People here might be interested in this blog post, in which we first give a historically-grounded explanation of why GRIB is so poorly suited to the Cloud, and then show how to create “virtual Zarr” stores as a interface to archival GRIB files, all without duplicating the data.