Per what?
This looks to be the total number of each category. If so, it’s a useless graph. How many L/day are used per single unit. Maybe per farm/datacenter.

Edit.
I did the math
3,562 Cattle farms
21,617,069 L/day/ranch

5,400 data centers
416,667 L/day/data center

45 AI data centers
600,000 L/day/AI data center

This does not take into consideration the heat produced nor fossil fuels consumed by each.

If water is your only metric, cattle ranches are worse. But this isn’t a good metric for many reasons, since livestock, being food, is a commodity, while data centers are not.

Let’s normalize to technology, ans ignore beef for a moment.
AI data centers consume about 33% more water than non-AI centers.
There is not a situation where this is fine.

Water is a distraction. The heat and fossil fueled power consumption is the real problem. They will burn us all alive and we have to sit back and let it happen because the fascists are in power, I guess.

27 million litres that did not have to be used to generate CSAM and shitty essays.

Your argument is basically “AI is good because the beef industry is worse.” Fucking moronic. Get the fuck out of here bootlicker.

Sadly you will die with the rest of us when the cascading environmental collapse gets even worse because of AI being powered by obscene amounts of fossil fuels. Fuck you, bootlicker.

The largest mass of datacenters are still in construction. Using old data to pretend it’s not a problem is just gross.

Oh boy, here we go again. We had this discussion two weeks ago over on !conservation@slrpnk.net (with a cross-post to a vegan community, but please don’t go there and start harassing vegans. It was enough of a shit-show already.), where I wrote way too long of a comment, and the post got deleted afterwards. What a waste. But since it is brought up here again, I’m gonna recycle some of that and make you relive it with me. [Turns out this comment is even longer than my original one. Sorry for that.]

The best source I could find to get a rough idea of what is using how much water (in the USA) was the USGS Estimated Use of Water in the United States in 2015. (yes, there were no AI data centres in 2015, but bear with me, this is gonna be a long ride.)

One of the main problems in discussing this is that there are different kinds of water use, which luckily the USGS publication differentiates - namely “consumptive use” (water that is not available after the process) and, well, “non-consumptive use” (they report it as “reclaimed waste-water”, i.e. water that, possibly after treatment, is immediately returned to rivers or similar for further use).

As the report compares total water use, “thermoelectric” (not sure why they use that term, it’s probably supposed to be “thermal”) power generation is the biggest reported category with 41%, followed by irrigation with 37%. When looking at consumptive use, the share is vastly different - only a fraction of the water used by thermal power generation is evaporated (see page 42), it consumes “only” 4.3 BGal/day, compared to 73 BGal/day for irrigation (page 30). That’s not even in the same ballpark!

Turns out water use != water use. However I’d argue, that people would rather agree, that the kind of water “consumed” by growing plants is more permanent than water being “consumed” by cooling power plants, where most of the water is put back into the water cycle immediately.

So now we have a rough picture of how much water is used for agriculture, and how much for (thermal) power generation.

My point would be that it is very unlikely, that AI data centres use more water than the power plants that generate their electricity (on average; not all electricity is generated in thermal power plants), especially as thermal power plants generate at least as much heat as electricity, so the power plants already need as much cooling as you’d need if you just turned their electricity into heat (which data centres basically do). But even if the data centres used as much water as all the thermal power plants in the USA, it would still only be a fraction of the water used for irrigation.

Now considering that feeding animals is less efficient in converting farmed area and water to calories for human consumption instead of directly feeding people, and it should be quite obvious, that animal products are responsible for a big portion of easily avoidable water usage (For some fun comparisons: calorie supply by food group - In the USA 72% of calories are provided by plants, the remaining 28% are some form of animal product vs global land use for food production - 80% of agricultural land is estimated to be for livestock, vs 16% for food crops (granted, livestock land includes land for grazing, but even without that, it’s 6 million km^2 for animal feed crops vs 8 million km^2 for crops for human food, so disproportionate to the amount of calories they provide))

And on top of all of that, if it wasn’t already enough - the water usage of animal agriculture hasn’t just recently started to become a problem, it has been a problem for quite some time. In the USA most aquifers are drying up including some very important ones for agriculture - and that was a problem before AI data centres sprung up everywhere

Water consumption of AI is not the main environmental problem, its energy use and especially the source of that energy is way more concerning.

