The data centers at the heart of the AI boom are producing so much heat that they're spiking land temperatures for miles around them by up to 16 degrees Fahrenheit, new research suggests. The effect is so pronounced that the researchers say they're creating entire "heat islands:"
The data centers at the heart of the AI boom are producing so much heat that they're spiking land temperatures for miles around them by up to 16 degrees Fahrenheit, new research suggests. The effect is so pronounced that the researchers say they're creating entire "heat islands."
The findings, detailed in a study that's yet-to-be-peer-reviewed, add to an already grim picture of the environmental impact of these sprawling facilities, the largest of which consume enough energy to power entire cities. Their commensurate greenhouse gas emissions, however, apparently aren't the only way data centers are heating up the world around them.
The researchers focused on roughly 8,400 so-called "hyperscalers," the term used to describe data centers of incredible size that offer cloud computing and AI services. Their construction has surged in the past decade, and the AI boom has pushed their demand and scope to new heights; Meta's new "Hyperion" data center, for example, cost $27 billion to build and has an expected computing capacity of five gigawatts, an appetite that takes ten gas-powered plants to sate.
[...] The effects were local, but far reaching. The researchers found that the temperature increases were felt up to 6.2 miles away — though they dropped off with distance — in all affecting more than 340 million people. CNN's coverage notes that the trend held globally: Mexico's burgeoning data center hub in Bajio saw an uptick of around 3.6 degrees over the past 20 years, as did Aragon, Spain, itself a hot new hub for hyperscalers.
Link to Study: The data heat island effect: quantifying the impact of AI data centers in a warming world
(Score: 4, Informative) by Mojibake Tengu on Sunday April 05, @05:20PM (16 children)
I just asked our new Overlords (whose I, for one, welcome) to compare steelworks and hyperscalers datacenters for energy consumption and it seems typical steelworks plant runs continuously at about 100MW while hyperscalers grow much bigger, up to the 1000MW.
Now the fanatics of the Green Cult have a serious problem.
Rust programming language offends both my Intelligence and my Spirit.
(Score: 1) by khallow on Sunday April 05, @06:13PM (12 children)
Incidentally, I could find only one datacenter [datacentermap.com] in or around Teresina, Brazil (this one [google.com]). Perhaps there are a bunch of hyerpscalers not in that database? Or perhaps it's deforestation in the region around Teresina? Hmmm, looks like this is the same database as what the study used: it claims over 11k datacenters world-wide and 1 datacenter in Teresina which doesn't visually appear very large. *Shrug.*
Ultimately, the paper seems weird because it has no real metric for measuring the extent of data center presence. It's just number of data centers (and maybe even "I hear there's a lot of datacenters here"). I think there would be two obvious metrics to use: land area used and of course, the power consumption of the facility.
I doubt it. Methinks the real problem here is finding new windmills to joust at. Over 11,000 data centers is a great windmill target especially with the shifty nature of some of the participants. Unpopular + money == excellent green cult target.
(Score: 1) by khallow on Sunday April 05, @08:48PM (7 children)
For an example of the targeting, 11 US states [stateline.org] are proposing legislation to slow or temporarily ban data center construction. Similarly, Senator Bernie Sanders has introduced [theguardian.com] a bill to create a moratorium on data center construction. There might be immense cognitive dissonance for a relatively clean industry that consumes electricity and produces heat, but that's never stopped them before.
(Score: 3, Insightful) by Bentonite on Monday April 06, @08:30AM (6 children)
There is nothing clean about datacenter computers and GPUs - producing such computer hardware is extremely polluting;
- Cobalt, tantalum, tungsten, copper and gold and lots of different plastics are used.
- Modern fabs use an incredible amount of electricity - modern EUV lithography equipment is especially bad, considering that 96% of the EUV light is absorbed by the mirrors before it reaches the target and there is a ~0.02% distribution-board-to-plug efficiency for EUV; https://en.wikipedia.org/wiki/EUVL?useskin=monobook#Tool [wikipedia.org]
- Chip production also uses a lot of toxic solvents and etchants.
