Post
We’re taking
1. the *most extreme purity* ever engineered in a mass production context and…
2. pushing it into a *rapidly degrading environment* and…
3. building out an entire economy on new levels of *waste* (which feeds back on 2)
“The utilization rates of GPUs for AI workloads in a datacenter run by cloud service providers is between 60% and 70%. With such utilization rates, a GPU will typically survive between one and two years, three years at the most”
@urlyman Getting harder and harder to think that AI is *not* a bubble.
Hmm, with that rapid depreciation, I wonder if you need to count it as OpEx instead of CapEx. If something is consumed during the course of a year, it isn’t capital expenditure.
Requiring companies to declare GPUs as OpEx would kill the bubble immediately.
@urlyman … and all of the ’magnificent 7’ devalue the GPU’s they ”bought” from Nvidia at 5-6 years in their financial reports, when they’ll actually burn out in 1-3 years..
It’s not a bubble at all.
…This is just new levels of stupid
@urlyman And I’m pretty sure no one in the design and production loop of these things give any thought to sustainability or material flows.
@mojala there are people really worried about it. e.g. https://semiengineering.com/ai-power-consumption-exploding/
But there is money to be made in the meantime 🤷♂️
A space for Bonfire maintainers and contributors to communicate
