There are a collection of private trackers run by motivated volunteers and paid for with meager donations which have collections far superior to the best paid services or archives anywhere or at any time in history.
They do this by erecting and maintaining virtual economies where what you can get is limited by how much you've shared.
You don't really need 100% uptime trackers, DHT lets you do peer to peer file discovery. There are trackerless torrents, the only problem being how you get the torrents in the first place, but all that takes is a magnet URI from the original uploader / creator somewhere.
So, you're suggesting that SoundCloud seeds all the files instead of offering the download option. Perhaps this could work if "play-in-browser" vs "download" ratio is not too big.
I haven't actually kept up to date recently on the project, but the ML is super active. I believe they now have a working localforage backend to store torrents - the implementation already works flawlessly, and there is a hybrid client in the same node package family that can seed the same files to both webtorrent cleints and traditional UDP clients.
But it is extensionless and seamless and just uses webrtc data channels and websockets.
There are implementations which approximate bittorrent written entirely in javascript. It's possible.
However, I really doubt infrastructure costs dominate SoundCloud's money problems. Significant, yes. But a major engineering overhaul developing an entirely different architecture for distribution which might either not work or alienate customers is probably not an appropriate move when in that situation.
Maybe it could be something to do as a clean shut-down to help things live on (an amazing thing when dying companies do things like this to end well) but not ... expected.
Mathematically, as the download count for each file approaches 1.0 (or even lower), a privately-funded torrent seed server starts to look more and more like a traditional file host.
There are a collection of private trackers run by motivated volunteers and paid for with meager donations which have collections far superior to the best paid services or archives anywhere or at any time in history.
They do this by erecting and maintaining virtual economies where what you can get is limited by how much you've shared.