Phosphorus (P) is a finite and dwindling resource. Debate focuses on current production and use of phosphate rock rather than on the amounts of P required in the future to feed the world. We applied a two-pool soil P model to reproduce historical continental crop P uptake as a function of P inputs from fertilizer and manure and to estimate P requirements for crop production in 2050. The key feature is the consideration of the role of residual soil P in crop production. Model simulations closely fit historical P uptake for all continents. Cumulative inputs of P fertilizer and manure for the period 1965–2007 in Europe (1,115 kg⋅ha ⁻¹ of cropland) grossly exceeded the cumulative P uptake by crops (360 kg⋅ha ⁻¹ ). Since the 1980s in much of Europe, P application rates have been reduced, and uptake continues to increase due to the supply of plant-available P from residual soil P pool. We estimate that between 2008 and 2050 a global cumulative P application of 700–790 kg⋅ha ⁻¹ of cropland (in total 1,070–1,200 teragrams P) is required to achieve crop production according to the various Millennium Ecosystem Assessment scenarios [Alcamo J, Van Vuuren D, Cramer W (2006) Ecosystems and Human Well-Being: Scenarios, Vol 2, pp 279–354]. We estimate that average global P fertilizer use must change from the current 17.8 to 16.8–20.8 teragrams per year in 2050, which is up to 50% less than other estimates in the literature that ignore the role of residual soil P.