in reply to Prof. Sam Lawler

A couple of these things I can do nothing about. Some I can.

One I need help with from you: in a few days I'm giving a talk on the environmental consequences of AI data centres (with a focus on orbit, but starting with ground-based).

So, fediverse! Please send me your favourite statistic (with citation please) on exactly how bad AI data centres are for the environment. I'm looking for: comparisons on CO2 emissions per centre, water usage per slop-video made, stuff like that.

in reply to Prof. Sam Lawler

The media in this post is not displayed to visitors. To view it, please go to the original post.

ERCOT is (politely and wonkishly) ringing alarm bells in Texas. ercot.com/news/release/0415202…

This graph is ludicrous. It shows the requests for power to all providers in ERCOT - the growth in the right bars is driven almost entirely by datacenters (particularly in DFW).

Yes, they're requesting Texas *triple* its power generation over the next six years.

in reply to Prof. Sam Lawler

AI data centres create localized heat islands, with temperatures going up by several degrees.
arxiv.org/pdf/2603.20897

News article: uk.news.yahoo.com/ai-data-cent…

in reply to May Likes Toronto

@mayintoronto They also make a constant humming/whirring noise eesi.org/articles/view/communi…
youtube.com/watch?v=JflFFqbZ1X…

And there's a lot more you can find about the hardware side. I started my career by working in the first megadatacenters. I can tell you also that investment in hardware is not one-time. We don't think of it this way, but computers wear down the same way cars or anything else do, and these machines are being pushed to their limits, and the amount of e-waste for churning through parts is huge.

in reply to Christine Lemmer-Webber

@mayintoronto Here are two good articles from Carbon Brief

On the one hand, in 2024 data centers were estimated to only use 1% of global electricity carbonbrief.org/ai-five-charts…

But on the other hand, from the same article, 10% of new growth was expected to be data centers.

And the problem is: much of the new data center power usage coming online is setting up *new* non-renewable power consumption, especially the gas generators built specifically for new data centers

Which leads to:

Analysis: CO2 from UK data centres could be ‘hundreds of times’ higher than thought carbonbrief.org/analysis-co2-f…

in reply to Christine Lemmer-Webber

@cwebber @mayintoronto There's a good video on this from Benn Jordan - youtube.com/watch?v=_bP80DEAbu…. They're functioning not unlike acoustic weapons.
in reply to Bonnie

@BatsInLavender @cwebber @mayintoronto
I'm just throwing this out here, but firsthand it looks well researched (I haven't yet read all of it): blog.andymasley.com/p/contra-b…
The author was mentioned in this video I'd recommend watching: youtu.be/faOD7v0Opq8
And here are some other (maybe useful) posts by him, like "Using ChatGPT Is Not Bad for the Environment":
blog.andymasley.com/p/a-cheat-…
blog.andymasley.com/p/the-ai-w…
in reply to Christine Lemmer-Webber

@cwebber @mayintoronto
We closely manage the lifecycle of all our infrastructure. I'm mostly familiar with our HPC environments where we only have 1 or 2 GPU Compute Nodes. Most components are leased for 3 years. If we purchase them it's 4 years. For large environments we rotate ½ of it every 2 years as long as things are compatible

The leasing company we use in North America is very good at selling the parts that roll out of service

But once things get old enough, ewaste

in reply to Prof. Sam Lawler

Sending hopeful and sympathetic thoughts.

My computer science background stops me at the O(N⁶) training time and O(N²) running time for LLM type AI. Increase their training data by 10, their demand for hardware, energy and time goes up by
about a million. If the resulting model is 10 times bigger, then it will cost about 100 times more hardware, energy and time to run.

These aren't things we can fix by "nerding harder". These factors are baked into the algorithms and hardware of almost all of the AI that is being pushed on us.

Nobody can build O(N⁶) computers, nobody can provide them power and water, and it will never do what the promoters are telling us they can.

(Kevin O'Leary's work on Reader Rabbit does not give him credibility on his data center projects.)

Ed Zitron gets it right (I enjoy his "geek-out" brain dump style, I know it isn't everyone's cup of tea).

wheresyoured.at/how-to-argue-w…

in reply to Prof. Sam Lawler

Just bumped into this: "New Gas-Powered Data Centers Could Emit More Greenhouse Gases Than Entire Nations"

The full article is behind a paywall though:
wired.com/story/new-gas-powere…

in reply to Prof. Sam Lawler

a good article with lots of citations about why narrow stats can be misleading here:

ketanjoshi.co/2025/08/23/big-t…

This pairs well with articles like these:

datacenterdynamics.com/en/news…

capitalbnews.org/musk-xai-memp…

britbrief.co.uk/environment/cl…

Stats are good (and perhaps more what you need for your purposes), but to those who parade misleading stats as a "see, it's not that bad" message, I think it's useful to ask "If it's not that bad, why are these terrible local effects happening at all?"

