This is true specifically for Pandas DataFrames. Numpy arrays are themselves just wrappers around c arrays, which are contiguous chunks of memory. Polars supports operations on files larger than memory.