In recent years, phenotyping approaches in plant breeding have expanded in both methodology and data collection capacity. One such tool, Near-Infrared Spectroscopy (NIRS) generates a wealth of reflectance values for biological samples. To test the potential of NIRS-based predictions, a hundred grain sorghum hybrids generated from a 10 × 10 factorial mating design were evaluated across eight environments. Hybrids were phenotyped for grain yield, days to anthesis, plant height, kernel hardness index, kernel diameter, and kernel weight. Hybrid grain samples were scanned with NIRS to generate phenomic data while parental lines were genotyped using genotyping by sequencing. Three different predictive models: genomic prediction (GP), phenomic prediction (PP), and GP + PP were fitted. Three different cross-validation schemes of untested hybrids in characterized environments (CV1), tested hybrids in uncharacterized environments (CV2), and untested hybrids in uncharacterized environments (CV3) were completed. GP + PP significantly improved over GP for days to anthesis, kernel hardness index, kernel diameter, and kernel weight for CV1. Prediction accuracy of GP + PP was also significantly improved for the kernel hardness index and kernel weight for CV2 and CV3. Depending on logistics, phenomic prediction has the potential to complement or supplement genomic data for predictive strategies in sorghum.