How to scale recommendations to extremely large scale using Apache Flink. We use matrix factorization to calculate a latent factor model which can be used for collaborative filtering. The implemented alternating least squares algorithm is able to deal with data sizes on the scale of Netflix.