Except ASS streams really aren't that big and don't have to be stored with each encode. They can just be in a separate file. And this is how cr used to serve them. Before they used hardware drm you could just download all of the separate sub tracks.
You don't need to multiply anything here, except the number of sub streams. One is ass, the other the primitive standards Netflix and other surges use.
Someone else was saying they maintained burn-in subs for devices that didn't handle ASS renderers. Even without accounting for the burn-in versions, using non-standard subs still bumps them off of commodity subtitling services and limits distribution/syndication.
Edit: and to the peer comment regarding S3 vs self host: regardless of 10x cloud cost, it's still 10x volume. Where 1TB local would do, now you need 10TB (10x the cost).
The labor of ass to ttml is there yeah. But the the factors are n_videos * languages * 2 Formats. And considering these are pretty compressible text(34MB->4MB for a completely bonkers sub track that includes animations, animated fonts and otherwise transformed text). I can't imagine that hosting costs cost more than their analysis.