Thank you for spending the time to write this when most sensible people would just go "It's obviously stupid" and ignore it. If it weren't for people like you, it would go unchallenged.
When I was originally reading the paper, I was sketched out that they didn't include any specific example satellite photo example comparisons or anything.
For the pixel areas, it would be better to write 0.5km × 0.5km or 250,000m² etc. instead of .5km² and 500m². (You can get away with 1km²!) This doesn't affect your conclusions.
A fine example of which would be Google Mayes County, Oklahoma. It is pretty big but built inside an old Gatorade bottling plant, which was also big, and in the middle of nowhere with bare ground on all sides for miles.
Tried to analyze the heating effects of the data center there by downloading similar NASA data but it seems mostly random without any visible change in temp at all
I wasn’t around for the initial water panic arguments, but I hope to use the knowledge I gained from this to challenge my anti ai friends. Hopefully this disinfo doesn’t spread as easily, thanks for writing this.
Hah, I saw the headline this morning, read that the study hadn't been peer reviewed yet, and filed it under "check back in when it's been peer reviewed". Thanks for your service.
That's not so, but I can see how that may have been misleading. I thought about fact checking the article but didn't bother until it had at least been peer reviewed (which usually means reviewed prior to appearing in a journal, just to define terminology - and some journals have higher standards than others) - Andy fact checked it instead and satisfied my curiosity.
> the red line is the average, and the error bars represent the 95th percentile bounds, meaning the vast majority of measurements fell within those much smaller bounds
the paper itself seems a bit unclear about what the range represents, but if those are error bars for the mean, then all datapoints could in theory be outside of that range.
From the paper: "The shaded areas show the interval between the maximum and minimum value of LST increase that has been recorded across the considered AI hyperscalers. Finally, the bar across the average line identifies the limit of the 95th percentile of the distribution we compute."
I am with you on "this is wrong". I am with you on "buildings are hotter than grass". But I think there is still something missing here.
Look at that goofy graph again. The timeline on the x axis shows that ~all of the effect happens in ONE MONTH. The normalized temperature increase goes from 0 to 0.2 C from -10 months to t=0, and then jumps up to 2.2 C from month zero to month one and then after that it stays pretty flat, only increasing by half a degree over the remaining 9 months.
That's not what I would expect from an increase in development. That looks like some kind of coding error. With no publicly available code and data right now (MODIS LST is public but it wouldn't be trivial to recreate the specific areas they looked at), I can't see what it is but I think there is some other problem here than what you've identified so far.
I suspect this might actually be because the main thing the satellite is picking up, the roof, is one of the last things to go up, and they're setting the baseline a few months before the DC went up when it's still being constructed, so the baseline is already where there's building present, but the roof causes the jump?
The majority of changes in the physical landscape are not due to the buildings themselves, but the large amounts of roads and buildings around it. Do you really think adding a roof on at the end will make enough of a material difference to produce the massive jump in temperature shown on the graph?
The physical construction time for a data-centre is something like 1-2 years: if the physical concrete of the buildings was what caused the effect you would expect to see temperatures steadily rising toward the maximum value many months before the data-centre actually started operations: not a miniscule rise followed by a massive jump.
If this graph were true it very much would support the papers conclusions that AI datacentre operations were to blame, rather than the buildings themselves. But the finding seems pretty odd given what we know about urban heat island effects and the energy emitted by datacentres: skepticism is warranted.
The vast majority of the change I'm seeing in these images is grass -> dirt, which I wouldn't imagine would have much of an effect? It seems like they expanded the roads a bit but I'm not seeing the roads take up anywhere near as much surface area as the DCs
I think grass to dirt is actually the biggest part of the effect! Plants have a major cooling effect, because they use some of the sunlight to do the work of photosynthesizing, and more importantly use heat to evaporate water from their leaves, causing capillary action to suck water up from their ground, which is their equivalent of pumping blood to get nutrients from one part of their body to another. Grass doesn’t use as much water this way as trees, so it doesn’t convert as much heat to water evaporation, but it still does a good amount, and dirt doesn’t do any.
A quick search found one article trying to compare the albedo effect to the transpiration effect, and finds that in particularly dry climates, the albedo effect might be larger. But in the kinds of places where the ground is normally covered with vegetation, transpiration is going to be larger.
Even if you exclude the dirt, about half the area is coming from roads and parking lots. And even without the roof, the data centre is still there, being a giant chunk of concrete: it seems implausible to me that a roof alone would make that much of a difference.
