Ocean biogeochemical models are integral components of Earth system models used to project the evolution of the ocean carbon sink, as well as potential changes in the physical and chemical environment of marine ecosystems. In such models the stoichiometry of phytoplankton C:N:P is typically fixed at the Redfield ratio. The observed stoichiometry of phytoplankton, however, has been shown to considerably vary from Redfield values due to plasticity in the expression of phytoplankton cell structures with different elemental compositions. The intrinsic structure of fixed C:N:P models therefore has the potential to bias projections of the marine response to climate change. We assess the importance of variable stoichiometry on 21st century projections of net primary production, food quality, and ocean carbon uptake using the recently developed Pelagic Interactions Scheme for Carbon and Ecosystem Studies Quota (PISCES‐QUOTA) ocean biogeochemistry model. The model simulates variable phytoplankton C:N:P stoichiometry and was run under historical and business‐as‐usual scenario forcing from 1850 to 2100. PISCES‐QUOTA projects similar 21st century global net primary production decline (7.7%) to current generation fixed stoichiometry models. Global phytoplankton N and P content or food quality is projected to decline by 1.2% and 6.4% over the 21st century, respectively. The largest reductions in food quality are in the oligotrophic subtropical gyres and Arctic Ocean where declines by the end of the century can exceed 20%. Using the change in the carbon export efficiency in PISCES‐QUOTA, we estimate that fixed stoichiometry models may be underestimating 21st century cumulative ocean carbon uptake by 0.5–3.5% (2.0–15.1 PgC).