The DRM implementation and algorithm could be "perfect" in a mathematical sense, but as you point out they tend to rely on a PUF in the silicon hardware. Currently very hard to extract but not completely so. However, say a system had a quantum based PUF then it could be unclonable due to QM. Such a system could still be potentially cracked by causing issues in the processor itself like with the ESP32s. Which was my point, since there's a physical system to work in it'd be impossible in a practice to make perfect. Hence it boils down to economics.

It doesn't boil down to economics at all. Even if you push a googolplex dollars into perfecting it. If you wanted you can still relatively easily snoop the electrical signals that control an LCD to reconstruct the video. This is not possible to encrypt and never will be.

It could only be described as "mathematically perfect" in the sense that without the decryption key, the encrypted data is no more useful than random bytes.

But DRM fundamentally needs to have the decryption key available at the end user's device - which at least in my opinion, makes it better described as "provably mathematically impossible".

It can depend on online keys that can be changed by the rightsholder at will.

"The Internet is the ultimate dongle" - John Carmack

You still have to deliver the new keys to the clients. As long as the client can decrypt the file, the client can also decrypt and dump it to storage. And if the top pirates keep their methods secret, you can change the keys all you want and not know how they are extracting them each time.