The thing is that yes, we should not use AI for all the reasons that have been mentioned numerous times, but if you focus on environmental problems, then the best thing you can do is stop eating animal products. Heck, it doesn’t even cost anything. On the contrary, it would be cheaper. It doesn’t even require new technologies! Going vegan is degrowth in action.

That being said, I’ll still condemn you if you use AI, even if you are vegan.

Yes.

AI data centers are disruptive from the context. They’ve been built haphazardly and quickly. Evaporative coolers, gas turbines, big lots in urban areas; all this is done because they literally cannot wait 6 more months to construct things properly.

It’s billionaire FOMO.

So they end up spewing gas into neighborhoods or running water dry not because of sheer volume, but because they just plopped the data centers down with zero mindfulness, nor any time for infrastructure to adapt.

…But.

On the whole, their water use is minuscule.

So is their energy use compared to, say, metal foundry industries.

Or emissions compared to concrete or ranching industries. Or shipping.

Not that it makes those industries unproblematic.

And I’m counting the hardware construction here, too. Fabs are intense, the datacenter supply chain is long, but the scale really isn’t that big. Just as an example, literally every Nvidia and AMD AI GPU comes from just one complex in Taiwan. The memory comes from three. And the other components on the PCB are pretty modest too.

How do you think that compares to, say, every steel foundry on the planet? And all the coal they ingest?

And the notion that “AI” will somehow require a fraction of all human economic output is propaganda from Sam Altman to trick people into giving him money.

It’s a lie.

It’s a lie for many nested reasons. We cannot possibly use that much “AI,” and even if we did, we shouldn’t, and even if we somehow do anyway, it’s getting exponentially less intensive to run. Not more expensive. It’s all just nonsense out of his mouth, and for some reason he doesn’t get called out for it.

I’m not trying to excuse the buildout, to be clear.

But I think the resources need to be put into perspective.

That doesn’t seem likely. And the pollution done to said water will be much different.

Without an as-of date, there’s no way I’d trust it. AI data centers are pretty new so it wouldn’t need to go very far back for it to be negligible.

Even if the graph is true, it’s misleading . Whatever water beef is using hasn’t really changed, but AI data Centers are a recent significant change much more likely to affect me. Depending on what you’re trying to show that is probably the wrong type of graph

The numbers for data centers are likely a lot higher than this. It should include consumption for creating the building and hardware but I have a hard time believing that this is included in that number. Let alone the amount used for electricity generation.

Yes. I’m vegan and anti-AI.

I hate this kind of reasoning.

“Let’s ignore this big problem because there are other bigger problems”

The same people say shit like: “Why are you sad? There are so many people who have it even worse than you!”

All this makes us do nothing until the “smaller” problems become so big that the damage is done and then those people will say “Well, we are in too deep. It it what it is. Can’t change it now 🤡”

OP is an AI booster, check their profile.

This is an attempt to disregard the damage that’s done by slop generators, comparing it to most inefficient food source.

Hope you have a terrible day, op.

If no source: not true

yes, beef is comically inefiecient (most meat production is too, but beef takes the crown for ineffiency), and data centers have a big incetive to have closed loops water cooling systems, although specially wih loops this big there will always be some significant ammount of lost water

yes agriculture and especially beef is a big problem. however, this does not mean we should ignore the newly emerging problems - just because they are not as big as the biggest problem (yet). i also doubt, that any agricultural business uses as much energy as datacenters.

additionally beef is edible. the only edible thing about datacenters is the rich owner.