"In some instances, AI data centres are powered completely by renewable energy. Unfortunately, unless the data centre builds new renewable energy sources to supply 100% of its power, this still results in an increase in fossil fuel usage. Why? Because if the data centre is using pre-existing sources for renewable energy, it is taking that energy away from other consumers that need it. To pick up the slack, we must generate more energy, and most of that generation is done using coal or natural gas."
-- strangely I had never actually thought about this
Because it's a very strange way to view the economics. Yes, AI uses energy. But using energy itself isn't an inherently bad thing. It'll also incentive people to invest more in renewable energy too.
> Additionally, many residents who live near AI data centres are seeing increased electricity bills due to the amount of electricity those data centres use.5 That's right, trillion dollar companies are using you to subsidize their power bill!
I'm glad the author put this paragraph early into this article though. Saved me a few minutes reading the rest.
His argument proves far too much: it would apply even if every data center built an entirely renewable dedicated solar farm to power it. After all, energy is fungible: the newly built solar farm could be going to help consumers transition to renewable.
There is a root issue here: prices of nonrenewables don't incorporate the costs of climate change. But that's a much broader issue, and otherwise the market price system is excellent at allocating energy to its most economically beneficially uses.
>it would apply even if every data center built an entirely renewable dedicated solar farm to power it. After all, energy is fungible: the newly built solar farm could be going to help consumers transition to renewable.
No, because that solar farm would have never been built in the first place without the datacenter. You either get both or you get none of them. It's completely different than using an existing solar farm that was already there before anyone started planning for that datacenter
There is a reason for that. Its not especially true. In fact, the more renewables you deploy, the more natural gas you use. And that's before you consider that most renewables are sited so poorly that it likely doesn't reduce CO2 emissions at all. This guy also seems to think that water is somehow destroyed when its used for cooling. This also isn't true. I really wish the author hadn't started with these claims that he clearly doesn't understand. There are good points much farther down, but he buried the lead under 20 feet of misunderstandings.
Yes, AI datacenters do actually use evaporative cooling. The water isn't destroyed obviously but it is made unavailable for other purposes (like keeping people, animals, and plants alive) and depletes water tables that can't be easily replenished
I hope you are also inveighing against industrial agricultural use of water, which uses orders of mangitude more water in significantly more wasteful ways.
Eh, somewhat reductive thinking. Demand for renewables causes an increase in the build out of new renewables. Yea, it's not instant, but that's how markets work. Coal/fuel is expensive, so this drives demand for more wind solar, especially during high demand times that bring the highest power prices.
The problem with this essay is thinking power demand is bad. It is not when we can deploy massive amounts of renewables. Not only for AI, but for other industrial workloads that user power.
If the rest of the dipshits around here hadn't elected the dumbest mother fuckers on earth, we'd be pushing out subsidies to build out renewables at 10x the rate we are now. But hey, coal jobs matter.
The text said "some instances". If you're next to a hydropower dam and are willing to pay for it, you can outbid others, who then switch to non-renewable sources. The dam can only put out so much energy. Ideally you'll switch to solar, but as various sources point out, it's not just adding solar power but also interconnect capacity, which is why a lot of data centers are using natural gas. Your "not instant" is many years, and longer than the timeframe these companies want to wait.
"Traditional hubs are at or near saturation, which has created long connection queues with waits of typically seven-to-ten years and in some cases 13-15. The International Energy Agency (IEA) estimates that globally, nearly 20% of planned data center projects risk delay due to grid bottlenecks, and as a result developers are building ‘behind the meter’ primary generation. This offers speed to market in months rather than years.”"
"gas is mainly chosen because it’s the fastest and easiest way to generate your own power. Turbines and engines can be set up and begin providing reliable energy quickly, while renewable power generation can be intermittent without costly storage – and takes time and money to be set up."
"Global data center electricity consumption is projected to more than double by 2030, driven primarily by AI training and inference workloads. The IEA estimates data centers could consume over 1,000 TWh annually by 2030 — roughly equal to Japan's total electricity consumption."
"Grid interconnection queues are now 4–7 years in many US markets, making behind-the-meter gas turbine generation the fastest path to powering new hyperscale campuses."
"Microsoft, Amazon, and Google have all signed gas turbine power agreements or acquired generation assets in 2024–2026, signalling a structural shift in how hyperscalers think about electricity procurement."
"The gas turbine data center trend is not temporary. Even as renewable procurement grows, the intermittency gap and the sheer scale of AI power demand mean firm, dispatchable generation will remain essential for Tier III and IV reliability standards."
