[0] https://github.com/OpenVisualCloud/SVT-AV1

When H.264 was introduced, you needed a state of the art CPU just to play it back... You can imagine how slow encoding it was.

AV1 is designed to give you the ability to throw more CPU cycles at the encode side to achieve higher quality/byte while maintaining reasonable decoding performance. You don't have to use it that way, you can encode faster but give up quality/byte. But without AV1 we would not even have that choice, the previous generations reach diminishing returns at some point.

And hardware decoders are not that great. They use less power and can achieve realtime encoding, but if you want the maximum quality/byte (at the expense of encoding time) then software encoders generally reign supreme due to years of iterative improvements that you don't see in hardware. This is important for streaming services which spend those cycles once and then streams to millions. The asymmetry makes it worth it.