Plus, do you have any evidence that data centre operation starts within a month of the roof going on the top? Like, it seems like it would take longer than that to install all of the internal equipment and data-racks and so on.
It does not seem likely to me that your theory alone explains the graph.
Your assumption is wrong. The bulk of building material that would be picked up from a thermal/satellites perspective is BY FAR the concrete (and it isn't even close). That is all on site/in place during the first half of a project. If anything the roof would insulate and "hide" the thermal change from the satellites but at that point in the project the roof going up is basically irrelevant.
I'm an architect with an understanding of materials and building science. Data centers are like the perfect building science study because they are simple structures with extremes for the inputs so things are much more binary than a building with a complicated footprint, an inconsistent envelope or mechanical loads. You are correct in thinking that the passive heat loads on these sites is massive, but the roof going on at the end of a project isn't making a difference. Roofs specifically are made from high albedo material (if its white this is the case), and if anything would reduce the heat being measured from a satellite.
As with most things, the truth is likely somewhere in between.
Hard to not look at the waste heat from the cooling systems. All of that would come online at the completion of the building-the only time the building starts turning megawatts into waste energy. Nothing else late in the building process (or at any time during construction to be honest) is really going to change anything that much. If this were all just essentially heat island effect, the graph would look like a staircase, not a cliff.
I gave both the paper and your post a very quick read, I'll try to be more thorough later.
i dont think i understand how the building construction itself could result in such a sudden (<1 month) change in temperature? a large building takes more than a month to construct, right?
I suspect this might actually be because the main thing the satellite is picking up, the roof, is one of the last things to go up, and they're setting the baseline a few months before the DC went up when it's still being constructed, so the baseline is already where there's building present, but the roof causes the jump?
We're already discussing this in the other thread - but I want to be clear, this is completely wrong from a construction timing and building science standpoint. The baseline already exists before the roof goes on. If anything, the roof will reduce the heat signature when it goes on.
You and Claude mention "dark-roofed" buildings, but *are* they dark? When I see large commercial buildings, the roofs are generally white.
Given how good the LLMs can be at arguing a case (any case), it might be useful to prompt it the other way, or have it critique its critique. Only prompting it one way (to find flaws in the paper) may leave you with some confirmation bias.
It would be interesting to try to reproduce but substitute large warehouses or something for data centers and see how large the effect is. If you get 98% of the effect (or whatever it is) from a warehouse, that's a simple story to tell to substantiate the "it's just construction" explanation.
This whole thread is a very interesting read as someone who is interested in AI and tech but has a professional background in buildings and building science.
I think the bigger effect is removal of vegetation - plants move nutrients by forcing water evaporation (which is why they need so much water) and this process converts a lot of heat into vapor.
As always, thank you for taking the time and effort to pick this apart.
Activists really seem to be losing their minds about data centers and AI in general, either to the point that they're just not thinking clearly (the "system 2 motivated reasoning" phenomenon) or worse, that ends justify means. It's an article of faith that "all this will pass", and the longer it doesn't, the more the panic rises.
Among the many depressing harms that this type of poor research is doing, it's already destroyed many people's heuristic - including mine - that science with a pro-environmental slant can be taken to be largely trustworthy, honest mistakes aside.
Thank you for spending the time to write this when most sensible people would just go "It's obviously stupid" and ignore it. If it weren't for people like you, it would go unchallenged.
The before/after pics of the Bajio industrial park from Google Earth are crazy and like 2% of it is data centers and the other 98% of it is car parts.
When I was originally reading the paper, I was sketched out that they didn't include any specific example satellite photo example comparisons or anything.
For the pixel areas, it would be better to write 0.5km × 0.5km or 250,000m² etc. instead of .5km² and 500m². (You can get away with 1km²!) This doesn't affect your conclusions.
Was debating this, thanks!
Yeah, as written it is confusing
This is confusing in the paper itself. It says
"We used a reconstructed MODIS LST dataset (produced
by NASA) acquired worldwide from 2004 to 2024 over an enhanced 500m resolution grid"
The lowest resolution public MODIS LST is 1 km. (Table 1 here lpdaac.usgs.gov/documents/715/MOD11_User_Guide_V61.pdf)
They link to this paper (https://www.researchgate.net/publication/262974152_Surface_Temperatures_at_the_Continental_Scale_Tracking_Changes_with_Remote_Sensing_at_Unprecedented_Detail) which talks about a spatial resolution of 250m. I'm guessing they are using a statistical modeling technique that was created for that paper, but they wanted smaller error bars so they went bigger?
it's like they have never heard of the urban heat island!
