Ideally you’d used Parquet or ORC if querying from Athena/Trino/Presto. Since they are columnar formats you will enjoy considerably faster queues and lower query costs for most query patterns other than “select *…” since the query engine can just retrieve the columns needed for your query instead of the entire row.
100 percent correct.
Ideally you will also partition the data, so your queries can use partitions. Athena charges $5 per terabyte of data scanned, so it's important to get this right.