> Eh, somewhat reductive thinking. Demand for renewables causes an increase in the build out of new renewables. Yea, it's not instant, but that's how markets work.
Funny how this line of thinking is indistinguishable from crypto.
> It is not when we can deploy massive amounts of renewables. Not only for AI, but for other industrial workloads that user power.
Funny how this line of thinking is indistinguishable from crypto. "Yes, we build a lot o power which is immediately consumed by us, but see how this is good? See? SEE?!"
> we'd be pushing out subsidies to build out renewables at 10x the rate we are now.
And these would be gobbled up by crypto and AI data centers, right?
I get that these data center companies are in a race right now and speed is the name of the game. But if they really want long-term acceptance for this technology, I think they will need to basically be fully self-sufficient with renewable energy, in a completely isolated power grid, so that we don't have to deal with the above issue.
AI is clearly useful, but usefulness doesn't automatically justify unlimited deployment, opaque training practices or turning every public service and workplace into an experiment
I don't understand the water argument? I thought places like data centers would use something akin a closed loop and a radiator. Are they taking water from the town pipe, heating it once, and then launching it into the sun? Is there some other use for the water other than cooling?
Apparently, data centers typically use evaporative cooling (I guess like a Swamp Cooler but on a whole different scale). This is cheaper than using compressed systems like we use on our homes. The water evaporates.
At a global level the water stays around but at a local level it "vanishes into thin air".
Most AI data centres use evaporative cooling. One of the linked sources talks about this; that 7 billion cubic metres metric is how much water gets evaporated into the air. (which makes that water no longer usable for drinking/watering crops/etc)
Don't forget about my favorite part of chilled water loops—the toxic chemicals that lowest bidders are contracted to dispose of.
Source: Environmental and Energy Study Institute (EESI)
> Evaporation Process: Roughly 80% of the water evaporates, cooling the servers, while minerals, microorganisms, and chemical additives (like anti-corrosion or anti-scaling agents) remain in the remaining 20% liquid, which is later discharged.
A large portion of data centers use open loop cooling. That is to say evaporative cooling. So they take fresh drinking water, make it hot, and dump it in the air.
It's a lot cheaper than closed loop cooling.
I’d like to see some evidence of that, adding evaporative cooling towers to a chilled water loop can almost double the COP. I don’t see datacenter operators foregoing much cheaper opex unless they’re in a water-critical area.
>Do you think that people like Sam Altman, Elon Musk, or Dario Amodei really have our best interests at heart...
The great thing about market competition is the bosses may be bastards but they are forced to make good products so as to compete with other companies, hence pretty good cars and satellite coms from Musk for example.
I'm cheered that the AI companies have little moat and stiff competition amongst themselves.
>Asking ChatGPT a question uses approximately 10x as much energy as a traditional search engine
I stopped reading here. It is more energy efficient to have ChatGPT write the code than a developer on their workstation. It in fact can save energy by having AI do it.
Unlike a lot of anti-AI writing I have seen which can drift into the territory of reactionist rhetoric, this feels very grounded. As a student entering the job market, I personally am quite conflicted about my usage of AI, due to all the effects you have said. Especially when testing out more recent “agentic” coding tools where they do literally all the work, not only are there the moral qualms but they’re also not enjoyable to use.
However, in the tech bubble I live in to use these tools for as much as possible in order to “make it” in some sense and actually be employable. Like you said it is a game of chicken. Perhaps the best strategy (for me) is to campaign for institutional guardrails on usage while continuing to individually try to be competitive?
So much written about this but nothing about anti-ai-psychosis, which this article is a clear example of.
There is no possible response to any of the issues raised would make a whit of difference to the author, since they have made a quasi-religious and certainly political determination that anything associated with AI is bad and should be avoided.
So the article itself can be basically ignored, it’s not written in good faith, it’s essentially propaganda in service of a fixed political viewpoint.
I disagree, if you'd like to propose solutions to these problems as well as an implementation strategy, I'd be very happy to hear them! Source: am the author.
"In some instances, AI data centres are powered completely by renewable energy. Unfortunately, unless the data centre builds new renewable energy sources to supply 100% of its power, this still results in an increase in fossil fuel usage. Why? Because if the data centre is using pre-existing sources for renewable energy, it is taking that energy away from other consumers that need it. To pick up the slack, we must generate more energy, and most of that generation is done using coal or natural gas."
-- strangely I had never actually thought about this
Because it's a very strange way to view the economics. Yes, AI uses energy. But using energy itself isn't an inherently bad thing. It'll also incentive people to invest more in renewable energy too.
> Additionally, many residents who live near AI data centres are seeing increased electricity bills due to the amount of electricity those data centres use.5 That's right, trillion dollar companies are using you to subsidize their power bill!