An interesting control would have been to find a brownfield data center that has been built on land that wasn't farmland/woodland before hand.
A fine example of which would be Google Mayes County, Oklahoma. It is pretty big but built inside an old Gatorade bottling plant, which was also big, and in the middle of nowhere with bare ground on all sides for miles.
Tried to analyze the heating effects of the data center there by downloading similar NASA data but it seems mostly random without any visible change in temp at all
Which seems to confirm your initial hypothesis
They specifically mention the urban heat island in the first line of the intro to their paper. The data heat island effect is named after it.
I wasn’t around for the initial water panic arguments, but I hope to use the knowledge I gained from this to challenge my anti ai friends. Hopefully this disinfo doesn’t spread as easily, thanks for writing this.
Hah, I saw the headline this morning, read that the study hadn't been peer reviewed yet, and filed it under "check back in when it's been peer reviewed". Thanks for your service.
This isn't what peer reviewed means. This is an opinion piece propped up by chatbots themselves.
He means the original study?
He seems to consider your blog post here a peer review.
That's not so, but I can see how that may have been misleading. I thought about fact checking the article but didn't bother until it had at least been peer reviewed (which usually means reviewed prior to appearing in a journal, just to define terminology - and some journals have higher standards than others) - Andy fact checked it instead and satisfied my curiosity.
Andy wrote an opinion piece on it and showed pretty basic misunderstanding of building science and construction in general.
Great piece. I really hope this becomes the common understanding as opposed to the nonsense in the paper. I think it will
nitpick: i dont think this sentence is true:
> the red line is the average, and the error bars represent the 95th percentile bounds, meaning the vast majority of measurements fell within those much smaller bounds
the paper itself seems a bit unclear about what the range represents, but if those are error bars for the mean, then all datapoints could in theory be outside of that range.
From the paper: "The shaded areas show the interval between the maximum and minimum value of LST increase that has been recorded across the considered AI hyperscalers. Finally, the bar across the average line identifies the limit of the 95th percentile of the distribution we compute."
I am with you on "this is wrong". I am with you on "buildings are hotter than grass". But I think there is still something missing here.
Look at that goofy graph again. The timeline on the x axis shows that ~all of the effect happens in ONE MONTH. The normalized temperature increase goes from 0 to 0.2 C from -10 months to t=0, and then jumps up to 2.2 C from month zero to month one and then after that it stays pretty flat, only increasing by half a degree over the remaining 9 months.
That's not what I would expect from an increase in development. That looks like some kind of coding error. With no publicly available code and data right now (MODIS LST is public but it wouldn't be trivial to recreate the specific areas they looked at), I can't see what it is but I think there is some other problem here than what you've identified so far.
I suspect this might actually be because the main thing the satellite is picking up, the roof, is one of the last things to go up, and they're setting the baseline a few months before the DC went up when it's still being constructed, so the baseline is already where there's building present, but the roof causes the jump?
Take a look at an image of a finished datacentre, like the one here: https://www.distilled.earth/p/these-data-centers-are-getting-really?hide_intro_popup=true
The majority of changes in the physical landscape are not due to the buildings themselves, but the large amounts of roads and buildings around it. Do you really think adding a roof on at the end will make enough of a material difference to produce the massive jump in temperature shown on the graph?
The physical construction time for a data-centre is something like 1-2 years: if the physical concrete of the buildings was what caused the effect you would expect to see temperatures steadily rising toward the maximum value many months before the data-centre actually started operations: not a miniscule rise followed by a massive jump.
If this graph were true it very much would support the papers conclusions that AI datacentre operations were to blame, rather than the buildings themselves. But the finding seems pretty odd given what we know about urban heat island effects and the energy emitted by datacentres: skepticism is warranted.
The vast majority of the change I'm seeing in these images is grass -> dirt, which I wouldn't imagine would have much of an effect? It seems like they expanded the roads a bit but I'm not seeing the roads take up anywhere near as much surface area as the DCs
I think grass to dirt is actually the biggest part of the effect! Plants have a major cooling effect, because they use some of the sunlight to do the work of photosynthesizing, and more importantly use heat to evaporate water from their leaves, causing capillary action to suck water up from their ground, which is their equivalent of pumping blood to get nutrients from one part of their body to another. Grass doesn’t use as much water this way as trees, so it doesn’t convert as much heat to water evaporation, but it still does a good amount, and dirt doesn’t do any.
