This ignores capital and opportunity cost. Building a GPU data center or chemical plant costs a lot. If you only use it 20% of the time, you're effectively paying 5x more for that capital equipment.

> make synthetic fossil fuels from atmospheric CO2.

that would actually be my preferred solution (if only it was less energy inefficient, sigh).

The problem is the capital cost of any of that type of equipment sitting around idle or under-capacity, ready to go when the electricity price goes down. It's likely more profitable to run them most of the time, even with positive electric rates, and then only stop using them when rates are exceptionally high ("load shedding").

This is why you see most opportunistic electricity consumption systems doing resistive heating - this equipment is inexpensive.