The short answer is: even though some of the per-query costs don't seem unreasonable, the same companies selling queries have made it their mission to exponentially explode the number of queries used, which is why the net impact is power- and water-starved communities.

in reply to Prof. Sam Lawler

Here in Oregon: "Google built its first data center in The Dalles in 2006. In 2012, the tech giant used 12% of The Dalles water supply. By 2024, a third of The Dalles’ water went to Google’s three local data center sites."

More context and data here: opb.org/article/2026/01/15/as-…

reshared this

in reply to Prof. Sam Lawler

I don't have any citations/stats at the moment, but I do wonder what numbers the businesspeople are looking at that gives them such a [redacted: crass euphemism for overexcited hyperfocal myopia] for such a stupid and short-sighted (in SO MANY ways) plan.

Part of me thinks they are only looking at how much investment they can raise, with not a single glance at actually following through.

But that can't be true for all of them. Can it?

in reply to Francis Bee

@Francis_Bee
Or at least drastically curtail the amount of cloud stuff we do.

I for one have reverted to paper agenda. And I avoid cloud services generally. And try not to use credit card. (When I use it, I have to make a special effort to remember to go pay it off, I go months without using it.)

If we all did this, the oligarchs would take quite a hit, and local economy might do better (with careful local spending).

in reply to Prof. Sam Lawler

arstechnica.com/ai/2026/04/gre… has some statistics, though at a quick glance there are no links to peer-reviewed literature.
And be well—serious illness is no joke (too many people I know are coping with too many serious medical issues), and do what you can to seek relaxation and relief, e.g., goats (or for that matter, actual astronomy).
in reply to Prof. Sam Lawler

How about 55? That is the number of data centers here in Santa Clara, a year ago… localnewsmatters.org/2025/06/0… I ride by many of these several times a week. And more are coming.

Why? Because we have a city owned utility and our rates are about half what PG&E charges.

in reply to Prof. Sam Lawler

Here is something I've posted here earlier.

A rising tide of e-waste, made worse by AI, threatens health, the environment and the economy phys.org/news/2024-11-tide-wor…

Generative AI: Uncovering its environmental and social costs phys.org/news/2025-01-generati…

Advocates raise alarm over Pfas pollution from datacenters amid AI boom theguardian.com/environment/20…

Google to tap into gas plant for AI datacenter in sharp turn from climate goals theguardian.com/technology/202…

AI infrastructure growth threatens water-stressed Thai regions news.mongabay.com/short-articl…

This entry was edited (2 weeks ago)
in reply to Prof. Sam Lawler

Does this help? Power estimates for UK data centres were nearly 100 times too low.

theguardian.com/technology/202…

This entry was edited (2 weeks ago)
in reply to Prof. Sam Lawler

I don't know why the focus is on orbit...

The idea of orbiting datacentres is so transparently a scam, that there's no reason to focus on it, it distracts from the actual problems we face on earth.

I'm legitimately disappointed that members of the press, tech bros, and even scientific community seem to be taking it seriously, but anyone remotely involved with space and computing can see with perfect clarity that this is not a real thing.

in reply to Prof. Sam Lawler

Starlink is different, it isn't a scam, it's a real service. Problematic, maybe, but not a scam.

The point of it being a scam is that it isn't going to happen at any meaningful scale. Maybe one or two test satellites get sent up to dupe some investors, but nothing remotely at scale will ever be deployed. There is no risk of harm from that.

In fact, one of the reason for the space datacenter scam is to distract people from the problems DCs are facing on earth. By focusing on it, we give the tech executives exactly what they want.

This entry was edited (2 weeks ago)
in reply to Prof. Sam Lawler

UK government underestimate AI carbon emissions by factor 100:
hachyderm.io/@inthehands/11647…


And this seems…uh, notable.

“officials raised their estimate of carbon emissions from AI by a factor of more than 100”

oops

theguardian.com/technology/202…


in reply to Prof. Sam Lawler

sad and complaining

Hey for the data center stuff, work with your water district to ensure your (and everyone's farms) are not getting impinged upon. Make sure they know how much static water storage the data center requires (usually a lot) and talk to the neighbor who is down stream, so to speak, of the data center in case their storage fails and they get flooded.

Make sure the people within the power district understand how much of a drain this thing will be and how that will drive rates. Make sure the power entity understands how much of a load this will be and that they will have to supply it.

I am sorry you have to fight more billionaires and their bullshit profiteering, it isn't cool at all. But I am here to tell you it is possible to fight back and I know you have that fight in you!

Also, if they can't get fiber connectivity to the data center in the first place, no point in having it.. That happens through easements and right of way (in the USA anyway) and maybe that is a point you can break their plans too..

Good luck! I hope some of that was informative or helpful! Sometimes just delaying civil or construction start is enough to kill big things like this. It sure won't bring money into your area, just extract natural resources..