Recent work suggests that AlphaFold2 (AF2)–a deep learning-based model that accurately infers protein structure from sequence–can also discern important features of folded protein energy landscapes, defined by the diversity and frequency of different conformations in the folded state. Here, we test the limits of its predictive power on fold-switching proteins, which assume two structures with regions of distinct secondary and/or tertiary structure. Using several implementations of AF2, including two enhanced sampling approaches, we generated >280,000 models of 93 fold-switching proteins, the experimentally determined conformations of these were most-likely present in the AF2 training set. Combining all models, AF2 predicted fold switching with a modest success rate <25%, indicating that it does not readily sample both conformations of fold switchers overall. Furthermore, both of AF2’s confidence metrics selected against models consistent with experimentally determined fold-switching conformations in favor of inconsistent models. AF2’s performance on seven fold-switching proteins outside of its training set was then assessed by generating >159,000 structural models with the enhanced sampling method AF-cluster. Among all models of these seven targets, fold switching was accurately predicted in only one. Furthermore, AF2 demonstrated no ability to predict alternative conformations of two newly discovered targets without homologs in the set of 93 fold switchers. These results suggest that AF2 has more to learn about the underlying energetics of protein ensembles and indicate the need for further developments of methods that accurately predict multiple protein conformations.