Tech companies will probably win. Copyright law is usually deployed to keep little people in line, and there's no danger of them being able to train SOTA AI models.

We're going to be in an awkward situation where pirating can net you a $150000 per mp3 fine but big companies can train on the said mp3s and produce a nearly identical product for you.

But that's why we have a giant police and propaganda machine.

I tend to think licenses will be required to train models made available to the public. The tech companies have deep pockets and lawyers, but so do the big media companies. And the big media companies have a more effective propaganda arm. My precedent would be Google Books, where a lawsuit similar to the ones we are seeing now shut down Google's plans.

Although it would be hilarious to think what would happen if the news industry went to war for instance, feeding false information to the AI scrapers making them look ridiculous.

I think it also ties into not wanting AI to be trained on AI generated content. The winners will want to license original sources, and licenses work both ways so you can sue if paid for clean data and got dirty.