That's hard to say; the zkEVM landscape is currently moving very fast with already three main contenders in the ring, and performances are being gained there and there.
Maybe in the end ASICs will be required, maybe they will be used as an intermediary solution, but there is no inescapable requirement on the long term for them yet. By this, I mean that the current use of compute power is not seen as a feature, but as a regrettable side-effect of current algorithms, that is being worked on to reduce as much as can be.
Yes, it is an interesting feedback loop and I'll tell you... there are a lot more than 3 main contenders in the ring. Looking over the spreadsheet tracking all of this, it is closer to 20. You probably just haven't heard of them, yet.
Maybe in the end ASICs will be required, maybe they will be used as an intermediary solution, but there is no inescapable requirement on the long term for them yet. By this, I mean that the current use of compute power is not seen as a feature, but as a regrettable side-effect of current algorithms, that is being worked on to reduce as much as can be.