Greetings,
Pandas is great and was a trail blazer but given that Polars is written in Rust and leverages a multi-threaded, lazy-evaluation engine with Apache Arrow memory format, it provides impressive performance and memory efficiency for large datasets compared to the single-threaded, eager execution model of Pandas. Polars was able to learn from and clean up Pandas' syntax by being built from scratch with a consistent, expression-based API, avoiding the legacy issues that led to Pandas' often-tricky, idiosyncratic behaviors.
Additionally, Polars does not appear to be a passing fad, but rather picking up steam.

<Figure as of mid-2025; source: Gemini>
Please consider supporting Polars in ESRI software APIs and shipping Polars with ArcGIS Pro in addition to the currently supported Pandas. Thanks for listening.
Tyler