Artificial intelligence is not virtual, clean, or weightless. It has a rapidly escalating physical cost in electricity, water, and emissions—and ordinary people will pay the price.
Research shows that AI data centres could soon consume electricity on the scale of entire nations. At the same time, AI cooling systems are diverting vast quantities of water in a world already facing severe shortages.
This video asks the questions politicians are refusing to confront: who pays for AI's energy and water use, who profits, and whether unlimited AI growth is compatible with planetary limits, democratic accountability, and basic human needs.
AI may promise growth—but at what cost, and to whom?
This is the audio version:
This is the transcript:
AI is threatening us on many economic fronts. Of course, the AI industry thinks otherwise, but the truth is that there are real challenges that AI is creating, which most of our politicians are totally unable to appraise or even talk about. One of those threats, the one I want to talk about now, is the crisis that it represents for energy and water.
The fact is that artificial intelligence has a physical cost. That cost comes in the form of electricity and water consumption and emissions created, and someone is going to have to pay for these things, and it could be you, and that's why this matters.
AI is sold as virtual, clean and weightless. It's sold as if it is a solution to all our problems, but in reality, it is massively energy-hungry, and the scale of the demand that it is making is exploding very fast.
Researchers at the Massachusetts Institute of Technology in the USA, MIT, one of the best-respected universities in that country, have estimated that the demand for energy from AI will, by 2026, represent the total energy usage of either Japan or Russia. And by 2030, they reckon that at least one 10th of global electricity demand could be for AI data centres. This is fundamentally destabilising of the world energy system and of the demands that we make upon the environment to create them.
As Noman Bashir, who is based at MIT, has concluded, the demand for new data centres cannot be met in a sustainable way, and that matters when the world is already facing a climate crisis. The pace of expansion of AI means that fossil fuels are going to have to fill the gap that it is creating for more electricity, and that directly contradicts every climate target that we've got.
The hyperscale of AI is changing everything. Traditional data centres, the ones that we've been using to date, have been using around 25 megawatts of power. The new hyperscale AI centres are going to 100 megawatts of power, and that by itself is enough to power 100,000 households, but these are now only the modest AI sites. The newest and largest announced data centres might consume enough power to provide electricity for 5 million households. This is not marginal demand; it is system-shifting.
We are facing a complete change altogether in the scale of electricity generation required to keep our economy going if it is to be based upon AI. And there is no way that we know how to deal with this.
US data centres are driving almost half of new demand for power in the USA.
In Japan. It's also more than half of new demand for power.
And in Malaysia, one fifth.
So this is a worldwide issue. It's also true that this is happening right across Europe, we know that. Energy planning is being reshaped around AI, but our national electricity systems can hardly bear the strain.
Nor can prices, because this is going to result in rationing, and rising demand is inevitably going to push up electricity prices. That's the inevitable consequence and reaction that we're going to see, not least because the energy infrastructure that is required by AI is enormous, and grid upgrades are going to be costly and somebody is going to have to pay, and that is going to include consumers and not just AI firms.
So we are all going to pay for this, and when affordability is already one of the key issues in elections around the world, this is going to matter electorally. People are not going to be happy.
We know that electricity prices will be a major US midterm election issue later this year, and alongside this electricity crisis, there is another crisis, and that is the hidden water crisis.
AI is going to consume vast volumes of water. Now, of course, it doesn't destroy it, but it requires it, and it has to be in the place where the data centre is, and the recycling of that water once it's been used to cool these places is going to take time and is not going to be easy to manage. The point is that water is going to have to be diverted to these places from where it is currently used, and we already know that water is in desperately short supply the whole world over. Climate change and the failure to do capital maintenance is already destroying supplies of water, not least in the UK, and AI is simply going to make this worse.
Data centres use chilled water for cooling. Every unit of AI output has a water footprint. It's thought that for every kilowatt hour of electricity consumed by a data centre, two litres of water is required. This is water usage on a scale that's very hard to imagine, given the vast amount of electricity that is going to be consumed by these places.
So we have a massive global water impact going on, as well as a massive global energy impact going on, and all of this is happening now. It is reckoned that very soon, AI in Europe will use six times more water than Denmark, that's enough water for 30 million people at least, and this can only grow. In a world where one quarter of humanity already lacks clean water, AI is simply going to tip the balance of power in favour of large corporations and away from ordinary people in a way that is almost unimaginable.
