The “out-of-memory” non-negative matrix factorization algorithm has shown impressive scaling results on data sets too big to fit in computer memory on devices ranging from laptops to supercomputers. The algorithm is used in data compression, data analytics and machine learning.