By Andrea Marinoni et al.. Coming to you and me via X. And of course, I begin by apologising for the intrusion of climate into this war-focussed blog.The basic idea seems perfectly sensible: from a MODIS dataset, identify datacenter locations and look at the temperature change after they turn on.
A nit: as we see from their figure, the average effect is a temperature jump, which isn't surprising: you don't expect a trend, unless the datacenter is ramping up over time; from the figure, it looks like that isn't usually the case. And yet they say stuff like "the Bajio region a great hub for AI data centres1. Nevertheless, it has been recorded [sic] a serious LST increase trend (in the order of 2°C) in the last twenty years in the region". But it isn't a trend, which they effectively admit by expressing it in degrees, rather than degrees-per-time. But this is just a carelessness of language, not a real flaw.
The flaw is in what they don't say: how does their datacenter heat island effect compare to the usual UHI? A random search for London and Paris says 5-10 oC compared to rural. So my suspicion is that worrying about the DHI is about as useful as worrying about their water usage. For bonus points, their "A way forward" section is comically silly; and their conclusion - " Our study shows a non negligible
and rather remarkable impact of the AI data centres on their local regions" - is simply not true; NN is correct, but RR is pure emotionalism and doesn't belong in a real paper. This is of course but a preprint.
Refs
Notes
1. It seems to me that they are rather casual about "datacenter" vs "AI datacenter". Obvs - no? - neither Bajio nor anywhere else has been an AI datacenter hub for the past 20 years. I doubt that much matters for the analysis.
No comments:
Post a Comment