I'm glad the author put this paragraph early into this article though. Saved me a few minutes reading the rest.
His argument proves far too much: it would apply even if every data center built an entirely renewable dedicated solar farm to power it. After all, energy is fungible: the newly built solar farm could be going to help consumers transition to renewable.
There is a root issue here: prices of nonrenewables don't incorporate the costs of climate change. But that's a much broader issue, and otherwise the market price system is excellent at allocating energy to its most economically beneficially uses.
>it would apply even if every data center built an entirely renewable dedicated solar farm to power it. After all, energy is fungible: the newly built solar farm could be going to help consumers transition to renewable.
No, because that solar farm would have never been built in the first place without the datacenter. You either get both or you get none of them. It's completely different than using an existing solar farm that was already there before anyone started planning for that datacenter
There is a reason for that. Its not especially true. In fact, the more renewables you deploy, the more natural gas you use. And that's before you consider that most renewables are sited so poorly that it likely doesn't reduce CO2 emissions at all. This guy also seems to think that water is somehow destroyed when its used for cooling. This also isn't true. I really wish the author hadn't started with these claims that he clearly doesn't understand. There are good points much farther down, but he buried the lead under 20 feet of misunderstandings.
"the more renewables you deploy, the more natural gas you use." -- source?
"most renewables are sited so poorly that it likely doesn't reduce CO2 emissions at all" -- source?
I don't think this even requires a source, it's completely made up.
Yes, AI datacenters do actually use evaporative cooling. The water isn't destroyed obviously but it is made unavailable for other purposes (like keeping people, animals, and plants alive) and depletes water tables that can't be easily replenished
I hope you are also inveighing against industrial agricultural use of water, which uses orders of mangitude more water in significantly more wasteful ways.
Shouts "AI wastes water!" with food in my mouth while eating an almond butter sandwich with a pizza lined up next.
Eh, somewhat reductive thinking. Demand for renewables causes an increase in the build out of new renewables. Yea, it's not instant, but that's how markets work. Coal/fuel is expensive, so this drives demand for more wind solar, especially during high demand times that bring the highest power prices.
The problem with this essay is thinking power demand is bad. It is not when we can deploy massive amounts of renewables. Not only for AI, but for other industrial workloads that user power.
If the rest of the dipshits around here hadn't elected the dumbest mother fuckers on earth, we'd be pushing out subsidies to build out renewables at 10x the rate we are now. But hey, coal jobs matter.
The text said "some instances". If you're next to a hydropower dam and are willing to pay for it, you can outbid others, who then switch to non-renewable sources. The dam can only put out so much energy. Ideally you'll switch to solar, but as various sources point out, it's not just adding solar power but also interconnect capacity, which is why a lot of data centers are using natural gas. Your "not instant" is many years, and longer than the timeframe these companies want to wait.
https://www.itpro.com/infrastructure/data-centres/gas-powere...
"Traditional hubs are at or near saturation, which has created long connection queues with waits of typically seven-to-ten years and in some cases 13-15. The International Energy Agency (IEA) estimates that globally, nearly 20% of planned data center projects risk delay due to grid bottlenecks, and as a result developers are building ‘behind the meter’ primary generation. This offers speed to market in months rather than years.”"
"gas is mainly chosen because it’s the fastest and easiest way to generate your own power. Turbines and engines can be set up and begin providing reliable energy quickly, while renewable power generation can be intermittent without costly storage – and takes time and money to be set up."
https://www.greengasturbines.com/blog/gas-turbines-for-data-...
"Global data center electricity consumption is projected to more than double by 2030, driven primarily by AI training and inference workloads. The IEA estimates data centers could consume over 1,000 TWh annually by 2030 — roughly equal to Japan's total electricity consumption."
"Grid interconnection queues are now 4–7 years in many US markets, making behind-the-meter gas turbine generation the fastest path to powering new hyperscale campuses."
"Microsoft, Amazon, and Google have all signed gas turbine power agreements or acquired generation assets in 2024–2026, signalling a structural shift in how hyperscalers think about electricity procurement."
"The gas turbine data center trend is not temporary. Even as renewable procurement grows, the intermittency gap and the sheer scale of AI power demand mean firm, dispatchable generation will remain essential for Tier III and IV reliability standards."
> Eh, somewhat reductive thinking. Demand for renewables causes an increase in the build out of new renewables. Yea, it's not instant, but that's how markets work.
Funny how this line of thinking is indistinguishable from crypto.
> It is not when we can deploy massive amounts of renewables. Not only for AI, but for other industrial workloads that user power.
