A tweet comes with a lot of metadata embedded (e.g. profile information containing follower numbers, etc.) so it‘s more like 1K per tweet. When I was working with the subsampled feed that you could get publicly, it was like 5GB of raw data per day.