The loss of agricultural biodiversity will compromise societal ability to proof the food system against abiotic and biotic perturbations. The steady decrease in planted area of sorghum [Sorghum bicolor (L.) Moench] in the United States is alarming. Recent studies attributed this decline to a lower rate of genetic gain in sorghum relative to maize due to the lower investment in grain sorghum breeding. While this is a reasonable interpretation, it is also plausible that sorghum breeding has reached a peak in the adaptation landscape for drought within the genetic and physiological boundaries imposed by the germplasm currently used by breeders. To test this hypothesis, we have conducted a breeding gap analysis. CERES-Sorghum was used to run a simulation experiment comprised of ∼1 billion genotype × environment × management combinations for the US sorghum belt. We estimated the 0.99 quantile of the response of yield to evapotranspiration (ET); this boundary defines the biophysical limits to yield based on water availability. We then projected data from multienvironment trials onto this yield-trait space. When trials were conducted in managed stress environments in the absence of water deficit at flowering time, we observed that modern sorghum hybrids reached the biophysical boundary. This result can explain the observed lack of genetic gain, which could be reverted by increasing investments in breeding efforts that harness novel sources of genetic diversity, phenomics, and genome-to-phenome technologies. We hypothesize that there are transfer learning opportunities to inform sorghum breeding strategies that can shift the yield-ET production front from successful crop improvement pathways identified in maize.