Could whatever actions have to be performed on the data also be performed in AWS?
Also while briefly looking into this I found that AWS has an egress waiver for researchers and educational instiutions: https://aws.amazon.com/blogs/publicsector/data-egress-waiver...
The other is for reproducibility - typically you need to preserve lots of in-between steps for peer review and proving that you aren't making things up. Some intermediary data is wiped out, but usually only if it can be quickly and easily regenerated.
You may use AWS yourself, but your collaborators have made their own choices for their own reasons. The data goes where it's needed, often crossing institutional boundaries.
As for leaving data in AWS, data is often (not always) revisited repeatedly for years after the fact. If new questions are raised about the results it's often much easier to check the output than rerun the analysis. And cloud storage is not cheap. But yes it sometimes makes sense to egress only summary statistics and discard the raw data.