And, of course, there are environmental consequences. There will be strains on local water supplies. There will be disruption of ecosystems. Agriculture will be disrupted. So will the biosphere, and nature, and species and everything else, whilst alongside this, there will be increased emissions from fossil fuels, so heat will be rising. None of this is priced into AI services. Nor is it in our thinking on AI and the waste of investment that it might represent, because we may never be able to see the data centres that they're planning put into use precisely because physical constraints will stop that happening.
There could be a financial crisis as a consequence, just to top these absurd environmental demands and who benefits and who pays? Well, the big tech firms are, of course, trying to capture the profits and energy companies will raise their prices, and we will pay. AI is not going to come cheap; AI is going to come at a cost, which is going to be enormous, and many communities will find that they tip over the edge. They will no longer have water security, and water is essential to life.
This is market failure on a dramatic scale. AI firms will never pay the full cost of the imposition that they are going to make upon us. Energy and water costs are going to be socialised, and profit is going to be privatised, and this is the recipe for a classic market failure, and the state cannot ignore this. Electricity grids are public systems. Water is a public resource, and climate stability is a public good. Leaving these things to the market is reckless, but that's what's going on with politicians supporting AI as if it is their salvation, because it delivers the promise of growth on which they hang all their hopes and expectations, quite recklessly, and the consequence is we are all going to suffer.
Questions have to be asked: very deep and real questions.
Should AI, for example, face energy rationing? Why not? It's creating the energy shortage. Why should they have a right to energy when real-life human beings may not get it?
Should data centres pay the additional costs that they will be imposing upon energy systems? Why not? They want the profits; they should pay the costs.
And should water use by AI be regulated or even capped so that there is a limit on the scale by which AI can grow simply because they will not be able to secure the water they need to cool the data centres that drive the whole process? Again, why not? Is AI such an important thing that we can afford to let the world go thirsty?
Planning, not panic, is required. I'm not anti-technology. I use AI; I can see some benefits from it. So let's be clear, I'm not saying that this thing is universally bad, but we do need to work out our priorities, and that is what planning is required for.
Should tech be allowed to make people jobless?
Should tech be allowed to deny people the basics of life that are essential for all?
Should tech be allowed to use these resources without accountability?
These are questions that are real because AI has to ultimately fit within our planetary limits, and all the indications are that it won't.
In that case, AI is quite simply not sustainable. This supposed information that it's going to generate is not worth the cost, and because AI is not immaterial, simply because the scale of investment that is being thrown at it, this is going to create massive physical consequences. And those consequences are accelerating, and more than that, they may be so unsustainable that the questions have to be asked now, and not be put on the agenda when it becomes apparent that the costs are real.
AI's energy and water costs are real. We know that they're going to happen. We have to address them. That's because they will hit consumers and ecosystems very hard indeed, and potentially fatally, and I mean the word literally in this case. Without state intervention, damage is inevitable.
The question is not whether AI will grow, because at the moment, it clearly will. The question is, who's going to pay for that growth? Who will benefit and who decides, and how do we decide the ratios of power that are implicit within the answers that we provide to those questions, because this is a core issue in political economy, as a result?
The question is also, what gives? Because something is going to have to, and that again is key. Do we care? Or do we literally focus upon the politics of might and power, which is what AI is all about? These are questions needing resolution.
I know where I am biased. What do you think? Is AI going to be worth the cost it might impose upon us?
There's a poll down below.
Poll
Loading ...Comments
When commenting, please take note of this blog's comment policy, which is available here. Contravening this policy will result in comments being deleted before or after initial publication at the editor's sole discretion and without explanation being required or offered.
Thanks for reading this post.
You can share this post on social media of your choice by clicking these icons:
There are links to this blog's glossary in the above post that explain technical terms used in it. Follow them for more explanations.
You can subscribe to this blog's daily email here.
And if you would like to support this blog you can, here:
Buy me a coffee!
Its interesting to see the current controversy around the use of Grok to create indecent images.
A reflection on the mindset of those behind it both in that Grok allows it AND they never saw the risk coming.
Also worth making the point that even the dear old internet comes with an energy cost BEFORE factoring in AI
Finally it is suggested that there could be a 25% rise in the price of ‘Tech’ phones, laptops etc because of the massive demand for AI chips – something that may well attract a reaction from Consumers and Governments
Thank you Richard, as always very clear and full of great action points. I think rhere needs to be more clarity on what AI itself is. I see a difference between LLMs ( essentially a scrape of the whole internet, stealing everyone’s inputs), which generate the output for Chat GP etc and analytical AI used for advancing scientific progress. Both need more thorough cost benefit analysis done, rather than this massive bet by the tech companies, driven by their endless need for “growth” to justify sky-high valuations.
