Discovering chemicals with desired attributes is a long and painstaking process. Curated datasets containing reliable quantum-mechanical properties for millions of molecules are becoming increasingly available. The development of novel machine learning tools to obtain chemical knowledge from these datasets has the potential to revolutionize the process of chemical discovery. Here, I comment on recent breakthroughs in this emerging field and discuss the challenges for the years to come. Toward chemical discovery revolution Computational design and discovery of molecules and materials relies on the exploration of increasingly growing chemical spaces 1,2 (see Fig. 1). The discovery and formulation of new drugs, antivirals, antibiotics, catalysts, battery materials, and in general chemicals with tailored properties, require a shift of paradigm to search in unchartered swaths of the vast chemical space 3. From the fundamental perspective of quantum mechanics (QM), this paradigm shift stems from the fact that molecular properties exhibit complex correlations 3 , which yields whole Pareto fronts of candidate molecules in multiproperty optimization algorithms, enabling "freedom of design". As an example, taking data for more than 100,000 small drug-like molecules, it is found that their molecular electronic (highest occupied molecular orbital-lowest unoccupied molecular orbital) gap is not correlated at all with their polarizability 3 , in contrast to widely quoted chemical rules. This implies that it is possible to design highly conductive and weakly interacting molecules, or molecules that exhibit stability to dielectric breakdown and yet are strongly interacting. Obviously, chemical discovery concerns not only with finding "this special molecule", but also predicting reaction pathways and interactions between molecules, optimizing catalytic conditions, eliminating undesired side effects, among many other important degrees of freedom. Given this vast space of possibilities, a statistical view on chemical design and discovery is mandatory (see Fig. 1). This is the main reason behind the current rise of machine learning (ML) techiques applied to molecular and materials science. The current situation can be compared to the huge advances made by the sustained development of quantum chemistry and solid-state electronic structure codes for modeling molecules and materials during the 1980s and 1990s. The development of steadily more accurate quantum-mechanical approximations and increasingly efficient electronic-structure codes lead to the "chemical modeling revolution". In a similar vein, the development of novel ML methods, combined with first principles of quantum and statistical mechanics, and fed with increasingly available molecular big data, could lead to the "chemical discovery revolution".