Why economic growth will not go crazy with AI

Essentially saying that Amdahl's Law dominates. Speed up part of a process; its total time taken will be dominated by the parts that can't be sped up. In this case the process is the economy, and time is money.

I look from a macroeconomic point of view. To a first approximation, GDP is household spending (and the reason for the other parts existing is also household spending. No business invests unless it expects to sell more)

Again to a first approximation, household spending is wages.

So saying "GDP will rise 20% per year" is approximately equivalent to saying "total wages will rise 20% per year".

That ... seems unlikely.