Might it be the case that preparations for the introduction/selling of AI presents a fundamental flaw/deceit by the government and the entourage of the main stream media which is to present only one side of the matter aka distorting single factor analysis?
AI does not break down because the models stop improving. It breaks down where electricity grids, water systems, ecosystems, and household budgets reach their limits. Those limits are social and political as much as technical.
The decision to expand AI is taken by a small group of firms and supported by governments chasing growth and prestige. The consequences land elsewhere: in higher household bills, water restrictions, degraded environments, and public infrastructure stretched beyond design limits. That gap between decision making and cost bearing is the defining failure.
Once energy and water affordability become politically intolerable, AI growth will not be constrained by innovation or market demand, but by public resistance and physical scarcity. That is what “social tolerance” means in practice.
Core conclusion is not anti technology. It is about accountability. If the people who decide AI’s scale do not pay for its infrastructure stress, then the system will continue to over expand until it collides with public limits. Planning, pricing, and democratic control are the only ways to prevent that collision.
Thanks. Agreed.
You are right – once again, a new market does not do its double entry book keeping and takes even more from the commons as it is supposed to give back. Result? Exactly what you have set out above.
In my job as housing development the developer has to pay for new drainage systems and even provide electricity substations to boost electricity supplies to not overload existing systems out of the costs of the schemes. Why should AI developers be any different? I suspect all sorts of slimy reasons will be found to get out of this.
Thanks for this and other comments this morning.
Thank you for enabling me to do so.
With 40-odd years in the energy industry covering O&G to nuclear to Renewables, I think the downside risks (as pointed out in.the article) massively outweigh the benefits (unless you consider Tech-bros making even more stupid amounts of money to be a benefit).
I think this very simple – close to home example – illustrates the danger: Data centres in Ireland consumed 21% of the nation’s total metered electricity in 2023. The consumption of urban households was 18% and rural homes 10%. Unsurprisingly, it is raising energy supply concerns. Forecasts are already suggesting data centres could use over 30% of Ireland’s electricity by the end of the decade…I think it will be higher. Now think of all the jobs created in these data centres, i.e., not many. Basically, building electrical and cooling water infrastructure to support foreign profits with little benefit to society and a significant negative impact on climate change. Please remember, as it stands, to have grid stability you still need conventional (or nuclear or hydro) power stations. In a word…bonkers!
Having said that, it less bonkers that the UK governments idea to create an AI super cluster between Oxford and Cambridge powered by SMRs (Small Modular Reactors). Putting aside the SMR technology challenge, where is the water coming from? Utterly bonkers.
I wish politicians would have a look at basic physics and available resources (such as water) and assuming it got past these basic tests do a decent techno-economic assessment. As is increasingly demonstrated, nearly every idea on using Hydrogen, Ammonia, etc., fails when you do the basic physics – far less the engineering – and the utterly fails when it gets to techno-economic viability. They need to put AI and data centres to the same tests at which point they utterly fail when it comes to overall societal benefits including climate impacts.
They believe William Nordhaus. See the economics questions series.
I agree with this, although I’ll need to re-read the piece properly to ensure I’ve not missed anything important.
Warning – incoming nonsense!
I have two thoughts, one with practical background, the other rather extreme. In your poll, I would agree that some action around limiting or regulating AI needs to be taken. Soon.
As you have indicated, data-centres do two things:
>> They provide internet services – not only AI. Although that’s a massive driver of data-centre expansion, ordinary internet services fall within what I’m saying
>> They turn massive amounts of electricity into heat. It almost feels like this is the primary product, with AI/Internet being a by-product!
The latter issue could be mitigated (not solved, btw) by harvesting this heat to give a second bite of the energy cherry. This is already being done, and here are two examples. The ‘Bunhill 2’ project in London harvests warm air from the underground to heat blocks of flats and leisure facilities. A company called ‘Deep Green’ has built a data-centre that uses oil instead of water for primary cooling and provides heat and hot water to a local leisure centre by secondary cooling (https://www.bbc.co.uk/news/technology-64939558). The sometimes-maligned heat pumps are used for this.
