denitrification in the water column and in the sediments together removing ~200 Tg N yr^-1 (each with ~100 Tg N yr^-1). N fixation and sediment denitrification dominate seasonal variability, changing local rates by up to 50%, while water column denitrification varies little seasonally. The seasonal cycle of N-fixation is driven primarily by temperature and nutrient availability, while the seasonality of sediment denitrification is a result of the seasonal variations of organic matter export along the continental margins. Year-to-year variability of the marine nitrogen cycle is dominated by water column denitrification (WCD), which responds very sensitively to the changes in the oxycline depth in the eastern tropical Pacific caused by El Niño/La Niña. Even though N fixation varies substantially on interannual timescales as well, these variations fail to compensate the WCD variations. As a result, the model simulations suggest that the global marine fixed nitrogen budget is characterized by temporarily existing global imbalances of up to 50 Tg N yr^-1 that persist over several years.  Overall, this reveals a rather dynamic marine nitrogen cycle, where the different key processes respond in isolation to the external perturbations, showing on these timescales no evidence of the negative feedbacks required to maintain homeostasis.