I think the point of the different storage tiers of AWS S3 is to get customers to classify their own data, then AWS can pick the right mix of hardware, softwar

I think the point of the different storage tiers of AWS S3 is to get customers to classify their own data, then AWS can pick the right mix of hardware, software, and compute that satisfies AWS’s requirements for availability and COGS. If the difference between standard S3 and S3 Glacier was just slower disk, then rate limiting the customer would suffice. But if there’s a significant amount of compute thrown at data de-duplication, compression, and indexing, then it starts to clarify why there’s a pricing penalty for using Glacier with the same access patterns as one would use on standard storage.