If you have 17TB of volume data you would like to replicate then you need at least this on the downstream array as it needs an initial replica of all the volumes you are replicating. And after that it depends on how many copies (snapshots) of data you would like to keep and how much your data changes.
Nimble is pretty smart and it will only replicate changed blocks and on the downstream array you will have a initial replica plus the changed blocks over time.
Like Rick mentioned, you can tune some settings on replication to control downstream space usage. The important thing to know is that Nimble replicates based on snapshots and snapshots are controlled by volume or volume collection. Thus, you can replicate some volumes but not others (maybe you have some volumes that you could lose and not be too bothered). And like Rick said, you can have a different number of snapshots on the source array than on the target but judging by my array stats, the snapshots are a small part of the overall data. Keeping 50 snaps on the source but only 5 downstream won't reduce space consumption by 90%, but only perhaps 10-20% depending on your data change rate.