There may be PUE. However AI adds significant net new energy expenditure. Even if you get higher efficiency per compute unit, you're still burning more coal and gas, or using renewable capacity that could be going towards replacing coal and gas. And then in this particular example 7T is a huge amount of capital that could move the needle on so many difficult climate problems like decarbonizing steel, concrete and ag. I think spending that on AI that will likely only accelerate us towards planetary ecosystem destruction is .. not great to say it politely. 😂 Climate change is an existential problem for our species. The lack of AI advancement, even the lack of AGI is not. If there was any likely solution to climate change that could come out of AI, I might think differently. However we have all the computation methods needed to analyze and solve what we can about climate change. We just need much more resources in doing those things now.

Speaking of compute optimization, someone who's actually built chips just chimed in on the matter.