Count operation slow only on Atlas instance


I have a candidates collection, 7.2GB, 4.5M documents.

                "$gte": ISODate("2023-08-21T12:53:02.038Z")
            "": ObjectId(xxx),
            "conversations.meta.valid": true

This query takes A LONG time (45-180 secs) on a M10 cluster and I’d like to understand why.
The three queried fields are indexed (simple indexes but I also tried a compound index).

If I fully dump the database on my computer and try the same query, it takes 0.2s! Even when I limit my mongo docker instance to 2G of ram like the M10 cluster it’s still as fast.

I’d be ok upgrading the cluster but I’d like to be sure it would really be faster. What other variables could there be? CPU? Disk speed?

Any ideas are welcome, thank you.