Standard Disclaimer: I work at AWS. Not only are all opinion my own. I’m so far away removed from the service teams, that I might as well be a just like any other customer - I work in the consulting department.

That being said, looking at things from the outside, I tend to think that objects are not stored in a compressed form. It would make many operations much slower.

1. There are APIs to fetch byte ranges from objects. Would that work without decompressing the entire object?

https://docs.aws.amazon.com/whitepapers/latest/s3-optimizing...

2. You can query over multiple objects using SQL either via Athena or S3 Select.

https://docs.aws.amazon.com/AmazonS3/latest/userguide/s3-gla...

Edit:

Someone has already explained how byte range requests would work.