The impact of climate change on the marine food web is highly uncertain. Nonetheless, there is growing consensus that global marine primary production will decline in response to future climate change, largely due to increased stratification reducing the supply of nutrients to the upper ocean. Evidence to date suggests a potential amplification of this response throughout the trophic food web, with more dramatic responses at higher trophic levels. Here we show that trophic amplification of marine biomass declines is a consistent feature of the Coupled Model Intercomparison Project Phase 5 (CMIP5) Earth System Models, across different scenarios of future climate change. Under the business-as-usual Representative Concentration Pathway 8.5 (RCP8.5) global mean phytoplankton biomass is projected to decline by 6.1% +/- 2.5% over the twenty-first century, while zooplankton biomass declines by 13.6% +/- 3.0%. All models project greater relative declines in zooplankton than phytoplankton, with annual zooplankton biomass anomalies 2.24 +/- 1.03 times those of phytoplankton. The low latitude oceans drive the projected trophic amplification of biomass declines, with models exhibiting variable trophic interactions in the mid-to-high latitudes and similar relative changes in phytoplankton and zooplankton biomass. Under the assumption that zooplankton biomass is prey limited, an analytical explanation of the trophic amplification that occurs in the low latitudes can be derived from generic plankton differential equations. Using an ocean biogeochemical model, we show that the inclusion of variable C:N:P phytoplankton stoichiometry can substantially increase the trophic amplification of biomass declines in low latitude regions. This additional trophic amplification is driven by enhanced nutrient limitation decreasing phytoplankton N and P content relative to C, hence reducing zooplankton growth efficiency. Given that most current Earth System Models assume that phytoplankton C:N:P stoichiometry is constant, such models are likely to underestimate the extent of negative trophic amplification under projected climate change.