- There is a lot of drinking water used as ultra-purifying clean water is much cheaper than ultra-purifying ocean water (many production step use ultra-purified water to wash waste away). Previously, polluted water was discarded, although re-processing is now used to reduce costs.
Maybe it won't be so bad if something that lasted was produced and used for many decades, but of course, almost all that is produced on EUV fabs is e-waste out of the factory, with digital handcuffs on the software the hardware runs via a signature check, to ensure the user can't repair the hardware by repairing the software and also can't get freedom by replacing the proprietary software with free software.
Current hardware is ludicrously fast and has ludicrous amounts of memory - while proprietary software keeps getting slower and more bloated on schedule, if the hardware could run fully free software, it could be supported to be used pretty much indefinitely, thus the hardware designers make sure that doesn't happen.
Datacentres typically use hardware for a few years, sell a few of the old computers if convenient and landfill the rest and then get new computers that generally have more processing power for the same amount of power consumption - that is a LOT of e-waste considering many datacenters have 10,000+ 42U racks.
(Score: 1) by khallow on Monday April 06, @02:01PM (5 children)
Those aren't manufactured on site at the datacenter and that lack of cleanness can be controlled at the places of manufacture. My point remains: data centers do extremely well from an environmental point of view for generating economic value.
(Score: 2) by Bentonite on Tuesday April 07, @04:18AM (4 children)
Generating economic value? The whole idea is to crush any value and rather extract the maximum possible profits for the shareholders and send the remainder to the CEO to blow on mansions etc.
It would be a lot environmentally cleaner for SaaSS to not be used and instead for processing to be done properly on the users computer (datacentre + users computers > users computers) .
A steelworks seems much better from an environmental point of view (although there is visible pollution at the mine, transporting the iron ore and also at the steelworks) and the steelworks also actually produces something (steel), which is also feasible to recycle.
(Score: 1) by khallow on Tuesday April 07, @04:39AM (3 children)
How are they "extracting" anything? Don't provide value to someone, don't get profits.
(Score: 2) by Bentonite on Tuesday April 07, @05:05AM (2 children)
Yes, if they don't con enough people, who fall for their scam, they don't get profits.
But scamming people is now regarded as "providing value" (I suspect there would be less harm caused if they instead gained profit via beating people up on the street and stealing their wallets, but I guess that won't be profitable for long).
(Score: 1) by khallow on Tuesday April 07, @01:45PM (1 child)
Aside from the shareholders themselves, who can fall for the scam? People using free search engine AI? Failing companies desperately reaching for anything that will hold their heads above the water? Scammers and spammers who will only use the cheapest AI products they can find? The only serious mark I can think of is the Trump administration. And they can only be scammed for so much.
There's a lack of potential marks here.
(Score: 2) by Bentonite on Wednesday April 08, @12:51AM
Many businesses and individuals have fallen for the scam of datacentres.
Many are paying huge subscriptions for cloudly storage, overpriced server hosting (Azure, AWS etc) and SaaSS etc - when it would be much cheaper and seemingly even more reliable for the businesses and individuals to do their computation on their own computer(s).
As for datacentre's dedicated to LLM's, considering that almost everyone who will fall for the "AI" scam, has fallen for it already and even then, at least 90% wouldn't be dumb enough to pay the actual cost of ~$10-1000/prompt, it's quite clear that such datacentres will not turn a profit.
It seems that the gamble is that with; "just one more datacentre of GPU's", the LLM will magically turn into "AGI" and "if" that fails, then comes the government bailout.
(Score: 4, Interesting) by JoeMerchant on Sunday April 05, @10:10PM (3 children)
Urban heat islands are a well documented phenomenon with 5-10 F temperature increase due to concrete and other human modifications of the landscape collecting and holding heat more than forest or even farm land use.
A drive in a convertible around sunset through an urban / agricultural boundary can demonstrate this viscerally, you feel the air temperature drop dramatically at the boundary.
