For marine biogeochemical models used in simulations of climate change scenarios, the ability to account for adaptability of marine ecosystems to environmental change becomes a concern. The potential for adaptation is expected to be larger for a diverse ecosystem compared to a monoculture of a single type of (model) algae, such as typically included in biogeochemical models. Recent attempts to simulate phytoplankton diversity in global marine ecosystem models display remarkable qualitative agreement with observed patterns of species distributions. However, modeled species diversity tends to be systematically lower than observed and, in many regions, is smaller than the number of potentially limiting nutrients. According to resource competition theory, the maximum number of coexisting species at equilibrium equals the number of limiting resources. By simulating phytoplankton communities in a chemostat model and in a global circulation model, we show here that a systematic underestimate of phytoplankton diversity may result from the standard modeling assumption of identical stoichiometry for the different phytoplankton types. Implementing stoichiometric variation among the different marine algae types in the models allows species to generate different resource supply niches via their own ecological impact. This is shown to increase the level of phytoplankton coexistence both in a chemostat model and in a global self-assembling ecosystem model.