Computation is a strange bedfellow for renewables.
Computation has 2 costs; the capital cost for the hardware and the energy cost to run the hardware. You pay for energy when you use the hardware. You pay capital costs weather you use the hardware or not.
If the hardware is cheap, then it makes sense to use the hardware intermittently, if you have a job to do 24x7 you simply have 4-6 copies of your hardware spread around the world and use solar power when solar production is locally at peak. The hardware is cheap so you don't care that its spending 80% of the time unused; you are essentially storing solar production in the form of completed calculations.
If the hardware is very expensive, then you have to pay lots of $ simply to have the hardware sitting unused on the shelf. You are desperate to use that expensive hardware 24x7. Your costs are dominated by the hardware costs, and while you care about energy you are focused on making best use of your capital investment.
My vague recall is that bitcoin hardware is getting cheap enough that it is worth following cheap energy, but that AI training hardware is devilishly expensive, the investors are pushing to gain market share and be 'ahead of the competition', and everyone want to run their AI hardware 24x7.
-Jonathan