Beth Plale, Inna Kouper, Kurt Seiffert, and Stacy Konkiel have self-archived "Repository of NSF-Funded Publications and Related Datasets: 'Back of Envelope' Cost Estimate for 15 Years" in IUScholarWorks.
Here's an excerpt:
The total projected cost of the data and paper repository is estimated at $167,000,000 over 15 years of operation, curating close to one million of datasets and one million papers. After 15 years and 30 PB of data accumulated and curated, we estimate the cost per gigabyte at $5.56. This $167 million cost is a direct cost in that it does not include federally allowable indirect costs return (ICR). After 15 years, it is reasonable to assume that some datasets will be compressed and rarely accessed. Others may be deemed no longer valuable, e.g., because they are replaced by more accurate results. Therefore, at some point the data growth in the repository will need to be adjusted by use of strategic preservation.