[abought]

Per the budget spreadsheet linked elsewhere, preprints seem to be 20+% of site traffic, but only 0.4-1% of storage costs for the overall site.

This could indicate that many preprints are posted as standalone artifacts (eg not every preprint would include rich datasets alongside the PDF). In effect, people could be skipping the workflow and just using it for the preprint.

For bandwidth usage, they're estimating 20-25% of site traffic. Hosting costs are not a trivial expense, but they're not the main driver of the budget, and the bigger uncertainty seems to be on labor costs.