Discrete choice models in economics are often used to mathematically model the heuristic approaches that people use in decision-making. When faced with a large number of choices, however, people face information costs that lead to choice overload. Within the discrete choice framework, here we formulate a quantization-theoretic approach to optimally cluster choices into categories. This is a non-asymptotic form of rational inattention theory. Drawing on a recent equivalence result between discrete choice models and Bregman divergences, and on properties of Bregman clustering, our main result is that the same clustering algorithm is universally optimal for any additive random utility discrete choice model. Examples are given and hierarchical clustering is also discussed.