This is where regulation comes in – IF we allow a data-centre to be built (any maybe we should be tougher), we should insist (via planning law, perhaps) that its heat management maximises harvesting for secondary, public, use (ideally, for free). It is even possible to generate some electricity from the hot cooling liquids.
That’s the ‘practical’ solution, where the technical capability exists already. I’m not sure how realistic it is, though, politically speaking. The IT industry seems to have a huge amount of power in terms of resisting being told how to behave. We seem to lack the political will to stand up to this. There’s also the issue of infrastructure (political will again). None of this is simple.
The extreme thought is to get the data-centres to be put up into orbit. There you will find limitless solar energy, coupled with easy dumping of heat. Yes, I acknowledge that we arguably have (more than?) enough hardware up there, but that kind of makes my point that the technical capability exists already. This is not really meant to be taken entirely seriously, btw, but…
Clearly secondary use helps. That is indisputable. Mike Parr has made the point here, often. But the secondary use is that – secondary and the laws of thermodynamics apply. And if the primary use was useless, is the secondary use that beneficial?
I agree with what you say. I will, though, add the following.
The rise in AI is a recent phenomenon, hugely accelerating the demand for computing and hence energy and cooling. It has served, if nothing else, to bring our attention to the issue. My comments about heat harvesting apply also to the server farms/data centres that existed before AI came to prominence. We’ve all been using email, reading your blog, using facebook or LinkedIn, doing our taxes, and ordering our stuff on-line etc. etc. for years, and for all of that time the servers have been using electricity and pumping out heat, without controls or much comment. I suspect that at least some of these services have some value to some people (point taken about memes, etc.).
It is about time to take seriously the secondary use of energy in the form of waste heat. The use of heat capture like this should be part of greening the economy, by reducing the demand for energy. Waste heat as a resource, and not just from data centres. I suspect that we are technically able to do more than we are doing.
This issue came to my attention 10+ years ago when I worked a little with an academic at the University of Sheffield, who was a technical adviser to a business I was working in. He had been involved in the development of a municipal waste incinerator system that generated heat and electricity, including hot water that was piped around Sheffield city centre to provide businesses with hot water year-round, with heating during the winter, and – IIRC – with energy to run aircon during summer. We need more of that kind of thinking, followed by action. I’m unsure where the impetus and political will might come from.
Combined heat and power has been on agendas for decades and has rarely gone very far.
Mainly it’s because the investment modelling ignores externalities
“Mainly it’s because the investment modelling ignores externalities ”
Business partner is an ex-UK civil servant (1990s) covering CHP. It was treated then as unimportant. The sub-agenda was: threat to nukes& big power.
The re-use of energy in imaginative ways remains unresolved.
Rockwool make insulation. Their factories produce vast quantities of heat – which could be re-used in some interesting ways – but are not (greenhouses?) . & I’m not blaming Rockwool – they ain’t in the greenhouse business – the guilty parties are energy ministries and the people @ the top.
CHP is extensivly used in the Netherlands to heat greenhouses. The elec output of the generators means that when it is cold & windy which means CHPs running @ max – on-shore wind output is constrained/reduced so that the gas-burning CHP systems can keep running. Pathetic? comical? you chose.
The failures above are because we continue to focus on energy vectors (elec, gas, heat) instead of energy systems.
Thanks.
Appreciated.
I believe this is happening in Switzerland.
https://www.euronews.com/green/2025/01/30/genevas-homes-will-be-heated-by-recycled-energy-from-this-revolutionary-swiss-data-centre
Really helpful Richard. As an environmentalist who’s always believed that the elephant in the room, which not even the Green Party will vocalise, is going forward we are all going to have to live less consumptive lives, I have failed to see where AI can sit with this. Of course it can’t. I work in education and we had a training session on AI on Monday, and I was pleased that a question was posed on the environmental issues raised by AI use. She got a non answer.
I note that at the end of the video you suggest using ChatGPT to write a letter to your MP! Is this use of AI necessary ? I have no intention of using AI for such things, but am interested in finding out about it’s use with minute taking.
I am not oppoosed to tech.
But do we need AI memes, songs, video, and more, all of which consume considerable energy. I doubt it very much.
