Pre-release this was a reason [1]. Not sure if its still a compatibility reason:
> There is also an option to let pandas know we want Arrow backed types by default. The option at the time of writing this article is partially implemented and has a confusing API. In particular, it's not yet working when creating data with pandas.Series or pandas.DataFrame. And for loading data from files it will only work when the parameter use_nullable_dtypes is set to True. For example, to load a CSV file with PyArrow directly into PyArrow backed pandas Series, you can use the next code:
> There is also an option to let pandas know we want Arrow backed types by default. The option at the time of writing this article is partially implemented and has a confusing API. In particular, it's not yet working when creating data with pandas.Series or pandas.DataFrame. And for loading data from files it will only work when the parameter use_nullable_dtypes is set to True. For example, to load a CSV file with PyArrow directly into PyArrow backed pandas Series, you can use the next code:
> pandas.options.mode.dtype_backend = 'pyarrow'
[1] https://datapythonista.me/blog/pandas-20-and-the-arrow-revol...