Arrow is an in-memory data format designed to allow zero-copy interop between languages, including Rust, C++, and Python. It's a bit more sophisticated than just some arrays, but ultimately everything is just arrays.

Polars is implemented in Rust and uses Arrow to represent some data in memory.

Pandas is often used in a way that results in a lot of processing happening in Python. Data processing in Python is comically inefficient and single-threaded. Beating that sort of pipeline by 10-100x is not that difficult if you have optimization experience and are able to work in a more suitable language.