A quick search found one article trying to compare the albedo effect to the transpiration effect, and finds that in particularly dry climates, the albedo effect might be larger. But in the kinds of places where the ground is normally covered with vegetation, transpiration is going to be larger.
https://www.science.org/doi/10.1126/sciadv.aea9165
Even if you exclude the dirt, about half the area is coming from roads and parking lots. And even without the roof, the data centre is still there, being a giant chunk of concrete: it seems implausible to me that a roof alone would make that much of a difference.
Plus, do you have any evidence that data centre operation starts within a month of the roof going on the top? Like, it seems like it would take longer than that to install all of the internal equipment and data-racks and so on.
It does not seem likely to me that your theory alone explains the graph.
Your assumption is wrong. The bulk of building material that would be picked up from a thermal/satellites perspective is BY FAR the concrete (and it isn't even close). That is all on site/in place during the first half of a project. If anything the roof would insulate and "hide" the thermal change from the satellites but at that point in the project the roof going up is basically irrelevant.
Do you have a source on this? Most of what I'm reading seems to imply the opposite
I'm an architect with an understanding of materials and building science. Data centers are like the perfect building science study because they are simple structures with extremes for the inputs so things are much more binary than a building with a complicated footprint, an inconsistent envelope or mechanical loads. You are correct in thinking that the passive heat loads on these sites is massive, but the roof going on at the end of a project isn't making a difference. Roofs specifically are made from high albedo material (if its white this is the case), and if anything would reduce the heat being measured from a satellite.
As with most things, the truth is likely somewhere in between.
Any thoughts on what's up with the jump?
Hard to not look at the waste heat from the cooling systems. All of that would come online at the completion of the building-the only time the building starts turning megawatts into waste energy. Nothing else late in the building process (or at any time during construction to be honest) is really going to change anything that much. If this were all just essentially heat island effect, the graph would look like a staircase, not a cliff.
I gave both the paper and your post a very quick read, I'll try to be more thorough later.
i dont think i understand how the building construction itself could result in such a sudden (<1 month) change in temperature? a large building takes more than a month to construct, right?
I suspect this might actually be because the main thing the satellite is picking up, the roof, is one of the last things to go up, and they're setting the baseline a few months before the DC went up when it's still being constructed, so the baseline is already where there's building present, but the roof causes the jump?
We're already discussing this in the other thread - but I want to be clear, this is completely wrong from a construction timing and building science standpoint. The baseline already exists before the roof goes on. If anything, the roof will reduce the heat signature when it goes on.
You and Claude mention "dark-roofed" buildings, but *are* they dark? When I see large commercial buildings, the roofs are generally white.
Given how good the LLMs can be at arguing a case (any case), it might be useful to prompt it the other way, or have it critique its critique. Only prompting it one way (to find flaws in the paper) may leave you with some confirmation bias.
It would be interesting to try to reproduce but substitute large warehouses or something for data centers and see how large the effect is. If you get 98% of the effect (or whatever it is) from a warehouse, that's a simple story to tell to substantiate the "it's just construction" explanation.
This whole thread is a very interesting read as someone who is interested in AI and tech but has a professional background in buildings and building science.
This is the exact intersection I need!
I think the bigger effect is removal of vegetation - plants move nutrients by forcing water evaporation (which is why they need so much water) and this process converts a lot of heat into vapor.
Thank you for doing the work most of us don't bother to do! It'd be really useful to know how you personally use AI in research, in depth.
As always, thank you for taking the time and effort to pick this apart.
Activists really seem to be losing their minds about data centers and AI in general, either to the point that they're just not thinking clearly (the "system 2 motivated reasoning" phenomenon) or worse, that ends justify means. It's an article of faith that "all this will pass", and the longer it doesn't, the more the panic rises.
Among the many depressing harms that this type of poor research is doing, it's already destroyed many people's heuristic - including mine - that science with a pro-environmental slant can be taken to be largely trustworthy, honest mistakes aside.
Curious, did you actually try Gemini for this one? I plugged it into 3.1 Pro and it claims it is legit, but that may be a different response.