In our own yard we have a roughly 15' x 30' x 20' high bamboo patch which consistently runs 2-3 F cooler than any other part of the yard, while even shaded concrete slabs will hold heat through the night being around 5F warmer on a cold morning as compared with vegetation covered areas of the yard. Meanwhile, a temperature sensor located within 8' of our HVAC condenser heat exchanger in the shade of the north side wall might run 1.5F hotter while the condenser is putting 36000BTU into the nearby air, yet a similar sensor high on a south facing wall will run 10-15 F hotter when direct sun is hitting the wall under it.
So, all in all, these "study" numbers from TFA aren't surprising at all, as compared with any human construction, data center or warehouse or shopping center with parking lot.
🌻🌻🌻🌻 [google.com]
(Score: 1) by khallow on Sunday April 05, @11:35PM (2 children)
Note that 5-10 F is 2.8-5.6 C. So the documented cases of alleged data heat island effect from data centers are at or below the base of the urban heat island range you mention above. Glancing around, it appears that power density of a serious data center is around 1000-2000 W per square meter - similar to sunlight which is ~1000 W per cross-sectional square meter (strictly for sense of scale). Then bump it up some more for power consumed to cool that much activity. Heat generation is easily an order of magnitude greater than an black asphalt parking lot at high noon in terms of heating. So sure it makes sense that a lot of data center is going to generate a substantial heat island effect.
Glancing at figure 3 on page 6 appears to be where the 9 C number comes from. You have to be right on top of the data center to get that level of heating. And how big is this data center that is generating this curve? I can't find an answer to that.
(Score: 2) by JoeMerchant on Monday April 06, @12:01AM (1 child)
Kind of like acid rain and nuclear fallout, your impact from the heat source depends on distance and wind direction.
Hard to say if data centers with mini nuke power stations would be an asset to a neighborhood or not.
🌻🌻🌻🌻 [google.com]
(Score: 0) by Anonymous Coward on Monday April 06, @01:47AM
Probably not. All that machinery makes a lot of noise.
(Score: 4, Interesting) by ikanreed on Sunday April 05, @06:30PM (2 children)
I don't have a problem, and I like to think about as extremist as it comes in my environmentalist views*. China is pioneering electrified steel plants, and aim to have most of their metalurgy powered by renewables within 10 years, so they can have the steel plant cake and eat it too.
Compared to horrendous ad copy gibberish generators, I know what I'd prefer.
*Noting of course that there are people whose beliefs incorporate things that are utterly untrue like "organic farming is better for the environment" or "nuclear pollutes the air"
(Score: 2) by JoeMerchant on Sunday April 05, @10:18PM (1 child)
What is different about the Chinese electric steel plants vs the ones used for the previous 100 years?
Widespread construction and industrial adoption of induction-heated steel plants began gathering momentum during World War II (roughly 1939–1945) due to the need for rapid production of parts, though early development started in the 1900s-1920s. Initial practical applications emerged in the 1920s, with high-production induction hardening and melting gaining traction in the 1930s before expanding significantly post-war.
🌻🌻🌻🌻 [google.com]
(Score: 2) by ikanreed on Sunday April 05, @11:15PM
Not really a damn thing other than existing and being economically viable due to energy context
(Score: 2) by Captival on Sunday April 05, @09:06PM (6 children)
The study has been debunked long before you posted the article about it.
https://blog.andymasley.com/p/data-centers-heat-exhaust-is-not/comments [andymasley.com]
(Score: 5, Insightful) by janrinok on Sunday April 05, @09:54PM (5 children)
Your counter to the story if from a personal blog. You do not offer any statements explaining why you think the blog has 'debunked' this story. We have enough to do just checking each of the links that are provided in a submission. We have no time to go searching the web for someone I have never heard of to see if he has written anything on the topic. I am not going to read a full paper on a topic before releasing a story. I doubt any other editor will do so either. Your comment could have summarised the salient points and thus contributed to the discussion, and saved each other reader from having to spend a great deal of time doing the same thing.
