Decided to do some more reading on this topic. TIL:

TCP, the more common protocol; requires at least one side to have a port forwarded through their NAT to the client, so the other side can make a connection to that open port.

uTP on the other hand, can ‘holepunch’ by sending a packet to a known IP, which opens a port through the sending clients NAT, specifically for that IP. That port can then be used to send and receive by either side until it closes due to inactivity.

So, torrent clients can use uTP holepunching to open a port without requiring manual forwarding, then advertise that open port to public trackers. Client ‘A’ will try to connect to an IP+port it got from the tracker and get ignored (because the recipient NAT isn’t expecting data from that IP and drops the packets). Then when client ‘B’ decides to connect to client ‘A’, 'A’s port will now be open and allowing data from 'B’s IP, thus establishing a connection.

This is slower than a direct connection because both clients need to be made aware of each other and decide to attempt to connect at reasonably similar times. It also requires public trackers with peerexchange enabled and the torrents cannot be flagged as private.