AI & Water & Data Centres. Cooling for data centres in north west europe can be: natural – if it is 5C outside and you need 30C in your data centre (typical) then the diff is enough to cool. Water is then used in a closed circuit (no loss). In the summer, you might need chillers. You will certainly need them in hot places. Chillers work through evaporation & thus consume water. I don’t have data on natural vs chiller cooling – but Data Centres with chillers can easily be IDed. (Google Earth).
Data centres also consume integrated circuits (ICs) which need large amounts of water in the mfu process (a cause celebre in the 1990s and 2000s). Many ICs come from Taiwan – where water recycling is almost an art form.
As the late John Doyle observed: data centres are +/- large radiators of heat. Why are we not using/resuing this useful heat?
In terms of size, there are plenty of 100MW and 200MW data centres. I would not say they are “standard” – there are quite a few around Dublin, Paris & of course London. Speaking of London, Telehouse was +/- the worlds first datacentre. Built by the Japanese in the 1980s for telephone & data traffic in Londons Docklands. It quickly morphed into a data hub. Lots of appartments just next to Telehouse, remind me why the “waste heat” is not used to heat the appts? Markets? Lack of imagination? Lack of political will? … what?
We are heading into Matrix territory (the film). I assume everybody is familiar with the overall premise?
Thanks
So spot on Mike, so spot on. I can visualise a s.106 planning gain condition for data centres that required them add benefit to existing infrastructures as we have to fulfil in new house building.
Oh for some joined up thinking………
Again, truth if it were ever needed that contemporary society is too driven by standardised market monopolist thinking. This is why markets are not even close to being efficient allocators of resources without a lot of help.
The true cost of AI usage is not currently being passed on to consumers, AI firms are gambling on the future. When AI firms start passing on the true cost, we might see a reduction in superfluous usage.
Indeed. Despite huge valuations, companies like OpenAI aren’t making a profit on ChatGPT generally, and are burning vast amounts of investor capital building and updating their models.
Given that when any competitor hits the market with a better model, most consumers can almost instantly substitute the better product, the loss leader approach is a massive gamble. Like the dotcom boom many will fail having burned investor money developing these models in the hope that being a market leader will lead to future profits.
Meanwhile, by keeping the cost so low, this encourages both speculative and frivolous usage. Some speculative usage will become serious, useful application, and will remain as costs are fully passed on, but some of the speculative and much of the frivolous usage should fade as costs are passed on, or shift to cheaper, more efficient alternatives that are ‘good enough’ for such non-critical usage.
Now hang on a minute?
What costs of AI are you talking about that are not being passed onto to users or society?
I am not sure about such a bold statement as that to be honest and it has got me thinking.
Already, orgs’ are investing in AI and making other cuts in budgets such as personnel budgets. And that could mean less jobs or lower wages for mouths to feed and shelter.
But when we look at the total entirety of money invested in the world a hell of a lot of capital deployed has been into AI development. My understanding is that AI development capital has outstripped other areas of development for some time. One opinion I read a while back was that had a cure for cancer or dementia had as much money poured into it, we would be much further along. Which would be more beneficial to society and who? There is a cost impact there on you and me and our kids because of arguably too many resources going into AI.
This investment in AI is driven by the returns it promises and those returns are predicated on what?
Data. Our data used to sell us stuff and get us to use AI platforms. So there are some costs here too – one cost being a loss of privacy.
But also, this data is OUR data, it belongs to us but no one is paying us for it? It is just taken without our permission.
And who is charging the AI platforms for learning from us every time we use it? Sentient beings giving their cognition away for free to robots and algorithms?!
So, users and wider society are not footing the cost of AI? No, sorry, that does not stack up. We are a free resource of AI developers.
We must get used to resource accounting and double entry book keeping. This obsession with outputs and ignorance of real inputs and costs is frankly making us one-eyed, giving away resources for free to capital and is likely to either render us surplus to requirements and wipe us out as well. Cost!? Do we really know what we are talking about, when we talk about ‘cost’? I seriously doubt it.
There are various AI tasks that use significant energy. However, if you look at the operations that they replace, they can represent an overall reduction in energy required. For example, if you compare the energy required to produce an AI video to the total energy required to create and export all the frames manually, then the latter may use significant (perhaps more) energy in total. That comparison should be researched to see where AI is efficient or inefficient.
Additionally, one of the challenges that we have with renewables is where energy can be produced in one location, and that’s not where it’s needed. If data centres follow the cheapest energy and build where renewable energy is produced, then it may reduce scenarios where we are paying wind farms not to operate while simultaneously paying fossil fuel plants to generate the same power elsewhere.
