The high quality semantic segmentation of kidneys and kidney tumors and cysts from contrast-enhanced CT images hold great potential for enabling the large-scale radiomic analysis of kidney tumor imaging and its relationship to the tumor's molecular characteristics and disease-specific outcomes. Unfortunately, manual semantic segmentation of these structures is prohibitively time-consuming for use in routine clinical care. Highly accurate and generalizable automatic segmentation of these structures remains an unmet need. Herein, we propose the third iteration of the Kidney and Kidney Tumor Segmentation Challenge (KiTS23). The previous iterations of KiTS in 2019 and 2021 introduced and expanded upon the first large-scale publicly-available semantic segmentation benchmark for kidney tumor segmentation, but significant opportunities remain. KiTS23 aims to substantially expand the size of the dataset, and increase its heterogeneity by including cases in the previously excluded portal venous and nephrogenic contrast phases. Our design decisions for KiTS23 are informed by the feedback we've collected from the participants of its predecessors, and by the needs articulated to us by a growing number of clinical partners who have made use of the KiTS dataset, with its wide array of associated clinical data, in their own research.