Funny how this line of thinking is indistinguishable from crypto. "Yes, we build a lot o power which is immediately consumed by us, but see how this is good? See? SEE?!"
> we'd be pushing out subsidies to build out renewables at 10x the rate we are now.
And these would be gobbled up by crypto and AI data centers, right?
I get that these data center companies are in a race right now and speed is the name of the game. But if they really want long-term acceptance for this technology, I think they will need to basically be fully self-sufficient with renewable energy, in a completely isolated power grid, so that we don't have to deal with the above issue.
AI is clearly useful, but usefulness doesn't automatically justify unlimited deployment, opaque training practices or turning every public service and workplace into an experiment
when is something automatically justifible
AI, like the Y2K "bug", has been extremely useful for driving tech refresh.
> Asking ChatGPT a question uses approximately 10x as much energy as a traditional search engine
Traditional search engines use vanishingly little energy, and 10x vanishingly little is still very small.
> It is estimated that AI alone will evaporate between 4 and 7 billion cubic metres of water per year in 2027
ChatGPT has 1 billion users every week, so AI is adding 1% to their water usage?
I don't understand the water argument? I thought places like data centers would use something akin a closed loop and a radiator. Are they taking water from the town pipe, heating it once, and then launching it into the sun? Is there some other use for the water other than cooling?
Apparently, data centers typically use evaporative cooling (I guess like a Swamp Cooler but on a whole different scale). This is cheaper than using compressed systems like we use on our homes. The water evaporates.
At a global level the water stays around but at a local level it "vanishes into thin air".
Most AI data centres use evaporative cooling. One of the linked sources talks about this; that 7 billion cubic metres metric is how much water gets evaporated into the air. (which makes that water no longer usable for drinking/watering crops/etc)
Don't forget about my favorite part of chilled water loops—the toxic chemicals that lowest bidders are contracted to dispose of.
Source: Environmental and Energy Study Institute (EESI)
> Evaporation Process: Roughly 80% of the water evaporates, cooling the servers, while minerals, microorganisms, and chemical additives (like anti-corrosion or anti-scaling agents) remain in the remaining 20% liquid, which is later discharged.
A large portion of data centers use open loop cooling. That is to say evaporative cooling. So they take fresh drinking water, make it hot, and dump it in the air. It's a lot cheaper than closed loop cooling.
A "large portion" of data centers needs to be quantified.
Particularly, newer data centers are much more likely to used closed loop systems. And, the bigger they are, the more likely they're on closed loop.
I’d like to see some evidence of that, adding evaporative cooling towers to a chilled water loop can almost double the COP. I don’t see datacenter operators foregoing much cheaper opex unless they’re in a water-critical area.
Yeah, this article is full of misinformation. The water argument is only one example.
>Do you think that people like Sam Altman, Elon Musk, or Dario Amodei really have our best interests at heart...
The great thing about market competition is the bosses may be bastards but they are forced to make good products so as to compete with other companies, hence pretty good cars and satellite coms from Musk for example.
I'm cheered that the AI companies have little moat and stiff competition amongst themselves.
I think the case is well-made, but it's not a game of chicken, it's an arms race.
>Energy Use
>Asking ChatGPT a question uses approximately 10x as much energy as a traditional search engine
I stopped reading here. It is more energy efficient to have ChatGPT write the code than a developer on their workstation. It in fact can save energy by having AI do it.
Unlike a lot of anti-AI writing I have seen which can drift into the territory of reactionist rhetoric, this feels very grounded. As a student entering the job market, I personally am quite conflicted about my usage of AI, due to all the effects you have said. Especially when testing out more recent “agentic” coding tools where they do literally all the work, not only are there the moral qualms but they’re also not enjoyable to use.
However, in the tech bubble I live in to use these tools for as much as possible in order to “make it” in some sense and actually be employable. Like you said it is a game of chicken. Perhaps the best strategy (for me) is to campaign for institutional guardrails on usage while continuing to individually try to be competitive?
> AI Psychosis
So much written about this but nothing about anti-ai-psychosis, which this article is a clear example of.
There is no possible response to any of the issues raised would make a whit of difference to the author, since they have made a quasi-religious and certainly political determination that anything associated with AI is bad and should be avoided.
So the article itself can be basically ignored, it’s not written in good faith, it’s essentially propaganda in service of a fixed political viewpoint.
I disagree, if you'd like to propose solutions to these problems as well as an implementation strategy, I'd be very happy to hear them! Source: am the author.
You are the person I’ve written off as unreachably deluded per my comment above, so I don’t see much purpose in engaging with you.