You've heard the term "carbon footprint", no? We should consider the entire energy consumption of a human worker. If you want to compare lifecycle costs, then you have to account for the energy & resource consumption including all of the nonproductive years of a human.
Taking all of this into account, wet brains look remarkably inefficient!
A 2000 calories diet equals to 2.2w/hr per day, or 803w/hr per year, or 64kw/hr in a 80 years life time. This is the energy cost (25% variance) of keeping a human alive and thinking.
Our industrial economy exponentially scales this number, to an extent, out of need, as we need to artificially grow food through industrial agriculture, and keep people warm or refreshed in overexpanded habitats, and the costs of healthcare, and the educational and transport costs to keep all the previous going.
However, from the 5000-10000kw/hr per person to keep a modern lifestyle going to the for example 78000kw/hr somebody in the United States uses, which includes now making AI handle more things that we were already handling before, there is a huge difference.
Technology needs to improve further before it is prudent to use AI in tasks that humans were already doing, and most efforts should be focused in things our non-binary brains cannot handle well, instead. The only reason it makes sense to use AI as people replacement is due to our very broken economic system.
And in a similar way, is due the same that high-level and api redundancy is taking over optimization, too.