Re: Environmental impact of Debian's AI usage

Matthias Urlichs <[email protected]> Wed, 19 Aug 2026 18:58:59 +0200
Newsgroups	gmane.linux.debian.devel.vote
Message-ID	<[email protected]>
On 19.08.26 17:13, Jonas Smedegaard wrote:
> which is about a 4 degree change of a city:

Fahrenheit. Not Celsius. While 2°C is still a lot, I'm not done here.

Also, let's look at the actual paper instead of some (IMHO 
factually-challenged) reporting about it.
https://arxiv.org/abs/2603.20897 for the curious.
TL;DR: Yes, the fact that it's on arxiv instead of someplace 
peer-reviewed is significant:

* they studied 6000 data centers? that's a lot, where did they even get 
their locations and startup times from? none of the cited sources seem 
to tell us. They cite "sources such as" datacentermap.com, but you only 
supplies the startup year from them. Claude or Perplexity can't figure 
it out either.

* they show a singular step increase in ambient temperature at "time of 
start of operations", which they get by averaging the ten months before 
vs. after the data center begins operations (fig.2). Sorry, but that 
doesn't make sense (data center startups are multi-month, if not -year, 
affairs) — unless you intentionally want as dramatic a graph as 
possible. Even then the slow increase after startup is not plausible IMHO.

* they talk a lot about the "hyperscalers" they "considered", but what 
they're actually considering is data centers. These are distinct 
concepts. Given that (a) most data centers are ~ an order of magnitude 
below 10 MW and (b) the paper doesn't even say how many they disregarded 
(and why), or their average load, or in fact *any* information I'd need 
to judge this study's quality. To sum it up, I have no idea exactly 
which real-world objects the authors are talking about.

* Claude:: 2°C is a believable number [context: a rather large data 
center], but the causal story is almost certainly "UHI-like 
land-development effect + localized surface heat rejection," not "the 
giant volume of heated air we calculated is warming the region." (UHI 
= urban heat island).


As to job loss … yes I can operate search engines myself, but the onus 
of supplying sources for assertions made in the course of a discussion 
is on whoever injects them into said discussion, not on everybody else.

But if you insist, let's take e.g. 
https://sqmagazine.co.uk/ai-job-loss-statistics/ :

* *85 million* jobs are estimated to be displaced globally by AI and 
automation by the end of 2026.
[ which IMHO is a gross overestimate, but whatever ]
…

* In contrast, AI is set to create *170 million* new roles globally by 
2030, with *1.3 million* already added recently.

Let's face it, *nobody* has *any* idea what effect AI will *actually* 
have on the job market. We can see how many people get displaced, that's 
rather obvious because [some company replaces a bunch of people with AI] 
reliably generates huge headlines every time it happens, but nobody's 
counting the new jobs.

Kindof reminds me of the billions the German government pumped into 
subsidies for the coal industry (~20k jobs, well documented) a couple 
years ago … while more-or-less-systematically killing off the 
manufacturers of renewables (several 100k jobs, not showing up in any 
statistics).

-- 
-- regards
-- 
-- Matthias Urlichs
OpenPGP_signature.asc (application/pgp-signature, 840 B)