The core argument is fair. Data centers are objectively better than scattered on-prem compute, but the author is comparing the wrong baseline. The real question is whether the marginal societal value of another H100 cluster training for the next AI wrapper justifies the resource burn during what's clearly a speculative hype cycle. I don't believe that is clear or even measurable, yet. Gold rush != net good for society.
The core argument is fair. Data centers are objectively better than scattered on-prem compute, but the author is comparing the wrong baseline. The real question is whether the marginal societal value of another H100 cluster training for the next AI wrapper justifies the resource burn during what's clearly a speculative hype cycle. I don't believe that is clear or even measurable, yet. Gold rush != net good for society.