Who is Andy Masley? What are his credentials? He admits in his info page that "I’m a full-time writer supported by a grant from Coefficient Giving (formerly Open Philanthropy) to do deep dives on AI, policy, and other topics." He states that he is not obliged to write supporting their views, but that doesn't mean that he does not support them. Being a 'full-time writer' as a single statement does not convince me that any thing he has written is worth reading. It might be - but having never heard of him or of any other of his work I could just as easily ask my local butcher, or the woman who drives the school bus, for their opinion.
I'm sorry that you feel that way. We depend on submissions from the community. If you make a submission I'm sure that it will bring the site back to life again - after all Easter Sunday is to celebrate such miraculous things.
[nostyle RIP 06 May 2025]
(Score: 2) by VLM on Sunday April 05, @10:27PM (1 child)
Why even have a discussion site, or at least participate on one, if the measure of quality of an argument is if boring traditional sophistry techniques and job title authoritarianism automatically beat common sense and basic high school physics?
I knew the original article was propaganda the instant I saw "up to 16 degrees Fahrenheit" because 0.000001 F is indeed true under "up to" criteria.
Building a massive bubble of buildings that accomplish little other than environmental damage IS bad, but I think the well is being poisoned by crazy ranty fake stuff being portrayed as the sole argument against such behavior. "Why only the crazies oppose it for nonsense reasons so we should build more" There are better arguments against feeding a bubble than perpetual motion machines and star trek physics and authoritarianism.
The original article is, unfortunately, misinformation and propaganda, and I think for nefarious purposes.
There's no magic in physics, specifically in thermodynamics. A couple megawatts here and there cannot heat things up 16 degrees kilometers away even using somewhat optimistic laser weapons. Unless its the bogey man and then everything bad that witchcraft can be imagined to accomplish must be true. Naah.
I suspect what went on here is the conclusion was determined first, then data was gathered to fit the predetermined conclusion. That technique appears a lot in "policy" research like diet or medical research, climate studies, etc.
Funding sources are missing from the original article. How predictable.
(Score: 2) by JoeMerchant on Sunday April 05, @10:46PM
I agree, TFA is slanted propaganda, however:
>up to 16 degrees Fahrenheit
In common parlance means: a rise of 16F was measured somewhere in the study. Yes, you can go overboard on semantics and ask "why not say 'up to 1000F'?" but that's not a meaningful criticism.
All I in all, I would rather have neither, but if I had to choose, a data center would be a lot less impactful to the neighborhood than a Walmart distribution center. This one https://maps.app.goo.gl/JS1qLdiiekMThT4K7?g_st=ac [app.goo.gl] is, I believe, 10 acres under roof and conveniently located just over the county line so the county it pays property taxes to is not responsible for the road maintenance for those semi trucks' access to the highway.
🌻🌻🌻🌻 [google.com]
(Score: 1) by khallow on Monday April 06, @12:02AM (2 children)
It's a personal blog versus an unpublished paper. I'd say the two sides are in balance. And well, the personal blog seems better written, knowledgeable, and unbiased. I'll note that the author used Claude to come up with a series of serious criticisms of the paper. So even AI can find the holes in this thing.
As to the "debunk" thing? I debunked that the world was round this morning. I looked out the window. It was flat. Debunked! Even the worst thought and writing can debunk. Whether the debunking sticks depends way more on the argument than on the credentials.
(Score: 4, Insightful) by janrinok on Monday April 06, @05:47AM (1 child)
I agree - but you can only state that when you have bothered to read the paper which took you a specific amount of time. I also spent a similar amount of time, as did the original commentator, who could just have easily summarised the paper's contents and told us that it was a well written and convincing piece. However, rather than do that he just left a link. That is not a meaningful discussion. He thinks that the site is dead - leaving miscellaneous links to papers, musings, Youtube videos or whatever does not contribute to the content of this site. It will surely die if community members think that links suffice for intelligent discussion. Links should, IMHO, be used to support a specific point of view - not be the argument itself.
[nostyle RIP 06 May 2025]
(Score: 1) by khallow on Monday April 06, @05:51AM
Good point.