well the actual reason is obviously that companies love to lie through their teeth about what they're actually building. i should run the numbers again to make sure my assumption that nvidia's sales numbers are far too small for "AI data centers" to actually be a thing that exists is still correct, haven't checked in a while
@leo i havent done the numbers but even with the same constant revenue let's say to keep it simple 1 billion month 1 is 1 bil month 2 is 1 bil even when that doesnt increase, that still means the amount of products that have existed has doubled (+100%) i can see how companies just buying the same stock of gpu would mean they'd run out of space, cus a single data center can like only fit that many gpus before you have to build a new one
@illyBytes well the other argument is companies started claiming they were spending absurd amounts on constructing "AI data centers" very suddenly these data centers would only actually become operational if there was demand for them demand would presumably have to have risen similarly rapidly for this to be the case building a data center takes several years this implies there would be a period of demand massively outstripping supply. there is no indication i have seen that this actually happened
@leo like someguy in procurement literally added a 0 by accident, that would have happened with enough of these stupid startups spawning out of nowhere like silverfish in an portal room
@illyBytes that would imply theyre just sitting there doing nothing which if true precludes most of the negative impacts being attributed to "AI data centers"
@linear@illyBytes yes but this still implies they aren't actually operational which precludes most of the negative consequences being attributed to them
@leo@linear oh alright, i forgot that im like the only person that gives a fuck about environmental effects of building buildings lol x3 :3 (yeah i genuinely forgot people literally only talk about the harms coming from operations x.x)
@leo@illyBytes ah, like, water and electricity use and such?
because yes, if you do the actual research, the water use is mostly an imaginary issue except where increased power usage drives it in power stations. and power usage is going up in areas that already have datacenters or the few that have had them built (like the region i live in).
there's still a hell of a lot of construction-related impacts going on, though, and construction is one of the larger oft-ignored sources of carbon emissions
@linear@illyBytes i think construction impacts are a plausible concern yes but are datacenter projects actually significant relative to the rest of the construction industry?
@leo@illyBytes quick estimate based on monetary spending figures on construction in the US suggests: doubtful? but not sure
according to census data the US spent upwards of $2,100 billion on construction projects in October of 2025
according to a few sources, the total amount spent in 2024 on "AI datacenter" construction was only $54 billion, globally
a few big companies made bigger commitments than that during 2025 but it's extremely unclear to me how many of those actually happened, which i think is part of the issue more broadly
@leo@illyBytes my concern is that our power grid is already inadequate and strained and i do not see the power delivery infrastructure being built out that is necessary to sustain even a fraction of what they are claiming to be deploying
and that some projects that already have been built solved this by running their own inefficient high-emissions natural gas plants onsite and finding loopholes in regulations meant to make this illegal
i am saying that while they build out these datacenters and wait for their orders of gpus that will never arrive, they are taking what gpus they can, refitting existing datacenters and other abandoned buildings that can be made into temporary datacenters (much like with crypto mining) and they are doing everything they can to find things for them to do to be running hot as much as possible, under VC subsidy, in order to addict people to the LLM and make it think that they can't live without it so that there is demand in the future when they think they will be able to power on their big projects.
which is similar to other tech trends that have happened, yes, but i think still is viewable from the general increasing power use of society, as a significant phenomenon in its own right
@leo@illyBytes from the data i've seen, until about 2023-2024, yes. then there was a significant spike in 2025, and if you look at datacenters in specific regions where these deployments are happening (like "datacenter alley" in Virginia, very near where i currently live), the power usage spike is even higher, upwards of 30% where 18% would be more typical. and anecdotally, the rate of increase is currently growing, not shrinking
@linear@illyBytes i think in my initial posting i phrased my real objection poorly, which is that everything people are attributing to "ai data centers" should be attributed to either the construction sector or tech sector at large, and "ai" is just a scapegoat
@linear@illyBytes i think arguing that in the coming years AI making up a significant portion of what the data center industry will do is inevitable or even plausible is harmful, even if (correctly) concluding that would be a bad thing
@leo@60228.dev
