I think there exists a point, where AI stops making sense in a cost equation. Currently we are well accustomed that machines are cheap and humans are expensive, and anything that can be done by machine instead of by a human means a cost reduction. At some point that probably stops, because the amount of compute you need to run larger and larger models means that humans can become cost competitive. It's a bit like in manufacturing, where cost of robotization is sometimes more than just the cost of employing humans.

And in the end, this long term predictions based on where we are now and how fast we are moving are always a bit iffy. Nobody really know what challenges must current AI overcome to become AGI (if that even means anything).