Having said that, there is also a lot of unhelpful AI work being done – systems running in the background generating unwanted interpretations through Copilot and other integrations, ‘AI slop’ image, video and text generation, etc. Some of this will fade away as fads change, some will continue as increased demand. With any significant new technology we have always tended to produce more rather than produce the same amount with less people, so in that sense energy usage is expected to remain higher.
For the time being, I do think energy demand will remain elevated. However, as systems reach sufficient quality to not need further improvement, energy usage will reduce. AI was initially about brute force scaling in early years, but has since added retaining accuracy with lower-precision arithmetic and ‘mixture of experts’ models that reduce energy usage for a given size of model.
Those efficiency improvements will be applied as people demand a reduced cost of processing, leading to energy usage dropping.
I also expect we’ll see more energy optimisation as people explore the DGX Spark and other local-only processing, shifting some of the current data-centre focus back to local. In the short term, high RAM prices will limit this, but data privacy as well as long-term cost drive this shift, further AI reducing energy demands and data centres needing water cooling.
I nhave to admit I think you are wildly optimistic. Sorry.
I don’t think we will get to a point where an AI is as energy efficient as the human brain.
In sensible world the AI focus would be on helping humans, not a race to replace humans.
Agreed
The tech bros are in a race to build a huge number of very large data centres fuelled by the tech share price boom.
The bros have no idea if AI will work and be profitable. But heho let’s gamble.
What are we looking at in the US:
1. Huge sites with the resulting massive carbon use during the construction process.
2. Huge electricity requirements.
3. The US electricity infrastructure is not able to meet the demand.
4. If the electricity cannot come to you. Bring the power production unit to you. The proposed way forward, build your own on site small nuclear reactor.
5. What about disposal of the nuclear waste? Not to worry by the time we need to dispose the problem will have been solved.
6. Huge amounts of water needed. Not to worry we take what we want and to hell with the local population. This is already happening.
7. What about the health impact on the local population of these centres? Who cares about the noise, pollution. Again already happening.
8. What about the massive environmental damage being caused? Who cares, that’s for the rest of you to sort out and pay for.
The tech bros must be brought to heel by regulation and the polluter pays principle rigorously enforced against them.
This article is relevant to Mr Fairhall’s comments:
https://www.utilitydive.com/news/ercots-large-load-queue-jumped-almost-300-last-year-official/808820/?_bhlid=60758c7173b3238ecc55d8e9ca8616fa3aa469d9
160GW of NEW electricity connections demanded for data centres JUST IN TEXAS.!!!
Amusingly Texas is not well connected to what passes for the US national grid. What could possibly go wrong?
Oh and water? well I notice that Texas is very hot in the summer – so chillers definitely – one wonders where the water comes from (Texas is also… quite dry).
I have seen lemmings with more sense.
🙂
Thanks
Decent coverage of recent history and prospects for AI:
https://www.lrb.co.uk/the-paper/v47/n23/john-lanchester/king-of-cannibal-island
Thanks
Quite ironic as you pretty much earn your living by regurgitating AI
That’s kind of amusing. Do you know for how long this blog has been going. Was AI around in 2006? Or are you showing you just write total BS?
Are you saying you operate the same today as you did in 2006? or have your practices changed to reflect the opportunities offered by Ai?
We use spell checkers.
Sometimes I smooth out ideas I have dictated by askinf AI to do a light edit to speed production.
But everything starts in my head as an idea, that I turn into words either typed or dictated, which I then edit, even if AI is used as a tool to speed production along the way. AI does not do much more than remove some of the typos, in effect. The exception is the economic questiins series – which always start with me dicating the notes on what I want to say but where a prompt I, of course, wrote has been used to deliver a consistency of editorial style – which I have always acknoweldged.
Your claim is quite simply wrong.
Great article and many great comments.
Regulation and planning absolutely needed. I watched some months ago, on the ‘More Perfect Union’ channel, a couple of videos about the impact of water and energy use by data centres in US.
More recently I read about Denmark’s local district heating plans, which are local and civic-led. A central part of their government’s energy transition programme, and involving house insulation as a prerequisite.
Ideally, where we want and can sensibly put a data centre (I appreciate they are for more than AI) then local government should be involved, to make it work for local people and environment. Ideally…
The regulation & planning of district heating in Denmark (a bit off-topic but interesting)
https://ens.dk/media/7309/download