Accurately predicting the next shopping basket of a customer is important for retailers, as it offers an opportunity to serve customers with personalized product recommendations or shopping lists. The goal of next-basket prediction is to predict a coherent set of products that the customer will buy next, rather than just a single product. However, if the assortment of the retailer contains thousands of products, the number of possible baskets becomes extremely large and most standard choice models can no longer be applied. Therefore, we propose the use of a gated recurrent unit (GRU) network for next-basket prediction in this study, which is easily scalable to large assortments. Our proposed model is able to capture dynamic customer taste, recurrency in purchase behavior and frequent product co-occurrences in shopping baskets. Moreover, it allows for the inclusion of additional covariates. Using two real-life datasets, we demonstrate that our model is able to outperform both naive benchmarks and a state-of-the-art next-basket prediction model on several performance measures. We also illustrate that the model learns meaningful patterns about the retailer’s assortment structure.